Slowing Moore’s Law: How It Could Happen

Weak points in the networks powering technological progress: chip factories
transhumanism, politics, predictions, survey
2012-03-162017-10-09 finished certainty: likely importance: 7

re­quires enor­mous com­put­ing pow­er; enor­mous com­put­ing power re­quires fur­ther pro­gres­sion of ; fur­ther Moore’s law re­lies on large-s­cale pro­duc­tion of cheap proces­sors in ever more-ad­vanced chip fabs; cut­ting-edge chip fabs are both ex­pen­sive and vul­ner­a­ble to state ac­tors (but not non-s­tate ac­tors such as ter­ror­ists). There­fore: the ad­vent of brain em­u­la­tion can be de­layed by global reg­u­la­tion of chip fabs.

Here I will de­fend the fol­low­ing chain of claims:

  1. that changes in Moore’s Law can affect the cre­ation of va­ri­eties of soft­ware, and AI soft­ware in par­tic­u­lar

  2. the semi­con­duc­tor in­dus­try is ex­tremely cen­tral­ized, phys­i­cally

  3. these cen­tral­ized fa­cil­i­ties are highly vul­ner­a­ble

    • But coun­ter­in­tu­itive­ly, not to small groups such as ter­ror­ists.
  4. and thus, Moore’s Law can be steered by affect­ing these fa­cil­i­ties

Why might one be in­ter­ested in this top­ic? Chip fab risks and costs may turn out to be a ma­jor lim­it­ing fac­tor on Moore’s law, and a fac­tor that has gone al­most en­tirely un­rec­og­nized (eg. who has ever heard of “Moore’s Sec­ond Law” gov­ern­ing chip fab­s?). The progress of Moore’s law is of ma­jor im­por­tance to any fu­ture fore­casts of eco­nomic growth, progress of both re­search & ap­pli­ca­tion of & (“big data” vs clever al­go­rithm­s), so­ci­o­log­i­cal de­vel­op­ments, and all fore­casts of a tech­no­log­i­cal Sin­gu­lar­i­ty. Any­one in­ter­ested in a bet­ter un­der­stand­ing of the top­ics, more ac­cu­rate fore­casts, or per­haps even affect­ing events di­rect­ly. The fragility of chip fabs is of in­de­pen­dent in­ter­est as it is not gen­er­ally ap­pre­ci­at­ed, and has po­ten­tially global con­se­quences—­for ex­am­ple, a con­flict be­tween China & Tai­wan (which is home to many cut­ting-edge chip fabs) would, at the least, in­ter­rupt semi­con­duc­tor chip de­liv­er­ies to the rest of the world, a con­tin­gency which may not have been ad­e­quately pre­pared for.


In par­tic­u­lar: here are 2 main pro­posed routes to a :

  1. (WBE)/upload
  2. de novo (AGI)1.

WBE is gen­er­ally be­lieved to be rel­a­tively straight­for­ward en­gi­neer­ing, but is ex­pected to take 1018 FLOPS2 (su­per­com­puter es­ti­mate: by 2019) to 1043 FLOPS3 (su­per­com­put­er: by 2111) to do in com­par­i­son with the lat­ter which ei­ther takes a sim­i­lar amount (if it is as in­effi­cient as em­u­lated brains and winds up effec­tively be­ing a syn­thetic up­load) or po­ten­tially much less if there are effi­cient al­go­rithms and op­ti­miza­tions un­avail­able to evo­lu­tion through brains or un­avail­able pe­riod with brains. Hence, we ex­pect that up­loads will be made more likely by pow­er­ful hard­ware and de novo AGI be made more likely by pow­er­ful soft­ware/al­go­rithm­s/­math­e­mat­i­cal in­sight. This ob­ser­va­tion im­me­di­ately sug­gests that any slow­ing in hard­ware de­vel­oped will re­duce the prob­a­bil­ity of up­loads com­ing be­fore de novo AGI (what­ever that prob­a­bil­ity is), and vice ver­sa, any slow­ing in soft­ware/­math will re­duce the prob­a­bil­ity of de novo AGI com­ing be­fore up­loads (what­ever that prob­a­bil­ity was). A slow­ing in both may not change the prob­a­bil­i­ties or rel­a­tive dates of them, but rather post­pone the ap­pear­ance of ei­ther or both.

Why might we want to affect the nat­ural de­vel­op­ment of ei­ther ap­proach? WBEs are fre­quently re­garded as be­ing po­ten­tially highly un­trust­wor­thy and dan­ger­ous, con­sti­tut­ing an .

Regulating Moore’s law

If an or­ga­ni­za­tion (such as a sin­gle­ton world gov­ern­ment like the UN Se­cu­rity Coun­cil) wanted to to­wards de novo AGI (the de­sir­abil­ity of up­loads is con­tentious, with pro and con), then they might with­hold sub­si­dies from hard­ware re­search & de­vel­op­ment, reg­u­late it, or even ac­tively op­pose it with penal­ties rang­ing from the fi­nan­cial to the civil to the mil­i­tary. (Or per­haps neo-Lud­dites wish to de­lay com­puter progress in the in­ter­ests of eco­nomic jus­tice.) This has been in­for­mally pro­posed be­fore; for ex­am­ple, Ber­glas 2009–20124.


Of course, from the util­i­tar­ian per­spec­tive, such a move would have enor­mous con­se­quences inas­much as it would prob­a­bly con­sti­tute a price floor on pro­cess­ing pow­er, dis­ad­van­tag­ing low-power ap­pli­ca­tions like em­bed­ded proces­sors or sup­plies to the de­vel­op­ing world; and many busi­ness arrange­ments are founded di­rectly or in­di­rectly on Moore’s law hold­ing for an­other decade or three. Poverty kills, and any de­lay in eco­nomic growth also kills. Since we can­not ob­serve the con­se­quences of ei­ther de novo AGI or brain em­u­la­tions be­fore they are cre­ated5, and once one is cre­ated we will cease to care about their tim­ings (from an ex­is­ten­tial risk point of view), we must re­sort to ar­gu­ments about what might affect their rel­a­tive ap­pear­ances; even in math, ar­gu­ments or proofs come with non-triv­ial er­ror rates—so how much more so in a le­gal or eco­nomic ar­gu­ment? To not be ba­nal, we would need to make a very solid case in­deed—a case I can­not make nor will make here. Let’s as­sume that the case has been made and ex­am­ine an eas­ier ques­tion, how fea­si­ble such strate­gies are at ac­com­plish­ing the goal at all.

General Feasibility

How fea­si­ble is this? The rel­e­vant mea­sure of pro­gress, Moore’s law, has op­er­ated re­li­ably since the 1950s, over more than 60 years, and has even sped up at times. This would sug­gest it is diffi­cult to slow down or re­verse this par­tic­u­lar kind of tech­no­log­i­cal progress.

On the other hand, at no point in those 60 years has any­one se­ri­ously at­tempted to slow down Moore’s law, and any at­tempt to do so will have been through or­di­nary com­mer­cial meth­ods, which are highly lim­ited in what co­er­cion can be ap­plied. In par­tic­u­lar, that 60 year pe­riod has been his­tor­i­cally un­usu­ally fa­vor­able for tech­no­log­i­cal de­vel­op­ment, with no ma­jor wars in the rel­e­vant na­tions like Japan, Tai­wan, or Amer­ica (his­tor­i­cally un­usu­al—and a war in­volv­ing China was, and still is, per­fectly plau­si­ble). Had Moore’s law sur­vived a mul­ti­lat­eral ther­monu­clear ex­change, that would be ev­i­dence for ro­bust­ness. Had Moore’s law sur­vived le­gal as­sault like peer-to-peer file­shar­ing has, that would be ev­i­dence for ro­bust­ness. But in fact, Moore’s law has been heav­ily fa­vored by gov­ern­ment sub­si­dies to re­search as Amer­i­can com­puter & semi­con­duc­tor ca­pa­bil­i­ties were seen as a ma­jor ad­van­tage over the So­viet Union dur­ing the Cold War; and they have con­tin­ued, as with the end of the Cold War, com­put­ers also be­came a key part of the Amer­i­can econ­omy and self­-im­age. (‘We may not have man­u­fac­tur­ing’, the elites say, ‘but we’re still the best In­ter­net and com­puter econ­omy in the world! Just learn you some com­puter stuff and there is a good job wait­ing in Econ­omy 2.0 for you!’ There is enough truth to this.) Where there is suffi­cient po­lit­i­cal will, even the most dra­matic and ac­ces­si­ble tech­nol­ogy can de­cline—wit­ness the end of Ming Chi­na’s ex­tra­or­di­nary sea ex­pe­di­tions led by or Toku­gawa-era Japan’s . Through­out his­to­ry, there have been long pe­ri­ods of tech­no­log­i­cal stag­na­tion or even re­gres­sion: to draw a straight or ex­po­nen­tial line from a few points us­ing over­lap­ping chronolo­gies is just cher­ry-pick­ing.

To shut down Moore’s law, one can at­tack ei­ther the po­ten­tial for fu­ture im­prove­ments or the ex­is­tence of cur­rent ca­pa­bil­i­ty:

  1. If all re­search ended to­day, then Moore’s law would quickly die: some effi­ciency can be ironed out of ex­ist­ing pro­duc­tion ca­pa­bil­i­ty, but these price drops would quickly hit an as­ymp­tote where new dis­cov­er­ies would be re­quired to make any no­tice­able im­prove­ments.
  2. If re­search con­tin­ued, but all chip fabs were de­stroyed, the knowl­edge would be ster­ile as every chip would be pro­duced in small hand-made quan­ti­ties; and the com­plex­ity of every ad­di­tional ad­vance would make each batch more ex­pen­sive. (Price cuts must be re­al­ized at ; see lat­er.)

Targets for Regulation

When ex­am­ing a sys­tem to speed it up or slow it down, one wants as much lever­age as pos­si­ble, to ma­nip­u­late cen­tral nodes. In a highly dis­trib­uted net­work, ma­nip­u­la­tion may be diffi­cult to im­pos­si­ble as no node is es­pe­cially im­por­tant: the sys­tem can only be ma­nip­u­lated by ma­nip­u­lat­ing a large frac­tion of the nodes. But many net­works are not very dis­trib­ut­ed, and may looks more like a but­ter­fly or fol­low the bow tie or­ga­ni­za­tional ar­chi­tec­ture seen in cel­lu­lar me­tab­o­lism:

…a myr­iad of nu­tri­ent sources are ca­tab­o­lized, or ‘fan in’, to pro­duce a hand­ful of ac­ti­vated car­ri­ers (e.g. ATP, NADH and NADPH) and 12 pre­cur­sor metabo­lites (e.g. glu­cose 6-phos­phate, fruc­tose 6-phos­phate, phos­pho­enolpyru­vate and pyru­vate), which are then syn­the­sized into roughly 70 larger build­ing blocks (e.g. amino acids, nu­cleotides, fatty acids and sug­ars). The pre­cur­sors and car­ri­ers can be thought of as two ‘knots’ of sep­a­rate bow ties that are both fed by ca­tab­o­lism, but whereas the for­mer ‘fan out’ lo­cally to the biosyn­the­sis of uni­ver­sal build­ing blocks, the lat­ter fan out to the whole cell to pro­vide en­er­gy, re­duc­ing power and small moi­eties.

While the In­ter­net is fa­mously dis­trib­uted and does not at first glance fol­low any bow tie net­work, the semi­con­duc­tor in­dus­try is amaz­ingly cen­tral­ized: <14 com­pa­nies make up the ma­jor­ity of global man­u­fac­tur­ing. (like or ) does not ac­tu­ally pos­sess man­u­fac­tur­ing fa­cil­i­ties; they fo­cus on re­search, de­sign, and li­cens­ing. The fa­cil­i­ties re­quire myr­i­ads of skills, re­sources, and tools, and also as in a bow tie, the out­puts of these few fa­cil­i­ties are shipped world-wide to be used in count­less ap­pli­ca­tions in every field of en­deav­our. The chip fabs look very much like the knot of a bow tie and where we might want to start.

Fab costs and requirements

The costs they bear to build each chip fab­ri­ca­tion plant is as­tound­ing, and in­creas­ing even as rev­enue growth slows (squeez­ing out many old com­pa­nies)6; the ba­sic equip­ment alone be­gin in the hun­dreds of thou­sands of dol­lars, lith­o­g­ra­phy ma­chines were $40 mil­lion a piece in 2009, and the most ex­pen­sive sin­gle pieces of equip­ment (like step­pers) can reach prices as high as $50 mil­lion dol­lars. The soft­ware li­cens­ing and en­gi­neer­ing costs that go into a cut­ting-edge proces­sor are equally stag­ger­ing; Brown & Lin­den 2005 (see also An­dreas Olof­s­son):

Cost re­duc­tion via off­shore in­vest­ments in low-wage coun­tries was not a fea­si­ble strat­egy be­cause fab­ri­ca­tion is so cap­i­tal-in­ten­sive that la­bor typ­i­cally ac­counts for 16% of costs (in­clud­ing de­pre­ci­a­tion) in U.S. fabs pro­duc­ing 200mm wafers, and less than 10% in the newer 300mm fabs, which un­der­cuts the ma­jor la­bor cost ad­van­tage of most in­dus­tri­al­iz­ing coun­tries.30…The eco­nomic char­ac­ter­is­tics of each step of the process differ sig­nifi­cant­ly. De­sign is skill in­ten­sive, and re­quires ex­pen­sive EDA () soft­ware, which is typ­i­cally li­censed per de­sign en­gi­neer. Fab­ri­ca­tion re­quires a huge fixed in­vest­ment (cur­rently on the or­der of $2 bil­lion [c. 2004]) to build a plant (called a fab) that holds a wide va­ri­ety of ex­pen­sive equip­ment and that meets ex­treme re­quire­ments of clean­li­ness. As­sem­bly also re­quires ex­pen­sive equip­ment, but the over­all costs of plant and equip­ment are much lower than for the fab, as are the av­er­age skill re­quire­ments. Over­all, worker skill re­quire­ments go down along the value chain (i.e., de­sign is more skil­l-in­ten­sive than man­u­fac­tur­ing, which is more skil­l-in­ten­sive than as­sem­bly)…The avail­abil­ity of out­sourc­ing (for­eign or do­mes­tic) is par­tic­u­larly im­por­tant for small com­pa­nies and start-ups be­cause of the rel­a­tively large fixed cost of EDA tools, which are typ­i­cally li­censed per en­gi­neer. One con­sul­tant es­ti­mated that the min­i­mum an­nual soft­ware ex­pense for a small com­pany is $10 mil­lion.85 For the in­dus­try as a whole, EDA ex­pense runs close to 1% of rev­enue. In that case, a com­pany earn­ing less than $1 bil­lion in rev­enue would be be­low the effi­cient scale for in­-house de­sign. Only the nine largest fa­b­less com­pa­nies met that cri­te­rion in 2004. One con­sul­tant es­ti­mated that out­sourc­ing even within the United States would save a small start-up that does fewer than five de­signs a year up to two-thirds the cost of do­ing the work in­-house.86

…Chip de­sign is highly skil­l-in­ten­sive, since it em­ploys only col­lege-trained en­gi­neers. A cou­ple of medi­um-size chip de­signs will em­ploy as many elec­tri­cal en­gi­neers as a fab for a year or more (although the skills are not di­rectly trans­fer­able). A com­plex chip de­sign like In­tel’s Pen­tium 4, with 42 mil­lion tran­sis­tors on a 180nm linewidth process, en­gaged hun­dreds of en­gi­neers for the full length of the five-year pro­jec­t.[“Comms held Pen­tium 4 team to­gether”, EE Times, No­vem­ber 1, 2000. “Linewidth” refers to the size of the fea­tures etched on a wafer dur­ing the fab­ri­ca­tion process. Each semi­con­duc­tor process gen­er­a­tion is named for the small­est fea­ture that can be pro­duced.]

…The soft­ware effort it­self has in­creased by 375%. Ac­cord­ing to one soft­ware ex­ec­u­tive, a typ­i­cal chip in 1995 went into a stand-alone prod­uct and re­quired 100,000 lines of code. In 2002, a typ­i­cal chip for a net­worked pro­gram­ma­ble prod­uct re­quires a mil­lion lines of code.60 [Jerry Fid­dler, chair­man of Wind River Sys­tems, cited in “Keynoter says chip value is in its in­tel­lec­tual prop­er­ty,” EE Times, June 14, 2002.] The soft­ware, plus the greater com­plex­ity of chips them­selves, has caused de­sign val­i­da­tion hours to grow by 90% for each mil­lion tran­sis­tors. By com­par­ison, the growth lev­els for the ac­tual de­sign en­gi­neer­ing jobs of logic and phys­i­cal de­sign for each mil­lion tran­sis­tors are a rel­a­tively mod­est 17% and 52%, re­spec­tive­ly. This is largely be­cause, as chips have got­ten more com­plex, the process of chip de­sign has be­come more au­to­mat­ed.61

The fa­cil­i­ties are even more im­pres­sive: Fab 11X has 400,000 square feet of a quar­ter-mile on a side. Chip fabs use up­wards of 40 miles of for their (: the pipes ex­pand in di­am­e­ter each gen­er­a­tion) with the in­ter­nal mono­rail trans­porta­tion can be 3 miles long, and the clean rooms must be con­structed with cus­tom pre-cleaned con­struc­tion ma­te­ri­als. (Ce­ment con­sump­tion is so high that In­tel just builds ce­ment plants on their sites.) Chip fab en­ergy con­sump­tion is mea­sured in megawatts, 55–65 megawatts in one case. In­tel cheer­fully notes about its Fab 42 con­struc­tion:

First of all, In­tel is us­ing the largest land-based crane in the world—one that can pick up and place mas­sive roof trusses that weigh ap­prox­i­mately 300 tons each. The crane is so large it had to be de­liv­ered on trucks to the site in pieces—ap­prox­i­mately 250 truck loads in to­tal. Ad­di­tion­al­ly, Fab 42 will re­quire 24,000 tons of steel re­bar and 21,000 tons of struc­tural steel. And to make room for the fab, 875,000 cu­bic yards of dirt had to be ex­ca­vat­ed. When all is said and done, ap­prox­i­mately 10.5 mil­lion man hours will be re­quired to com­plete the pro­ject.

A fab cost ~$1.5b in 1998, $2b in 2004, $3b in 2007, and $5b by 2010. Ju­rvet­son wrote in 2004 that

An­other prob­lem is the es­ca­lat­ing cost of a semi­con­duc­tor fab plant, which is dou­bling every three years, a phe­nom­e­non dubbed Moore’s Sec­ond Law. Hu­man in­ge­nu­ity keeps shrink­ing the CMOS tran­sis­tor, but with in­creas­ingly ex­pen­sive man­u­fac­tur­ing fa­cil­i­ties—cur­rently $3 bil­lion per fab.

Ross 2003 did­n’t fore­see the com­ing ex­plo­sion in prices in de­scrib­ing Moore’s Sec­ond Law, or Rock’s law as he prefers:

Some­times called Moore’s Sec­ond Law, be­cause Moore first spoke of it pub­licly in the mid-1990s, we are call­ing it Rock’s Law be­cause Moore him­self at­trib­utes it to Arthur Rock, an early in­vestor in In­tel, who noted that the cost of semi­con­duc­tor tools dou­bles every four years. By this log­ic, chip fab­ri­ca­tion plants, or fabs, were sup­posed to cost $5 bil­lion each by the late 1990s and $10 bil­lion by now.

Not so. VLSI Re­search es­ti­mates that fabs cost $2 bil­lion apiece, the same as in the late 1990s, even as their pro­duc­tiv­ity has gone up. “In the 1980s, the fabs in­creased their yield; in the 1990s, they started [in­creas­ing] their through­put,” Hutch­e­son says. (Through­put refers to the num­ber of wafers a fab pro­duces in a given time.) Wafer through­put rose from 20 per hour in the early 1990s to about 40 to 50 an hour to­day.

Any­how, the fo­cus was wrong­head­ed; what mat­ters is not the cost of the fab but the value of its prod­uct. If a $100 bil­lion fab made so many tran­sis­tors per penny that it could un­der­cut the prices of a $10 bil­lion com­peti­tor, it would be eco­nom­i­cal (if, of course, you could get the seed cap­i­tal to­gether from a coali­tion of com­pa­nies-or con­ti­nents).7

In­tel’s Fab 32 cost an es­ti­mated $3b in 2007 (clean rooms: 184,000 square feet; to­tal: 1 mil­lion square feet), re­vised to $3.5b by 2011. A 2009–2010 up­grade to an In­tel fab, Fab 11X, cost $2.5b (on top of the $2b up­grade in 2007). The ‘first stage’ of New York 1.3 mil­lion square foot fab will cost >$4.6 bil­lion dol­lars ($1b re­port­edly sup­plied by New York State); Glob­al­Foundries CEO San­jay Jha es­ti­mated in 2017 that a 7n­m-ca­pable chip fab would cost $10-$12b and the 5nm $14–18b. In­tel’s Fab 42 (be­gun 2011–2012) is pro­jected at >$10b. Fab 15 in Tai­wan is es­ti­mated at >$9.3 bil­lion, and they are prepar­ing to start a fab in 2015 pro­jected at >$26 bil­lion; a $20b es­ti­mate for their next fab was re­peated in 2017. Con­struc­tion of a Ger­man chip fab has been blamed for con­tribut­ing to the fi­nan­cial hob­bling of for­merly com­pet­i­tive AMD8, and in­volved com­pa­nies are re­sort­ing to col­lab­o­ra­tions to cover the cap­i­tal costs, even for the largest play­ers (eg. In­tel & build­ing a $3b+ Flash fab to­gether, or still able to build mem­ory chip fabs with in­ter­nal fi­nanc­ing its mem­ory fac­to­ries—as a gov­ern­men­t-backed rep­re­sent­ing 1⁄5th of the world’s 15th largest econ­o­my). The trend shows lit­tle sign of abat­ing for a va­ri­ety of rea­sons91011, and vastly out­paces in­fla­tion. At cur­rent rates, it is not im­pos­si­ble that the to­tal cost of a bleed­ing-edge CPU/GPU chip fab may pass the $100b (in­fla­tion-ad­just­ed) mark some­where in the 2020s or 2030s—well be­fore a num­ber of the Bostrom-Sand­berg es­ti­mates for hard­ware power reach­ing brain em­u­la­tion lev­els. These bil­lions of dol­lars of ex­pen­di­tures are de­vel­oped & man­aged by hun­dreds of thou­sands of em­ploy­ees: TSMC has >38k and In­tel >104k.

To put these fi­nan­cial & in­fra­struc­ture in­vest­ments in per­spec­tive (par­tic­u­larly the pro­jected TSMC in­vest­ment of >$26b for one new fab), the —one of the largest, most ex­pen­sive (after ), and in­tri­cate sci­en­tific pro­grams ever un­der­tak­en, try­ing mul­ti­ple path­ways to the atomic bomb in par­al­lel—­cost $24$21945 bil­lion in 1945 or $25–30b in 2012 dol­lars, with 130k em­ploy­ees.

One can’t help but think that even if pos­si­ble, no one will en­gage in such cap­i­tal ex­pen­di­tures be­cause it will be bad busi­ness. (In Jan­u­ary 2014, In­tel halted de­vel­op­ment of its Fab 42—“touted as the most ad­vanced high­-vol­ume semi­con­duc­tor-man­u­fac­tur­ing fa­cil­ity in the world” and “among the world’s largest con­struc­tion projects in re­cent years”—after ~$1b of con­struc­tion was com­plet­ed.) A semi­con­duc­tor con­sul­tant shows a 2012 es­ti­mate about the cost per gate of the smaller processes (which may re­quire new chip fab­s):

Fig­ure 1: Cost per gate

3. Nex­t-gen­er­a­tion 20-nm pla­nar CMOS will have a range of ad­di­tional tol­er­ance con­trol chal­lenges com­pared to 28-nm. One likely im­pact is that cost per gate at 20-nm will be higher than at 28-nm. With the po­ten­tial for in­creased cost per gate, ad­di­tional com­paction will need to be done, which will lengthen de­sign com­ple­tion times. Cost per gate at 14-nm can also be higher than that at 28-nm.

…New li­braries will need to be de­vel­oped, IP tran­si­tioned to the FinFET struc­tures, test chips run, and pro­duc­tion vol­umes ramped up. At 14-nm, com­plex chips will cost $200 mil­lion to $500 mil­lion to de­sign, and re-spins will cost $20 mil­lion to $50 mil­lion. The cost of fail­ure will in­crease dra­mat­i­cal­ly. What’s more, 14-nm FinFETs are not likely to be in high­-vol­ume pro­duc­tion out­side of In­tel un­til 2016 to 2017. High­-vol­ume pro­duc­tion will re­quire lower power con­sump­tion and lower cost per gate than ear­lier gen­er­a­tions of tech­nolo­gies. After 14-nm, there will be a range of new chal­lenges (EUV, 450-mm, car­bon nan­otubes, etc). The semi­con­duc­tor in­dus­try must be re­al­is­tic that the sup­ply chal­lenges are be­com­ing more diffi­cult, and there will be a length­en­ing of the time to mi­grate to smaller fea­ture di­men­sions.

Con­sis­tent with squeeze on rev­enue and es­ca­lat­ing cap­i­tal costs is the ob­served dis­tri­b­u­tion of man­u­fac­tur­ing. “Re­source Al­lo­ca­tion & Sched­ul­ing in Moore’s Law Twi­light Zone”, Benini July 2012, pg2; we see 20 man­u­fac­tur­ers at the an­cient 130nm, but just 5 man­u­fac­tur­ers at 22/20nm:

“Mar­ket vol­ume wall: only the largest vol­ume prod­ucts will be man­u­fac­tured with the most ad­vanced tech­nol­ogy”

The scale thesis

Given all this, a nat­ural ob­jec­tion is that chip fabs are only cen­tral­ized like this be­cause it’s slightly bet­ter than the de­cen­tral­ized al­ter­na­tives. There’s no point in reg­u­lat­ing chip fabs be­cause any se­ri­ous slow­down will sim­ply en­cour­age de­cen­tral­iza­tion and small­er-s­cale chip fabs. I con­tend that the above fig­ures are so ex­treme that this can­not be the case, and we have ex­cel­lent rea­sons to be­lieve that this cen­tral­iza­tion trend is ro­bust and fun­da­men­tal, and dri­ven by ba­sic eco­nomic facts and par­tic­u­larly bru­tal economies of scale; hence, chip fabs will con­tinue cen­tral­iza­tion as long as pos­si­ble, and any at­tempt to de­cen­tral­ize may well it­self drive prices up and slow down proces­sor de­vel­op­men­t—ex­actly as in­tend­ed.

are one of the more ro­bust ob­ser­va­tions in man­u­fac­tur­ing: the more you make & for longer, the more ex­pe­ri­ence or in­tel­li­gence builds up in your fa­cil­i­ties & hu­mans, and the cheaper or bet­ter they can make them. ob­serves the curve in the 61–62 ar­eas com­piled in the Per­for­mance Curve Data­base. The curve may be re­lated to Moore’s law (eg it is seen in Ko­rean semi­con­duc­tor pro­duc­tion).

One early ex­am­ple of the ex­pe­ri­ence curve effect in com­put­ing is given on pg171 of Tur­ing’s Cathe­dral (2012) by while dis­cussing the choice of com­mod­ity com­po­nents in the con­struc­tion of the (one of the in­flu­en­tial early im­ple­men­ta­tions of a ):

“If the 6J6, which was the twin , had not ex­isted dur­ing the [WWII] war and had not been widely used, I don’t know what we would have used for a tube,” says Willis Ware. Not only did the wide­spread use of the 6J6 mean that it was avail­able in­ex­pen­sive­ly, but it was found to be more re­li­able as well. One of ’s last as­sign­ments at the Sta­tis­ti­cal Re­search Group at Co­lum­bia had in­volved the re­li­a­bil­ity of mu­ni­tions. “There had been a lot of ac­ci­den­tal ex­plo­sions of rocket pro­pel­lant units on air­planes in which the ex­plo­sion would take the wing off a plane,” he ex­plains. “And this would hap­pen in a very rare and er­ratic fash­ion. So we had some ex­cel­lent peo­ple in sta­tis­tics there, in­clud­ing no less than , who founded while work­ing with our group. Sta­tis­ti­cal think­ing had be­come a part of my way of think­ing about life.” It turned out that the most re­li­able tubes were those pro­duced in the largest quan­ti­ties-such as the 6J6. As Bigelow de­scribed it, “We learned that tube types sold at pre­mium prices, and claimed to be es­pe­cially made for long life, were often less re­li­able in re­gard to struc­tural fail­ures than or­di­nary tube types man­u­fac­tured in larger pro­duc­tion lots.”60

That higher qual­ity did not re­quire higher cost was not read­ily ac­cept­ed, es­pe­cially since IBM, who had used the 6J6 as the com­put­ing el­e­ment in its pop­u­lar model 604 elec­tronic cal­cu­la­tor, had re­cently es­tab­lished its own ex­per­i­men­tal tube pro­duc­tion plant in Pough­keep­sie, New York, to de­velop spe­cial com­put­er-qual­ity tubes at a much higher cost. There was in­tense de­bate over whether the choice of the mass-mar­ket 6J6 was a mis­take. Of the fi­nal to­tal of 3,474 tubes in the IAS com­put­er, 1,979 were 6J6s. “The en­tire com­puter can be viewed as a big tube test rack,” Bigelow ob­served.61

“It was con­sid­ered es­sen­tial to know whether such minia­ture tubes as the 6J6 have rad­i­cally in­fe­rior lives com­pared to other types, to an ex­tent ren­der­ing their use in de­sign a ma­jor blun­der; and ac­cord­ingly a crude life-test set up was de­vised and op­er­ated to get some sort of a sta­tis­ti­cal bound on their re­li­a­bil­i­ty,” Bigelow re­ported at the end of 1946. Four banks of 6J6 tubes, twenty in each bank, for a to­tal of eighty tubes, were in­stalled in a test rack so they were ori­ented up, down, and in the two hor­i­zon­tal po­si­tions (cath­ode edge-wise and cath­ode flat). The en­tire rack was mounted on a vi­brat­ing alu­minum plate, and the tubes left to run for three thou­sand hours. “A to­tal of six failed, four within the first few hours, one about 3 days and one after 10 days,” was the fi­nal re­port. “There were four heater fail­ures, one grid short and one seal fail­ure.”62

Financial fragility

This leads to an in­ter­est­ing ques­tion: if a chip fab were de­stroyed, how well would the com­pany weather it? It is diffi­cult to an­swer this, but I will note that In­tel’s 2010 rev­enue was $54b, TSMC’s was $14b and Glob­al­Foundries’s was $3.5b. It is not clear that chip foundry com­pa­nies could sur­vive the de­struc­tion of one or two of their fabs now, much less how fi­nan­cially ro­bust they will be after an­other cost dou­bling or two.

Or will the dou­blings con­tin­ue? If it ceases to be­come profitable to build chip fabs ca­pa­ble of build­ing faster chips, or to up­grade the fabs, this sug­gests that Moore’s law may come to an end on its own with­out any kind of in­ter­ven­tion. One an­a­lyst is al­ready fore­cast­ing the death of new fabs and hence, a slow­ing or end to Moore’s law12. Some sim­ple eco­nomic mod­els put the shut­down be­tween 2015 and 202513.

(The R&D efforts needed to sup­port fabs may be sim­i­larly in­creas­ing: “…R&D costs are rapidly in­creas­ing. In 2009, around $30 bil­lion, or 17% of rev­enue, went to R&D across the in­dus­try–a 40% in­crease over 1999.”.)

Effects of fab disruptions

Can we make any es­ti­mates about the fragility of the net­works sur­round­ing the chip fabs as well as the fabs them­selves?

Semi­con­duc­tor chips are per­haps the epit­ome of an : so­phis­ti­cated prod­ucts where a sin­gle mis­take any­where in a process with dozens or hun­dreds of steps ren­ders the prod­uct val­ue­less. A sin­gle de­fect in a 4-core proces­sor may—at best—dis­able one of its cores, forc­ing the man­u­fac­turer to sell it for half or less, per­haps di­vert­ing it from high­-end con­sumers like gamers to mass-mar­ket game con­soles or other cus­tomers. (For ex­am­ple, the has 8 cores but ships with just 7 func­tion­ing be­cause the man­u­fac­tur­ing yield of func­tion­ing Cells is so low.)

Case studies

Sumitomo Chemical fire

An ob­scure in­ci­dent sug­gests what we might see in a dis­rup­tion; from “Real Chip Short­age Or Just A Pan­ic, Crunch Is Likely To Boost PC Prices”:

The dis­rup­tions now rag­ing in world com­puter chip mar­kets started when an ex­plo­sion at a Sum­it­omo Chem­i­cal Co. fac­tory in the town of Ni­ihama, Japan, on July 4 wiped out the source of 60% of the world sup­ply of an epoxy resin called cresol…When news of the Japan blast sur­faced in such in­dus­try pub­li­ca­tions as Elec­tronic Buy­ers’ News and In­foWorld, spec­u­la­tors started rush­ing to buy up the world’s sup­ply of DRAM chip­s…­Dataque­st’s Giu­dici said big com­pa­nies such as In­ter­na­tional Busi­ness Ma­chines Corp. and Ap­ple Com­puter In­c., which typ­i­cally buy the chips un­der long-term con­tracts, are likely to get through the cur­rent price swings with only small price in­creases of around $30 per ma­chine…Even these big com­pa­nies are pay­ing sharply higher prices when they buy on the spot-mar­ket, though, he ex­plained. DRAM prices for these large com­pa­nies have jumped from $37 on the av­er­age in May to an av­er­age of $55 to­day. “Some of these com­pa­nies have paid as high as $70 in the past week,” Giu­dici said. The hun­dreds of smaller com­pa­nies that have flour­ished in the cur­rent com­puter boom are likely to be forced to add $100 to $150 to the cost of each ma­chine be­cause they must buy on the spot-mar­kets, Giu­dici said. Those hit hard­est, how­ev­er, are peo­ple like Ah­mad who want to up­grade. Barry Lebu, pres­i­dent of 50/50 Mi­cro­elec­tron­ics Inc. in Sun­ny­vale, Calif., said DRAM spot mar­ket prices are av­er­ag­ing $89 per megabyte, up from $39 just four weeks ago: “Some are as low as $69 and some are hit­ting $119, but the av­er­age price is $89.”

and “In­dus­try an­a­lyst still un­sure of the sig­nifi­cance of Hynix fire: Ex­tent of fire dam­age still un­known at Hynix fa­cil­ity in Chi­na, Jim Handy of Ob­jec­tive Analy­sis fills us in on his take of the events”:

There are strong sim­i­lar­i­ties be­tween this in­ci­dent an an­other fire in 1993. In July of that year a fire and ex­plo­sion in a Sum­it­omo Chem­i­cal plant re­moved over 90% of the world’s sup­ply of a cer­tain epoxy that was al­most uni­ver­sally used to at­tach DRAM dice to their pack­ages. The im­pact of this event was to gal­va­nize a DRAM short­age that was al­ready de­vel­op­ing at that time. The short­age lasted un­til the end of 1995, longer than any short­age in the his­tory of the DRAM mar­ket. The du­ra­tion of that short­age was not the re­sult of the fire—other fac­tors were at play. Still, the in­dus­try very quickly tran­si­tioned from the mild on­set of a short­age to a very solid short­age as a a re­sult of the in­ci­dent, even though abun­dant al­ter­na­tives to Sum­it­o­mo’s epoxy were iden­ti­fied within a week.

John C. Mc­Cal­lum’s “Graph of Mem­ory Prices De­creas­ing with Time (1957–2016)” shows an in­ter­est­ing abrupt plateau ~1993 for RAM, last­ing to ~1997, sug­gest­ing that the RAM short­age sparked by the Sum­it­omo Chem­i­cal fire had long-last­ing effects. The Sum­it­omo Chem­i­cal fire has passed into ob­scu­rity with­out much re­cent no­tice, so it’s hard to see how much im­pact it re­ally had.

Toshiba NAND Flash memory

As it hap­pens, his­tory re­cently gifted us with a beau­ti­ful pair of more re­cent, bet­ter doc­u­mented ex­am­ples which en­ables us to an­swer: yes, progress is frag­ile.

In one ex­am­ple, a 13-minute power out­age in a sin­gle Toshiba/WD fab in June 2019 cost an es­ti­mated ~16% of global NAND flash mem­ory out­put that quar­ter. On 2010-12-08, the power grid serv­ing a Toshiba-San­disk chip fab (spe­cial­iz­ing in Flash mem­o­ry) suffered a brief fall in volt­age be­yond what the fab’s lo­cal were de­signed for. This re­sulted in a fab-wide power in­ter­rup­tion of less than a tenth of a sec­ond (~0.07). Toshiba re­ported it took the plant 2 days to re­turn to 100% op­er­a­tion. This cut pro­duc­tion for the quar­ter by 20%, and world­wide pro­duc­tion by 7.5% (ac­cord­ing to iSup­pli’s Yang, quoted by the WSJ1415). Cov­er­age of this in­ci­dent men­tioned pre­vi­ous in­ci­dents in 2007 and 200016, and was rem­i­nis­cent of an­other in­ci­dent on 1993-07-04 where a fire at the only plant man­u­fac­tur­ing a cresol epoxy resin used in man­u­fac­tur­ing chip cases pro­voked an im­me­di­ate dou­bling of DRAM prices un­til the plant was re­paired over 4 months later, and a TSMC mal­ware in­ci­dent in 2018.

(Given the costs in­volved, one might ex­pect re­li­able UPSes to be de­fault, but also re­mem­ber that this fabs can be us­ing any­where up to the dozens of megawatts range of elec­tric­i­ty, which might be diffi­cult to com­pletely UPS. With­out a se­ri­ous and plau­si­ble risk, it would be un­re­al­is­tic to ex­pect the in­dus­try to in­vest mil­lions in sub­stan­tial lo­cal elec­tri­cal power ca­pac­i­ty. And backup power sys­tems them­selves have been the source of er­rors—a gen­eral truth in en­gi­neer­ing highly re­li­able com­plex sys­tems is that the com­plex­ity added to pre­vent er­rors is it­self a ma­jor source of er­ror.)

Kryder’s Law

The Oc­to­ber struck at the hub of a quar­ter of global hard drive man­u­fac­tur­ing ca­pa­bil­i­ty. West­ern Dig­i­tal restarted one flooded plant in De­cem­ber but oth­ers would not be on­line un­til March 2012. Shipped quan­ti­ties were not pro­jected to re­cover un­til Q3 2012 and it took un­til Sep­tem­ber 2012 for the vol­ume to ac­tu­ally re­cov­er; the same source pre­dicted that the 2011 prices would only be matched in 201417.

The floods en­able us to give a quan­ti­ta­tive mea­sure of how much progress was set back. Hard dri­ves fol­low : unit prices halve every 12 months. As it hap­pens, on 2011-03-28 I pur­chased one of the cheap­est hard dri­ves at that point, a 1.5tb Sam­sung hard drive for $51. By Kry­der’s law, in March 2012, I would be able to pur­chase 3tb for ~$50 (and ~$25 in March 2013). But as a mat­ter of fact, the cheap­est drive was $120 for 2tb. An equiv­a­lent 1.5tb drive is now not $25 as one would have pre­dicted be­fore the floods, but a cool $100. And the cheap­est 3tb drive is $200. So not only did one crank of Kry­der’s law (a halv­ing) not hap­pen, a pre­vi­ous turn of the law was re­versed, dou­bling prices; so we can es­ti­mate that Kry­der’s law has been set back by 2 years by some “slow-mov­ing”18 flood­ing con­ducted by brain­less nat­ural forces.

What is strik­ing is the long-term effect of the Thai­land floods. If one were a skep­tic of the “ex­pe­ri­ence curve effects” pre­vi­ously men­tioned, one might rea­son that R&D was con­stantly on­go­ing and so one would hope for a “catch-up effect” where the high prices were merely tem­po­rary and a sort of “su­per Kry­der’s law” op­er­ates briefly to re­store the orig­i­nal trend-line as the pen­t-up R&D im­prove­ments are im­ple­ment­ed. Yet, in­dus­try fore­casts did­n’t es­ti­mate a de­cline un­til 2014, and we can watch the de­vi­a­tion from Kry­der’s law in re­al­time: 6 months after March 2012 (Au­gust 2012), close to a year after the floods, a 3tb hard dri­ve’s price had fallen from $200 to… $150. Ex­actly as pre­dicted from half a crank of Kry­der’s law. A year later (Au­gust 2013), the re­sults are even more dis­mal: now 3tb costs… $130. No catch-up effect had yet oc­curred and it seems un­likely that chip fabs are re­mark­ably more ro­bust. Ex­trap­o­lat­ing Kry­der’s law and the ab­sence of any catch-up effect, we can make some pre­dic­tions out to Au­gust 2014:

  1. March 2013, a 3tb drive will cost: ~$113 (25% off $150) or 26.55g­b/$; wrong (ac­tu­al: 22.34g­b/$)
  2. Au­gust 2013: $75 (50% off $150) or 40g­b/$; wrong (ac­tu­al: 23.25g­b/$)
  3. March 2014: $56 (63% off $150) or 53.57g­b/$; wrong (ac­tu­al: 30g­b/$)
  4. Au­gust 2014: $38 (75% off $150) or 78.95g­b/$; wrong (ac­tu­al: 33g­b/$)

The first pre­dic­tion for March 2013 was blown: it re­quired 24–30G­B/$, while the cheap­est hard drive at Newegg was 22.4G­B/$. So the first pre­dic­tion was off sub­stan­tial­ly, and on the low end, sug­gest­ing that if any­thing, I un­der­es­ti­mated the slow­down in hard drive growth. This pat­tern re­peated it­self through the last pre­dic­tion­s—where I had hoped for as much as 84g­b/$, I must set­tle for a measly 33g­b/$. Look­ing for ex­pla­na­tions, I learned that the hard drive in­dus­try has seen (just like the chip fab in­dus­try), go­ing from scores of man­u­fac­tur­ers to . This con­sol­i­da­tion was partly re­spon­si­ble for the flood dis­as­ters by con­cen­trat­ing fa­cil­i­ties, but has other im­pli­ca­tions: fewer com­peti­tors means less com­pe­ti­tion, less pres­sure to un­der­cut the oth­ers, fos­ters cartel-like be­hav­ior, and sug­gests de­clin­ing profitabil­ity or di­min­ish­ing re­turns since the merg­ers may be dri­ven by economies of scale. Re­gard­less, the abrupt halt of Kry­der’s law seems to have caught oth­ers by sur­prise too, such as backup provider which wrote on 2013-11-26:

We are two years re­moved from the hor­rific flood­ing that caused the Thai­land Drive Cri­sis that cre­ated a world­wide short­age of hard disk dri­ves. Prices for hard dri­ves spiked and have re­mained stub­bornly high, only re­turn­ing to pre-cri­sis lev­els in the last cou­ple of months. The mar­ket and com­pet­i­tive forces that over the past 30 years have pre­dictably dri­ven the cost per gi­ga­byte of stor­age down took a va­ca­tion…In ad­di­tion, the cost per gi­ga­byte also de­clined in an amaz­ingly pre­dictable fash­ion over that time. Be­gin­ning in Oc­to­ber 2011 those 30-years of his­tory went out the win­dow…­Cost per GB for Hard Dri­ves: In Sep­tem­ber 2011, our cost per gi­ga­byte was $0.044. That low wa­ter mark would not be achieved again un­til Sep­tem­ber 2013. In that two-year pe­ri­od, our cost ran as high as $0.064 per gi­ga­byte…When the Drive Cri­sis start­ed, in­dus­try pun­dits es­ti­mated that the hard drive mar­ket would take any­where from 3 months to 1 year to re­cov­er. No one guessed two years. Was the de­lay sim­ply an is­sue in re­build­ing and/or re­lo­cat­ing the man­u­fac­tur­ing and as­sem­bly fa­cil­i­ties? Did the fact that the two in­dus­try lead­ers, Sea­gate and West­ern Dig­i­tal, had to in­te­grate large ac­qui­si­tions slow down the re­cov­ery and sub­se­quent in­no­va­tion? What about the dra­matic shift to­wards tablets and away from Desk­tops and Lap­tops, has that changed the hard drive mar­ket and the de­clin­ing cost per gi­ga­byte trend line forever? What­ever lies ahead, we’ll adapt.

Al­most 4 years lat­er, in July 2017, it has be­come clear to Back­blaze that not only has the re­cov­ery never hap­pened, but post-2011, the im­prove­ment curves for hard dri­ves have dras­ti­cally wors­ened:

Up through the 4 TB drive mod­els, the cost per gi­ga­byte of a larger sized drive al­ways be­came less than the smaller sized dri­ves. In other words, the cost per gi­ga­byte of a 2 TB drive was less than that of a 1 TB drive re­sult­ing in higher den­sity at a lower cost per gi­ga­byte. This changed with the in­tro­duc­tion of 6- and 8 TB dri­ves, es­pe­cially as it re­lates to the 4 TB dri­ves. As you can see in the chart above, the cost per gi­ga­byte of the 6 TB dri­ves did not fall be­low that of the 4 TB dri­ves. You can also ob­serve that the 8 TB dri­ves are just ap­proach­ing the cost per gi­ga­byte of the 4 TB dri­ves… the 6 TB dri­ves have been in the mar­ket at least 3 years, but are not even close to the cost per gi­ga­byte of the 4 TB dri­ves. Mean­while, back in 2011, the 3 TB dri­ves mod­els fell be­low the cost per gi­ga­byte of the 2 TB dri­ves they “re­placed” within a few months. Have we as con­sumers de­cided that 4 TB dri­ves are “big enough” for our needs and we are not de­mand­ing (by pur­chas­ing) larger sized dri­ves in the quan­ti­ties needed to push down the unit cost? Ap­proach­ing Ze­ro: There’s a Lim­it: The im­por­tant as­pect is the trend of the cost over time. While it has con­tin­ued to move down­ward, the rate of change has slowed dra­mat­i­cally as ob­served in the chart be­low which rep­re­sents our av­er­age quar­terly cost per gi­ga­byte over time.

“Back­blaze Av­er­age Cost per GB for Hard Dri­ves; By Quar­ter: Q1 2009–Q2 2017”; shows 2011–2012 price spike due to Thai­land floods fol­lowed by slower cost de­clines 2013–2017 than his­tor­i­cal­ly.

The change in the rate of the cost per gi­ga­byte of a hard drive is de­clin­ing. For ex­am­ple, from Jan­u­ary 2009 to Jan­u­ary 2011, our av­er­age cost for a hard drive de­creased 45% from $0.11 to $0.06 – $0.05 per gi­ga­byte. From Jan­u­ary 2015 to Jan­u­ary 2017, the av­er­age cost de­creased 26% from $0.038 to $0.028 – just $0.01 per gi­ga­byte. This means that the de­clin­ing price of stor­age will be­come less rel­e­vant in dri­ving the cost of pro­vid­ing stor­age.


State-actors: Why Not Terrorism

We have seen how diffi­cult fabs are to make, how few they are, how even small dis­rup­tions spi­ral into global shifts. Does this im­ply that reg­u­la­tion could be ac­com­plished by any mod­estly ca­pa­ble group, such as a im­i­ta­tor or a souped-up ITS?


  1. the size, scale, and re­mote­ness from neigh­bor­ing build­ings of chip fabs im­plies both that se­cur­ing them against con­ven­tional para­mil­i­tary as­sault is easy, and as a per­cent­age of con­struc­tion & op­er­at­ing costs, triv­ial. Se­cur­ing them against con­ven­tional mil­i­tary as­sault (cruise mis­siles, ar­tillery strikes, etc.) is highly non­triv­ial but also un­nec­es­sary as no ter­ror­ist groups op­er­at­ing in the rel­e­vant coun­tries has ac­cess to such equip­ment and per­son­nel. Ter­ror­ists are which are easy to ac­cess and im­pos­si­ble to de­fend.

    This may change in the fu­ture due to tech­no­log­i­cal ad­vance­ments like , which have been fore­cast by strate­gists like John Robb to rad­i­cally level the mil­i­tary play­ing field, but I sus­pect they and other changes may be a wash or even tilt the play­ing field against small ac­tors (as they have fa­mously done in Afghanistan & else­where).19

    If ter­ror­ists can­not go after the chip fab, they can go after sup­port­ing in­fra­struc­ture: at­tack­ing in­stead the power plants or power grid. But such an at­tack is still so­phis­ti­cated be­yond al­most all ter­ror­ist at­tacks ever done and is also eas­ily de­fended again­st: plants use up to 60 megawatts of pow­er, and 1 megawatt of pow­er-pro­duc­ing ca­pac­ity can be in­stalled at $1–7m20 de­pend­ing on sources, so in the worst case the power vul­ner­a­bil­ity can be elim­i­nated by chip fabs set­ting up their own on-site power sup­ply with an in­vest­ment of <$420m (). Of course, the in­vest­ment is not waste—one does­n’t need to buy huge quan­ti­ties from the lo­cal grid if one has one’s own plant stand­ing idle. Power plants are not an in­vest­ment the semi­con­duc­tor in­dus­try would make for no rea­son (e­spe­cially if a lo­cal grid is avail­able), but the first suc­cess­ful at­tack on an elec­tric­ity grid and shut­down of a chip fab would im­me­di­ately mo­ti­vate a shift away from the grid by the re­main­ing chip fabs (sim­i­lar to how 9/11 elim­i­nated the abil­ity of ter­ror­ists to hi­jack any Amer­i­can planes, be­cause the pas­sen­gers will at­tack them sui­ci­dally rather than co­op­er­ate). Even bean coun­ters won’t risk ru­in­ing a >$20b in­vest­ment be­cause they re­fused to pay for a <$400m lo­cal power plant!

  2. Ter­ror­ist groups are from the point of view of get­ting things done: they change their ide­o­log­i­cal po­si­tions, they refuse ne­go­ti­a­tions or com­pro­mis­es, they fail to claim re­spon­si­bil­ity for at­tacks, they cause ‘back­fire effects’, and have a sta­tis­ti­cal record of vic­tory in the sin­gle per­cent­age point range. In gen­er­al, their mo­ti­va­tion seems to be not any sort of ra­tio­nal choice of meth­ods (spec­tac­u­lar bomb­ings be­ing in the first place) but rather a mar­ginal & un­pop­u­lar form of so­cial bond­ing. Any per­son se­ri­ously con­cerned about ar­ti­fi­cial in­tel­li­gence would be pretty stu­pid (and heed­less of the many good rea­sons to err on the side of cau­tion in con­se­quen­tial­ist rea­son­ing) to con­clude that ter­ror­ism is the best strat­e­gy. Since vi­o­lence is a last re­sort, it will be re­sorted to only by those who think the threat is most press­ing, which for rea­sons sim­i­lar to the in auc­tions, will gen­er­ally be those who have made the largest er­rors in rea­son­ing. (This has been for­mal­ized as .)

    Ter­ror­ists mo­ti­vated by such ab­struse goals & rea­son­ing will likely be even more dys­func­tion­al; if highly ed­u­cated and in­tel­li­gent West­ern­ers in STEM fields can vo­cif­er­ously dis­agree on whether the cre­ation of AI is even pos­si­ble dur­ing the 21st cen­tury (much less dan­ger­ous), and ma­te­ri­al­ism still be a live is­sue in , it is highly un­likely that any no­tice­able num­bers of peo­ple will change their minds un­til an AI has ac­tu­ally both been cre­ated and also demon­strated its harm­ful­ness. And even then it is un­clear how well peo­ple would be able to co­or­di­nate: the “nu­clear taboo” took decades to de­vel­op. The only peo­ple who un­der­stand or care about these is­sues are geeks, and they are fa­mously bad at co­or­di­na­tion21 or agree­ment22 (“like herd­ing cats”). Con­trast the ut­ter fail­ure of eco-ter­ror­ists—who are mo­ti­vated by griev­ances backed by past cen­turies of ex­tra­or­di­nar­ily well-doc­u­mented en­vi­ron­men­tal dam­age and op­er­at­ing in a dom­i­nant West­ern ide­o­log­i­cal par­a­digm of —to do any­thing more than petty ar­son & van­dal­ism, with what our hy­po­thet­i­cal ter­ror­ist group would have to ac­com­plish.

  3. Tran­s-na­tional ter­ror­ist groups like Al-Qaeda have demon­strated in the 2000s that they are not re­silient to al­lied gov­ern­ment sup­pres­sion; this point is most vividly il­lus­trated by con­sid­er­ing video of Osama bin Laden in his Pak­istani safe house be­fore his as­sas­si­na­tion, watch­ing him­self on TV, and vainly writ­ing end­less emails to his re­main­ing sub­or­di­nates and wish­ful plot­ting.

    Where such groups cur­rently suc­ceed, it is by a kind of ‘fran­chise’ strat­egy where the lo­cal mem­bers and na­tion­al­ist griev­ances take dom­i­nance. At best, fa­cil­i­ties could be at­tacked in one coun­try and its neigh­bors, which would cer­tainly de­lay progress but on a time-s­cale mea­sured in years rather than decades, as the in­dus­try routes around the dam­age by re­build­ing else­where and in­creas­ing ca­pac­ity in ex­ist­ing chip fabs.

  4. There is no pos­si­ble ben­e­fit to the ter­ror­ist group suc­ceed­ing

    This is a highly spec­u­la­tive util­i­tar­ian strat­e­gy, which could win no fer­vent ad­her­ents (no sol­dier jumps on a grenade for the sake of a cost-ben­e­fit equa­tion), as suc­cess is mea­sured in just the sta­tus quo—no faster chip­s—and one of the few sce­nar­ios in which one could have ev­i­dence of suc­cess would be the cre­ation of de novo AGI, at which point any po­ten­tial chip fab ter­ror­ist has more press­ing con­cerns. Sim­i­lar­ly, ab­stract ideals like Com­mu­nism have al­ways been able to promise their ad­her­ents more con­crete ben­e­fits like groupies or high po­si­tions after the war, but what can be promised in this case?

All of the­se, how­ev­er, im­ply that a na­tion or coali­tion of na­tions could: na­tions have large con­ven­tional mil­i­tary forces, (some­what) func­tion­ing de­ci­sion process­es, in­ter­na­tional pull and com­bat ca­pa­bil­ity (e­spe­cially in the case of the USA’s Air Force or any na­tion pos­sess­ing cruise or bal­lis­tic mis­siles), and value the sta­tus quo in which they are top of the food chain. Some gov­ern­ments are even so far­sighted as to work on .

Covert fabs?

The spe­cial con­struc­tion, power & wa­ter con­sump­tion, and sheer scale of chip fabs sug­gest that it would be diffi­cult to covertly build a com­pet­i­tive un­der­ground chip fab; the Toshiba in­ci­dent sug­gests that weak non-nu­clear may be vi­able; the fragility of clean rooms sug­gest chip fabs could with­stand lit­tle over­pres­sure dam­age; the sen­si­tiv­ity of the chips & equip­ment to me­chan­i­cal dam­age (which will likely in­crease as smaller nanome­ter processes are used) sug­gest that at­ten­u­ated shock­-waves23 may cause out­sized con­se­quences. Worse, for com­mer­cial rea­sons, all ex­ist­ing fa­cil­i­ties ap­pear to be lo­cated within US al­lies and within short fighter range of Amer­i­can mil­i­tary bases (for ex­am­ple, all of In­tel’s fa­cil­i­ties).

We can con­trast chip fabs with par­al­lel ex­am­ples of suc­cess­ful black­-mar­ket pro­duc­tion:

  1. coun­ter­feit elec­tron­ics are not pro­duced at covert fac­to­ries, but by le­git­i­mate fac­to­ries which are ei­ther de­voted to the task and con­doned by lo­cal au­thor­i­ties, or by above-board fac­to­ries which are di­vert­ing part of their pro­duc­tion to knock­offs of their real pro­duc­tion dur­ing what is os­ten­si­bly down­time (“shadow shift pro­duc­tion runs”)

  2. il­licit drugs:

    drug lab­o­ra­to­ries are nu­mer­ous and pro­duce many tons of prod­ucts but they are gen­er­ally small and rel­a­tively low-tech: the process of one of the most chal­leng­ing nar­cotics to pro­duce, hero­in, was dis­cov­ered more than a cen­tury ago, and the chal­lenge stems more from the painstak­ing han­dling and ma­nip­u­la­tion by an ex­pe­ri­enced heroin chemist dur­ing the ex­plo­sive24 acid­i­fy­ing step. Metham­phet­a­mine can be cooked up in a bath­room fol­low­ing the “shake and bake” recipe, and is pop­u­lar in part for this rea­son. Mar­i­juana grow­ing is triv­ial (al­beit grow­ing the finest mar­i­juana is very chal­leng­ing). Co­caine pro­cess­ing was sim­i­larly de­vised around a cen­tury ago and has mul­ti­ple processes, nev­er­the­less, the processes can be pretty sim­ple: in one, gaso­line dis­solves the co­caine from the leaves, sul­fu­ric acid & wa­ter are stomped into the gaso­line, and then a base like lye is added to form a co­caine paste. (Sul­fu­ric acid, lye, wa­ter, and oils have been avail­able for cen­turies or mil­len­ni­a.) It’s worth not­ing that for both heroin and co­caine, re­quir­ing the pu­rifi­ca­tion of sub­stan­tial amounts of bulk raw ma­te­ri­als (poppy sap and coca leaves re­spec­tive­ly), even these rel­a­tively prim­i­tive processes take place in labs in law­less cartel-pro­tected ar­eas like the Golden Tri­an­gle.

  3. bioweapons:

    weaponized bac­te­ria pro­duced in a covert lab might be an ex­am­ple, but so far they have not been pro­duced. The best known ex­am­ple of re­cent biowar­fare, the , re­main mys­te­ri­ous but there seems to be gen­eral agree­ment they were pro­duced at or were re­lated to the large well-funded so­phis­ti­cated le­git­i­mate Amer­i­can mil­i­tary biowar­fare lab­o­ra­tory at . The sarin (toxin dis­cov­ered 1938) stand as some­thing of a coun­ter­point; they had fa­cil­i­ties for pro­duc­ing sarin which were de­scribed as ‘so­phis­ti­cated’, but while they were very in­ter­ested in more com­plex weapons like nu­clear bombs and bac­te­ria, they only made sarin and VX gas, a pack­age-bomb, and an at­tempt to man­u­fac­ture 1000 -style as­sault ri­fles ended in fail­ure with one work­ing pro­to­type (although the AK-47 de­sign is fa­mous for be­ing sim­ple and pro­ducible in even prim­i­tive coun­tries). It may be that bioweapons are much more de­mand­ing of “tacit knowl­edge” than gen­er­ally as­sumed.

  4. Nu­clear weapons are an in­ter­est­ing enough ex­am­ple they de­serve their own sec­tion.

Case-study: suppressing nuclear weapons

The task of slow­ing Moore’s law can be analo­gized to the task of slow­ing or stop­ping :

  • ob­jects: nukes :: proces­sors

  • eco­nomic val­ue: , , :: any in­for­ma­tion-pro­cess­ing task

  • trend: we can analo­gize in a few ways:

    1. num­ber of pos­ses­sors :: coun­tries with cut­ting-edge chip fabs
    2. mega­ton­nage of weapons :: speed of proces­sors
    3. num­ber of weapons :: num­ber of proces­sors

The trend brings out how mod­est our aims are with Moore’s law. What would a rogue state have to do to start or con­tinue a Moore’s law of nu­clear weapons? Let’s take North Ko­rea as an ex­am­ple. North Ko­rea has a few nu­clear bombs of a low kilo­ton­nage. To con­tinue any of the 3 anal­o­gous trends it would have to:

  1. spread nu­clear weapons to one coun­try in the next 18 months (South Ko­re­a?), 2 coun­tries the next 18 months (Viet­nam? Cam­bo­di­a?), 4 coun­tries the next in­ter­val (Thai­land? Laos? Japan? Mon­go­li­a?), …
  2. dou­ble the kilo­ton­nage to say 50kt in the first pe­ri­od, 0.1mt the next, 0.2 the next, 0.4, 0.8, 1.6, 3.2, and within 4 dou­blings it would be test­ing a and then have to be­gin de­sign­ing gi­ga­ton bombs of the sort that nei­ther the United States nor Rus­si­a—the pre-em­i­nent nu­clear weapons de­sign­er­s—ever dreamed of.
  3. be­gin dou­bling its weapon count, which would in­volve vast scal­ing up of its lo­cal ura­nium min­ing ac­tiv­i­ties and of course many new nu­clear re­ac­tors for pro­cess­ing said ura­nium ore. The orig­i­nal nu­clear weapons pro­gram was es­ti­mated to cost sev­eral per­cent­age points of North Ko­re­a’s an­nual GDP for decades, and while it would no doubt be cheaper to man­u­fac­ture bombs once the re­search is done, such dou­blings would quickly drive North Ko­rea (more) bank­rupt.

All of this is re­mark­ably de­mand­ing; to give an idea, Liber­man 2003 briefly dis­cusses and dis­misses as ab­surd the idea that the Amer­i­can gov­ern­ment (!) could build a cut­ting-edge chip fab for mil­i­tary use (em­pha­sis added):

…the De­part­ment of De­fense can at­tempt to achieve tem­po­rary so­lu­tions, such as build­ing its own next gen­er­a­tion gov­ern­men­t-owned chip fab­ri­ca­tion fa­cil­i­ty, but this is likely to be both ex­pen­sive and in­effec­tive. If the best re­search and de­sign ca­pa­bil­ity shifts to China along with man­u­fac­tur­ing, this ap­proach will not work past the next gen­er­a­tion or two of semi­con­duc­tor chip pro­duc­tion25. In ad­di­tion, such tem­po­rary so­lu­tions are not only un­work­able over time if the U.S. wishes to re­tain the best ca­pa­bil­ity that is re­quired for de­fense and in­tel­li­gence needs, but will be far more ex­pen­sive than the so­lu­tions pro­posed above. This is be­cause the op­por­tu­nity to lever­age off the com­mer­cial sec­tor (an ap­proach which the DoD and in­tel­li­gence com­mu­nity rely upon at pre­sent) for new ad­vances and cost sav­ings will be lost.

This anal­ogy is amus­ing, but more in­ter­est­ing is how well nu­clear pro­lif­er­a­tion has suc­ceed­ed. At the dawn of the Nu­clear Age, it was gen­er­ally be­lieved that pro­lif­er­a­tion was in­evitable: as the tech­nol­ogy diffused (eg. diffu­sion pu­rifi­ca­tion tech­niques) and ever more trained en­gi­neers came into be­ing and more na­tions ac­quired nu­clear weapons, still more na­tions would ac­quire nu­clear weapons, and so on and so forth. In par­tic­u­lar, it was “ob­vi­ous” that nu­clear strikes would be­come stan­dard in war­fare: the US would use a nuke on Rus­sia to fore­stall Rus­si­a’s ac­qui­si­tion, or they would be used in Ko­rea26, or in Viet­nam, or in the Mid­dle East, or ter­ror­ists would steal one from Rus­sia (or Pak­istan or…) or… But cu­ri­ously enough, 67 years after Hi­roshi­ma, not only have there never been any nu­clear at­tacks post-WWII, pro­lif­er­a­tion has ac­tu­ally been an in­cred­i­ble suc­cess, with the nu­clear club num­ber­ing a mere 8 mem­ber­s—with mul­ti­ple coun­tries aban­don­ing their nu­clear weapons or nu­clear pro­grams27.

Lit­er­a­ture on the de­vel­op­ment of nu­clear bombs afford an­other use­ful per­spec­tive: the in the de­vel­op­ment of ad­vanced nu­clear weapons. re­marked that just know­ing that it was pos­si­ble for the USA to build & det­o­nate a nu­clear bomb by 1945 made it easy for any pow­er­ful coun­try to de­velop one28, and MacKen­zie & Spinardi 1995 cite (pg10–11) sev­eral civil­ian ex­am­ples of what might be called rein­ven­tions of the atomic bomb (to which we could add the Ger­man team’s be­lated re­al­iza­tion of their crit­i­cal mass mis­take once they learned of Hi­roshi­ma), ob­serv­ing that the mere fact of suc­cess of var­i­ous pro­grams en­cour­ages ad­di­tional pro­grams29. But in no case do they come up with so­phis­ti­cated de­signs along the lines of suit­case nukes, effec­tive tri­tium-boosted mul­ti­-stage hy­dro­gen bombs, or com­pact ICBM-capable de­signs, and this holds true of na­tions at­tempt­ing to join the nu­clear club: their bombs are all ap­par­ently rel­a­tively crude and low ton­nage, and in­com­pa­ra­ble to the cut­ting-edge of Amer­i­can or Russ­ian de­signs.

So the les­son from nu­clear pro­lif­er­a­tion sup­ports our the­sis: high­ly-ad­vanced tech­nolo­gies on the cut­ting-edge are much eas­ier to sup­press and con­trol than prim­i­tive slow de­signs which rep­re­sent the state of the art from pos­si­bly many decades ago.

How­ev­er, the nu­clear ex­am­ple also em­pha­sizes the things coun­tries will not do. In the early Cold War, when the USA was still the sole nu­clear power and the USSR was ex­pected to take an­other 5–10 years for its ac­tive nu­clear weapons pro­gram to suc­ceed, there were a num­ber of pro­pos­als that the USA elim­i­nate the threat: launch a uni­lat­eral strike on the USSR, de­liver an ul­ti­ma­tum to cease nu­clear de­vel­op­ment, or form a world gov­ern­ment with ex­clu­sive con­trol over nu­clear weapons, etc. At this point, there was near-cer­tainty that nu­clear weapons were real & an ex­is­ten­tial threat to coun­tries and hu­man­i­ty, that the USSR un­der Stalin would seek nu­clear weapons, that any nu­clear stand­off would be pre­car­i­ous & dan­ger­ous, and that there would likely be pro­lif­er­a­tion to other coun­tries (not just al­lies of the USSR & USA like China or Eng­land). The USA chose in­ac­tion and an arms race, de­clin­ing any more ex­treme mea­sures (although it came close dur­ing a war for a mi­nor coun­try called Ko­rea, which strate­gi­cally makes no sense). Given that the case for pre­empt­ing the Cold War & MAD was in every way clearer & bet­ter than for chip fabs or AGI in gen­er­al, a for­tiori we can pre­dict with ~100% con­fi­dence that there will never be any con­ven­tional mil­i­tary at­tacks on chip fabs with the pur­pose of de­lay­ing AGI de­vel­op­ment.

Hardened non-covert fabs

Hard­en­ing does not seem to be an an­swer. Leav­ing aside the pre­vi­ous ob­ser­va­tion that it’s un­clear whether com­pa­nies (or even na­tion-s­tates) can sur­vive the con­struc­tion costs of un­hard­ened fabs, even ba­sic hard­en­ing is ex­pen­sive. Linger et al 2002 ar­gues that con­trary to the usual es­ti­mates (that or­di­nary un­der­ground con­struc­tion costs 3–5× nor­mal con­struc­tion cost­s), Nor­we­gian con­struc­tion re­ports (with their long ex­pe­ri­ence in un­der­ground con­struc­tion) for power sta­tions and waste treat­ment plants in­di­cate that un­der­ground con­struc­tion pre­mi­ums are more like 1.25× and may even­tu­ally save mon­ey. An Army overview agrees30. These hard­en­ing cost-es­ti­mates, how­ev­er, are not for struc­tures ro­bust against nu­clear or bunker-bust­ing bombs, which re­quire over­bur­dens up to 2000 feet. The com­plex is smaller than what might be needed to con­tain & de­fend a post-2012 chip fab against US as­sault; a 1997 page men­tions “du­pli­cat­ing this fa­cil­ity to­day with all of its mis­sions would cost about 18 bil­lion US dol­lars” (based on its 1965 con­struc­tion cost of $142m, which in­fla­tion-ad­justed is ~10 times larg­er). So op­ti­misti­cal­ly, hard­en­ing would cost an­other 25% (>$5b based on the TSMC Fab 15 es­ti­mate) or more than 500% (>$125b like­wise).

The China question

Like any good mer­can­tilist de­vel­op­ing East Asian coun­try, China has set its heart on mov­ing up the value chain of chip man­u­fac­tur­ing and heav­ily sub­si­dized lo­cal com­pa­nies31. While it may dom­i­nate prod­uct as­sem­bly and the sim­plest chip fab­ri­ca­tion, its ful­l-strength proces­sors like the is far from state-of-the-art (although this may give it an edge in power con­sump­tion at su­per­com­puter scale). Asia is a ma­jor semi­con­duc­tor con­sumer, and a great deal of gen­eral man­u­fac­tur­ing has al­ready shifted there; both Tai­wan and China are po­ten­tial com­peti­tors32.

If we ac­cept the Loong­son ar­chi­tec­t’s pro­jec­tion of par­ity with In­tel etc in 20 years (2031), this means that the cur­rent sit­u­a­tion of all cut­ting-edge chip fabs in reach of US power may face a se­ri­ous bar­ri­er: chip fabs lo­cated in main­land Chi­na. China is one of the few mil­i­tary pow­ers that can give the USAF a run for its mon­ey, and the other con­tender, Rus­sia, is en­tirely out of the chip fab busi­ness. What hap­pens in, say, 2041 with the Chi­nese chip fabs churn­ing out ma­ture op­ti­mized de­signs? At this point, the WBE Roadmap pro­jec­tions sug­gest that brain em­u­la­tion could be fea­si­ble by the 2040s and so the tar­get chip fabs may well have all been built in Chi­na.

Would China buy into any plan which the West some­how man­aged to agree on?

I con­sider the pos­si­bil­ity re­mote for many rea­sons when I project for­ward from the China of 2012:

  1. Eco­nomic growth is the chief man­date of the Chi­nese gov­ern­ment, and any­thing that might sab­o­tage that is op­posed.

  2. Ren­t-seek­ing and cor­rup­tion are com­mon, par­tic­u­larly with ‘pres­tige’ projects or ex­tremely large busi­ness in­ter­ests; if chip fab costs are brought un­der con­trol, they will still be enor­mous pool of cap­i­tal backed by even more enor­mous cor­po­ra­tions. Such en­ti­ties will be able to de­fend them­selves and de­feat any such agree­ments.

  3. In­ter­na­tional agree­ments are seen as a tool of an in­ter­na­tional sys­tem stacked against Chi­na. Any re­quests or “med­dling” in Chi­nese affairs is greeted with re­flex­ive na­tion­al­ism.

    Con­verse­ly, in­ter­na­tional agree­ments are best seen as a form of le­gal war­fare; an ex­po­nent of this view was the wide­ly-read 199 text .

  4. China feels en­ti­tled to any­thing the West did dur­ing its own de­vel­op­ment; if the West could pol­lute with car­bon emis­sions, China may pol­lute (one of its main ra­tio­nale for ig­nor­ing things like the Ky­oto agree­ment or sab­o­tag­ing cli­mate-change talk­s). A sim­i­lar ar­gu­ment may be used with chip fabs.

  5. Spec­u­la­tive­ly, there may be more cul­tural com­fort in China with the pos­si­bil­ity of ‘ro­bots’; at least, some Japan­ese have as­cribed Japan­ese in­vest­ment into ro­bots as be­ing due to such cul­tural traits (eg. Mu­rakami 2005), and this may carry over to Chi­na. With lit­tle anx­i­ety about the is­sue, why would they bother es­pe­cially when the costs are so con­crete and near?

All of these traits can change, and I fully ex­pect many to change be­fore 2040 (for ex­am­ple, I ex­pect na­tion­al­ism & xeno­pho­bia to de­crease as the pop­u­la­tion ages, and re­spect for in­ter­na­tional law will in­crease as China in­creas­ingly be­comes the coun­try that ben­e­fits from a tidy sta­tus quo), but all of them, on top of the orig­i­nal un­like­li­ness?

  1. Any ac­tual AI is likely to be a blend of ap­proach­es—w­hole brain em­u­la­tion and AGI form a con­tin­uum from a brute-force mol­e­cule-by-mol­e­cule em­u­la­tion of a dis­sected hu­man brain to a first-prin­ci­ples learn­ing al­go­rithm ca­pa­ble of generic cross-do­main learn­ing & ac­tion which bears as much re­la­tion to hu­man think­ing as a he­li­copter flies like an al­ba­tross or a sub­ma­rine (pace Dijk­stra) swims like a shark. A blend might be some­thing like a vast neural net­work based on hu­man imag­ing stud­ies but with en­tire chunks of the brain re­placed with sim­pli­fied pro­grams and hard­wired as­sump­tions im­posed on each neu­ron in the in­ter­est of op­ti­miza­tion & com­pu­ta­tional tractabil­i­ty, where the sim­pli­fied chunks are those parts of brain func­tion­al­ity we have un­der­stood to the point where we can reim­ple­ment them by hand.↩︎

  2. Ana­log net­work pop­u­la­tion mode↩︎

  3. Sto­chas­tic be­hav­ior of sin­gle mol­e­cules↩︎

  4. After writ­ing this es­say, I dis­cov­ered Ber­glas had briefly pointed out that proces­sors are a weak point:

    Try­ing to pre­vent peo­ple from build­ing in­tel­li­gent com­put­ers is like try­ing to stop the spread of knowl­edge. Once Eve picks the ap­ple it is very hard to put it back on the tree. As we get close to ar­ti­fi­cial in­tel­li­gence ca­pa­bil­i­ties, it would only take a small team of clever pro­gram­mers any­where in the world to push it over the line. But it is not so easy to build pow­er­ful new com­puter chips. It takes large in­vest­ments and large teams with many spe­cial­ties from pro­duc­ing ul­tra pure sil­i­con to de­vel­op­ing ex­tremely com­plex log­i­cal de­signs. Ex­tremely com­plex and pre­cise ma­chin­ery is re­quired to build them. Un­like pro­gram­ming, this is cer­tainly not some­thing that can be done in some­one’s garage.

    So this pa­per pro­poses a mora­to­rium on pro­duc­ing faster com­put­ers. Just make it il­le­gal to build the chips, and so starve any Ar­ti­fi­cial In­tel­li­gence of com­put­ing pow­er.

    We have a prece­dent in the con­trol of nu­clear fu­el. While far from per­fect, we do have strong con­trols on the avail­abil­ity of bomb mak­ing ma­te­ri­als, and they could be made stronger if the po­lit­i­cal will ex­ist­ed. It is rel­a­tively easy to make an atomic bomb once one has enough plu­to­nium or highly en­riched ura­ni­um. But mak­ing the fuel is much, much hard­er. That is why we are alive to­day. If some­one pro­duced a safe and afford­able car pow­ered by plu­to­ni­um, would we wel­come that as a so­lu­tion to soar­ing fuel prices? Of course not. We would con­sider it far too dan­ger­ous to have plu­to­nium scat­tered through­out so­ci­ety…(It might turn out that it is ac­tu­ally the patent trolls and at­tor­neys that are our sav­ior. In­tel­li­gence de­vel­op­ment would pro­vide a rich source of triv­ial patents and ag­gres­sive lit­i­ga­tion…)

    So­tala & Yam­pol­skiy men­tions “Ber­glas (per­sonal com­mu­ni­ca­tion) has since changed his mind, and no longer be­lieves that it is pos­si­ble to effec­tively re­strict hard­ware or oth­er­wise pre­vent AGI from be­ing cre­at­ed.”↩︎

  5. In dis­cus­sions of tech­no­log­i­cal fore­cast­ing like AI pro­jec­tions or or , the de­mands of pes­simist­s/skep­tics that the events in ques­tion al­ready have hap­pened be­fore they can be dis­cussed are quite ir­ri­tat­ing. I am re­minded of the sar­cas­tic com­ment by 2 re­searchers on cli­mate change:

    “Should we trust mod­els or ob­ser­va­tions?” In re­ply we note that if we had ob­ser­va­tions of the fu­ture, we ob­vi­ously would trust them more than mod­els, but un­for­tu­nately ob­ser­va­tions of the fu­ture are not avail­able at this time.

  6. “Na­tional Se­cu­rity As­pects of the Global Mi­gra­tion of the U.S. Semi­con­duc­tor In­dus­try”, Lieber­man 2003:

    A fun­da­men­tal change in the semi­con­duc­tor in­dus­try has been, in very sim­pli­fied form, that the price to per­for­mance curve has re­duced rev­enue in the in­dus­try dra­mat­i­cally over the last decade. Dur­ing the early 1960s, and con­tin­u­ing un­til about 1994, the com­pound an­nual growth rate in rev­enue of the in­dus­try was 16%. From 1994 to the pre­sent, the growth rate has been ap­prox­i­mately 8%.13 This sit­u­a­tion is com­bined with the very large costs as­so­ci­ated with the de­vel­op­ment of new 300mm fab­ri­ca­tion fa­cil­i­ties (“fabs”), as well as the in­creas­ing com­plex­ity and cost of re­search and de­sign as the in­dus­try must de­velop meth­ods other than the tra­di­tional scal­ing meth­ods (mak­ing all as­pects of the chips smaller and small­er) in or­der to in­crease per­for­mance. These fac­tors, and the cur­rent re­ces­sion, are dri­ving the in­dus­try to con­sol­i­da­tions.

    …The num­ber of state-of-the-art U.S. chip man­u­fac­tur­ing fa­cil­i­ties is ex­pected to sharply de­crease in the next 3–5 years to as few as 1–2 firms that now have the rev­enue base to own a 300mm wafer pro­duc­tion fab, and likely less than a hand­ful of firms.15 Al­though the U.S. cur­rently leads the world semi­con­duc­tor in­dus­try with a 50% world mar­ket share, the Semi­con­duc­tor In­dus­try As­so­ci­a­tion es­ti­mates that the U.S. share of 300mm wafer pro­duc­tion ca­pac­ity will be only ap­prox­i­mately 20% in 2005, while Asian share will reach 65% (only 10% of this from Japan).16 The re­main­ing state-of-the-art U.S. chip-mak­ing firms face great diffi­culty in at­tain­ing the huge amounts of cap­i­tal re­quired to con­struct nex­t-gen­er­a­tion fabs. This sit­u­a­tion stands in con­trast to that in Chi­na. To en­sure that they de­velop the abil­ity to build the nex­t-gen­er­a­tion fab­ri­ca­tion fa­cil­i­ties, the Chi­nese cen­tral gov­ern­ment, in co­op­er­a­tion with re­gional and lo­cal au­thor­i­ties, has un­der­taken a large ar­ray of di­rect and in­di­rect sub­si­dies to sup­port their do­mes­tic semi­con­duc­tor in­dus­try. They have also de­vel­oped a num­ber of part­ner­ships with U.S. and Eu­ro­pean com­pa­nies that are cost-ad­van­ta­geous to the com­pa­nies in the short­-term.

  7. A caveat that will be­com­ing in­creas­ingly im­por­tant in com­ing decades.↩︎

  8. “The rise and fall of AMD: How an un­der­dog stuck it to In­tel; Re­mem­ber when AMD could com­pete with In­tel in both speed and price?”, ars tech­nica:

    AMD’s suc­cess­ful Athlon chip­s—Ars named the Athlon its “CPU of the Year” in 1999—had fi­nally put the screws to archri­val In­tel, and in 2000 the com­pany earned nearly $1 bil­lion in profits…AMD has been on a no­table drop for nearly a decade now. To put it mild­ly, 2012 was a rough year: AMD lost over $1 bil­lion, effec­tively wip­ing out its $471 mil­lion profit in 2010 and its $491 mil­lion profit in 2011-its two most profitable years in the last decade. Over the last 15 years, AMD has sus­tained a net loss of nearly $7 bil­lion, and the com­pany has been down­graded by credit rat­ing agen­cies, burned by lower de­mand for PCs (and hence, for its prod­uct­s), and even called “un-in­vestable” by one Wall Street an­a­lyst…How­ev­er, just weeks after the K7 de­buted on June 23, 1999, Raza left AMD…As Raza tells the story to­day, his boss in­sisted on build­ing a fab in Dres­den, Ger­many, over Raza’s ob­jec­tions. (That fab, which still op­er­ates to­day as part of AMD spin-off Glob­al­Foundries, was com­pleted in the spring of 2000.) “The trou­ble in the en­tire eco­nomic model was that AMD did not have enough cap­i­tal to be able to fund fabs the way they were fund­ing fabs,” Raza said. “The point at which I had my fi­nal con­flict was that [Sanders] started the process of build­ing a new fab with bor­rowed money pre­ma­ture­ly. We did­n’t need a fab for at least an­other year. If we had done it a year lat­er, we would have ac­cu­mu­lated enough profits to afford the fab in Ger­many. He laid the foun­da­tion for a fun­da­men­tally in­effi­cient cap­i­tal struc­ture that AMD never re­cov­ered from. I told him: don’t do it….” Both Raza and Bar­ton re­called, in­de­pen­dently of one an­oth­er, one of Sanders’ mantras: “Real men have fabs.” Raza called this com­ment “si­mul­ta­ne­ously a sex­ist re­mark and the most stu­pid thing you can say,” and he saw the fab de­ci­sion as one of Sanders’ “sig­nifi­cant acts of ir­re­spon­si­bil­i­ty.”

  9. Kim 2008, “Brick and Mor­tar Chip Fab­ri­ca­tion”

    Tech­nol­ogy scal­ing has pro­duced a wealth of tran­sis­tor re­sources and cor­re­spond­ing im­prove­ments in chip per­for­mance. How­ev­er, these ben­e­fits come with an in­creas­ing price tag, due to ris­ing de­sign, en­gi­neer­ing, and val­i­da­tion costs of mod­ern chips [15]. The re­sult has been a steady de­cline in unique ap­pli­ca­tion-spe­cific in­te­grated cir­cuit (ASIC) de­signs that en­ter pro­duc­tion [21]. This ini­ti­ates a vi­cious cy­cle. Fewer unique chips means that fabs have fewer cus­tomers across which to amor­tize their costs, lead­ing to even higher costs for those who do man­u­fac­ture chips. The cy­cle com­pletes as higher chip man­u­fac­tur­ing costs ex­clude even more po­ten­tial man­u­fac­tur­ing cus­tomers.

  10. Kim 2008, con­tin­ued:

    While Moore’s Law has fu­eled the semi­con­duc­tor in­dus­try, it has also fu­eled this spi­ral of in­creas­ing costs and shrink­ing fab cus­tomer bases. As tran­sis­tors have shrunk, the cost of fab­ri­cat­ing a semi­con­duc­tor de­vice has grown com­men­su­rate­ly. While the fab­ri­ca­tion cost per tran­sis­tor has steadily de­clined [62], mul­ti­ple other ex­penses have bal­looned, con­tribut­ing col­lec­tively to the grow­ing to­tal. For ex­am­ple, small fea­tures are more sus­cep­ti­ble to process vari­a­tion than larger ones, in­creas­ing the range of vari­a­tion and the pro­por­tion of faulty chips. In ad­di­tion, the smaller the tran­sis­tor, the more of them that can fit in a given amount of sil­i­con. The re­sult is that cir­cuit com­plex­ity has been in­creas­ingly out­-strip­ping de­signer pro­duc­tiv­i­ty, in a phe­nom­e­non re­ferred to as Moore’s Law’s corol­lary of “com­pound com­plex­ity” [143].

    The in­dus­try has dealt with these chal­lenges by in­creas­ing the en­gi­neer­ing effort that goes into each chip. This effort man­i­fests it­self as larger de­sign teams, or longer prod­uct cy­cles, and often both at once. The vast ma­jor­ity of this en­gi­neer­ing effort is in­curred once per chip de­sign, and does not vary with the num­ber of chips pro­duced. Ac­cord­ing­ly, this ex­pense is called the non-re­cur­ring en­gi­neer­ing cost (NRE) of a chip. In­dus­try an­a­lysts es­ti­mate that the NREs for a typ­i­cal 90nm stan­dard cell ASIC can range from $5M up to $50M [113].

  11. Kim 2008, con­tin­ued:

    Main­tain­ing a par­tic­u­lar­quires larger and larger batches of chips. This is be­cause the sin­gle NRE is shared evenly across the pop­u­la­tion of chips pro­duced. The larger the pop­u­la­tion, the smaller the im­pact of the NRE on in­di­vid­ual chip cost…The re­sult of this sit­u­a­tion is that only high­-vol­ume chip man­u­fac­tur­ers, or those who can sell smaller batches at high prices, can afford to be in the chip busi­ness. More­over, at the same time that com­plex­ity and en­gi­neer­ing effort have been soar­ing, the com­mer­cial mar­ket has been de­mand­ing and re­ward­ing short chip de­sign cy­cles. This is due to shrink­ing prod­uct life­times and the in­creas­ing com­pet­i­tive im­por­tance of be­ing the first to mar­ket with a new pro­duc­t…One of the in­puts is the as­sumed NRE. The NRE in­cludes all en­gi­neer­ing effort…with an en­gi­neer’s time cost­ing up­wards of $380,000 per year [141], the en­gi­neer­ing cost is nearly al­ways a sev­en-fig­ure number…NREs also en­com­pass the cost of tools, IP li­censes if nec­es­sary, and pho­tolith­o­graphic masks. ASIC de­sign tools typ­i­cally cost more than $300,000 [146]. Mask cost has been roughly dou­bling every tech­nol­ogy node, re­sult­ing in a com­plete set of 90nm masks cost­ing be­tween $1M and $3M [146]…In 2000, the cost to test each tran­sis­tor was 10% of the cost to man­u­fac­ture it. How­ev­er, as tran­sis­tors be­come cheaper and test­ing be­comes more diffi­cult, it is pro­jected that by 2015 it will cost more to test a tran­sis­tor than to make it [73].

  12. “Moore’s Law reaches its eco­nomic lim­its”, Fi­nan­cial Times 2012:

    “The high cost of semi­con­duc­tor man­u­fac­tur­ing equip­ment is mak­ing con­tin­ued chip­mak­ing ad­vance­ments too ex­pen­sive to use for vol­ume pro­duc­tion, rel­e­gat­ing Moore’s Law to the lab­o­ra­tory and al­ter­ing the fun­da­men­tal eco­nom­ics of the in­dus­try,” wrote Len Je­linek, chief an­a­lyst for semi­con­duc­tor man­u­fac­tur­ing at the iSup­pli re­search firm, last month.

    Mr Je­linek pre­dicted that Moore’s Law would no longer drive vol­ume chip pro­duc­tion from 2014, spark­ing in­tense de­bate in Sil­i­con Val­ley. His rea­son­ing is that cir­cuitry widths will dip to 20nm (nanome­tres or bil­lionths of a me­tre) or be­low by that date. But the tools to make them would be too ex­pen­sive for com­pa­nies to re­cover their costs over the life­time of pro­duc­tion.

    The costs and risks in­volved in build­ing new fabs have al­ready dri­ven many mak­ers of logic chips (proces­sor or con­troller chips) to­wards a “fa­b­less” chip mod­el, where they out­source much of their pro­duc­tion to chip “foundries” in the Far East.

    The 14 chip­mak­ers who were in the game at the 90nm level have been re­duced to nine at the cur­rent 45nm lev­el. Only two of them—In­tel and —have firm plans for 22nm fac­to­ries. In­tel ar­gues that only com­pa­nies with about $9bn in an­nual rev­enues can afford to be in the busi­ness of build­ing new fabs, given the costs of build­ing and op­er­at­ing the fac­to­ries and earn­ing a de­cent 50 per cent mar­gin. That leaves just In­tel, Sam­sung, , and .

  13. “The Eco­nomic Limit to Moore’s Law”, Rupp & Sel­ber­herr 2011

    …The re­duced growth model (10) can be in­ter­preted in the fol­low­ing way; as long as fab costs in­crease with the same rate as they did in the past, the num­ber of tran­sis­tors per chip also in­creases at the same rate as in the past. How­ev­er, as soon as fab costs hit an eco­nomic bar­rier given by εg(t), fab costs can only in­crease at the same rate as the GWP [Global World Pro­duct] does. Con­se­quent­ly, tran­sis­tor counts will also grow at a re­duced rate…The pre­dic­tion of the time at which we run into eco­nomic lim­i­ta­tions is very sen­si­tive with re­spect to the fab costs pa­ra­me­ter ε; choos­ing ε = 0.03%, a growth re­duc­tion is pre­dicted around 2025, whereas the choice ε = 0.01% shows first signs of re­duced growth al­ready in 2015. Thus, with joint fund­ing of large fabs, an eco­nomic growth cap­ping can be shifted many years into the fu­ture so that we might face lim­i­ta­tions im­posed by physics first.

  14. “Power Blip Jolts Sup­ply of Gad­get Chips”, De­cem­ber 10, 2010, WSJ:

    Toshiba said the power out­age could cause a 20% drop in its ship­ments over the next two months or so of chips known as NAND flash mem­o­ry, which are used to store mu­sic, pho­tos and data in prod­ucts such as Ap­ple In­c.’s iPhone and iPad. Toshiba, which makes the chips in part­ner­ship with Sil­i­con Val­ley com­pany San­Disk Cor­p., rep­re­sents about a third of the mar­ket as the sec­ond-largest sup­plier of the chips after Sam­sung Elec­tron­ics Co. After the next cou­ple months, the out­age is­n’t ex­pected to have a [sub­stan­tial] im­pact on world-wide ship­ments of flash mem­o­ry. Some big buy­ers of the chips, such as Ap­ple, have long-term sup­ply arrange­ments with mul­ti­ple chip mak­ers. But the tem­po­rary dis­rup­tion comes as de­mand for NAND flash is surg­ing, no­tably from com­pa­nies hop­ing to offer new tablet com­put­ers to chal­lenge the iPad. Mar­ket watch­ers say some com­pa­nies could face tight sup­plies and higher prices just as they are try­ing to ramp up pro­duc­tion. “I don’t think it could come at a worse time,” said Kr­ishna Shankar, an an­a­lyst at ThinkE­quity.

    Toshiba’s trou­bles started early Wednes­day when, ac­cord­ing to power sup­plier Chubu Elec­tric Power Co., there was a sud­den drop in volt­age that caused a 0.07-sec­ond power in­ter­rup­tion at Toshiba’s Yokkaichi mem­o­ry-chip plant in Mie pre­fec­ture. Even the briefest power in­ter­rup­tion to the com­plex ma­chines that make chips can have an effect com­pa­ra­ble to dis­con­nect­ing the power cord on a desk­top com­put­er, since the com­put­er­ized con­trols on the sys­tems must effec­tively be re­boot­ed, said Dan Hutch­eson, a chip-man­u­fac­tur­ing an­a­lyst at VLSI Re­search in San Jose, Calif. For that rea­son, chip com­pa­nies typ­i­cally take pre­cau­tions that in­clude in­stalling what the in­dus­try calls un­in­ter­rupt­ible power sup­plies. Part of Toshiba’s safe­guards did­n’t work this time be­cause the volt­age drop was more se­vere than what the backup sys­tem is de­signed to han­dle, a com­pany spokesman said. Power out­ages fre­quently cause dam­age to chips, which are fab­ri­cated on sil­i­con wafers about the size of din­ner plates that may take eight to 12 weeks to process, Mr. Hutch­e­son said. Wafers that are in­side pro­cess­ing ma­chines at the time of an out­age are often ru­ined, he added, though many that are in stor­age or in tran­sit among those ma­chines are not. In some cas­es, a shut­down of the air-pu­ri­fy­ing and con­di­tion­ing sys­tem that keeps air in a chip fac­tory free of dust also could con­t­a­m­i­nate chips. Mr. Hutch­e­son com­pared the sit­u­a­tion to cut­ting off the power to an ar­ti­fi­cial heart ma­chine in the mid­dle of an op­er­a­tion. “You lose the pa­tient,” he said. On the other hand, he said that Toshiba’s es­ti­mate of the im­pact is a worst-case sce­nario that may wind up to be sub­stan­tially less. Toshiba es­ti­mated that its ship­ments of NAND flash mem­ory could de­cline by as much as 20% through Feb­ru­ary as a re­sult of the out­age. Based on the com­pa­ny’s share of the mar­ket, such a re­duc­tion would trans­late into a 7.5% cut in world-wide ship­ments over that pe­ri­od, but a much smaller per­cent­age for all of 2011, es­ti­mated Michael Yang, an an­a­lyst at the tech­nol­ogy mar­ket re­search firm iSup­pli.

  15. Con­tin­ued:

    “The im­pact for the year is noth­ing,” Mr. Yang said, though he added it could tem­porar­ily raise NAND prices. Such prob­lems are far from un­prece­dent­ed. In Au­gust 2007, a power out­age at Sam­sung’s mem­o­ry-chip plant near Seoul forced the com­pany to tem­porar­ily halt some pro­duc­tion lines. Mr. Yang said that cased prices for chips to rise for a few months. Tim­o­thy Luke, an an­a­lyst at Bar­clays Cap­i­tal, es­ti­mated the out­age would re­duce 2011 NAND flash chip sup­plies by 3% to 5%, pos­si­bly boost­ing prices to the ben­e­fit of ri­vals such as Sam­sung and Mi­cron Tech­nol­ogy Inc. A San­Disk spokesman de­clined to com­ment on the out­age. One mit­i­gat­ing fac­tor is that chip de­mand is typ­i­cally lighter in Jan­u­ary and Feb­ru­ary than other parts of the year, which could re­duce the chances of a short­age. On the other hand, de­mand for NAND chips has been ris­ing at an un­usual rate, dri­ven largely by sales of smart­phones and tablets. ISup­pli in Sep­tem­ber pre­dicted that global ship­ments of NAND flash-mem­ory chips by vol­ume would jump 70% next year.

    An­other in­stance of power out­age dam­ag­ing a frac­tion of global out­put hap­pened again to Sam­sung on 2018-03-09.↩︎

  16. “The Split Sec­ond Dis­rup­tion to the Sup­ply Chain”:

    In 2005, Pro­fes­sor Yossi Sheffi from the Mass­a­chu­setts In­sti­tute of Tech­nol­ogy wrote a sem­i­nal work en­ti­tled The Re­silient En­ter­prise that ex­am­ined dis­rup­tions in cor­po­rate sup­ply chains. The very first chap­ter in that book was en­ti­tled “Big Lessons from Small Dis­rup­tions.” It un­folds a story about how a St. Patrick’s Day (2000-03-17) light­ning strike in Al­bu­querque, New Mex­i­co, started a fire in Fab­ri­ca­tor No. 22 of a Phillips NV chip man­u­fac­tur­ing plant which led to un­fore­seen long-term con­se­quences. Alert plant em­ploy­ees and au­to­matic sprin­klers put the fire out in less than ten min­utes. Sheffi wrote, “A rou­tine in­ves­ti­ga­tion showed that the fire had been mi­nor. No­body was hurt and the dam­age seemed su­per­fi­cial. The blaze did not make head­lines in Eu­rope, did not ap­pear on CNN, and did not even ap­pear in the Al­bu­querque news­pa­pers.”…The fire had di­rectly ru­ined only eight trays of wafers but smoke from the fire spread be­yond the im­me­di­ate area and, along with soot spread by work­ers and fire­fight­ers, con­t­a­m­i­nated much of the rest of plant. A mi­nor fire had turned into a ma­jor dis­as­ter. Sheffi re­ported that Phillips no­ti­fied its 30-plus cus­tomers about po­ten­tial de­lays in chip pro­duc­tion but pre­dicted the de­lay would only be about a week…as soon as it [Nokia] re­al­ized that the de­lay was ac­tu­ally go­ing to be weeks or months, it took ac­tion…Er­ic­s­son was­n’t quite so lucky. Sheffi re­ported that Er­ic­s­son ex­ec­u­tives re­ceived the same tele­phone call from Phillips as Nokia but they re­acted very differ­ent­ly. They be­lieved that the de­lay would be a short one…Ac­cord­ing to Sheffi, Er­ic­sson’s lack of a Plan B cost the com­pany around half a bil­lion dol­lars.

  17. iSup­pli at­trib­utes the very slow re­cov­ery to, in part, long-term con­tracts lock­ing in PC man­u­fac­tur­ers and the hard drive in­dus­try con­sol­i­dat­ing into 2 oli­garchi­cal man­u­fac­tur­ers. One might be tempted to ar­gue that the Thai hard drive floods are a poor case-s­tudy be­cause of these idio­syn­cratic fac­tors; but that’s miss­ing the point: un­til a dis­rup­tion hap­pens, how can one know—e­spe­cially an out­sider—what will or will not be a prob­lem? Are there no long-term con­tracts for proces­sors, and is not In­tel largely dom­i­nant? And what are the idio­syn­cratic fac­tors rel­e­vant to proces­sors? There must be some, even if we would not ex­pect them to be the same as in the Thai hard drive floods.↩︎

  18. “Thai­land Flood­ing Crip­ples Hard-Drive Sup­pli­ers”, The New York Times:

    The slow-mov­ing flood­wa­ters, which are an ac­cu­mu­la­tion from this year’s un­usu­ally strong mon­soon rains in north­ern Thai­land, are grad­u­ally drain­ing into the sea. At what is known as the Bang Pa-In In­dus­trial Es­tate, trucks have de­liv­ered mas­sive pumps. Work­ers said they would start try­ing to re­move wa­ter from the area on Mon­day. The flood­wa­ters de­scended to this area an hour north of Bangkok in early Oc­to­ber. Efforts to de­fend in­dus­trial ar­eas with sand­bags and other bar­ri­ers were fu­tile.

    “There was no way we could have held back the wa­ter,” said Sam­ruay Pakubol, a welder at an au­to­mo­tive parts fac­tory here. Now out of work, he takes pas­sen­gers on a wooden boat down the streets of the in­dus­trial zone. Work­ers have caught and killed croc­o­diles swim­ming in the area, he said.

    Dale Schudel, man­ag­ing di­rec­tor of In­triPlex Thai­land, a com­pany that makes com­po­nents for hard-disk dri­ves, said his fac­tory in nearby Ayut­thaya had flood­wa­ters al­most six feet deep. But pump­ing out the wa­ter, which will take about two to three weeks, is only the be­gin­ning of the cleanup. Mr. Schudel de­scribed the wa­ter as “highly cor­ro­sive.” “I think you have to ask your­self, if any fac­tory in the world were sub­merged in that much wa­ter, how much dam­age would there be?” he said.

  19. I think drones are an in­ter­est­ing is­sue, but right now they’re look­ing like a mas­sive shift to­ward gov­ern­men­t/­cor­po­rate pow­er: in 20 years, you may be able to afford an awe­some drag­on­fly drone which you’d like to pi­lot into a chip fab and up­load some evil (note I’m not even both­er­ing to think about the sce­nario in which you load the drone with 20kg of high ex­plo­sives and crash it into a tar­get), but your drone won’t be able to get through the thou­sands of pa­trolling se­cu­rity drones spread out over the en­tire in­stal­la­tion! En­thu­si­asts right now may be play­ing around with nifty quadropters etc… and the US gov­ern­ment is prob­a­bly play­ing around with drones that never come down, have radar & sur­veil­lance/de­tec­tion ca­pa­bil­i­ties that us civil­ians can’t even guess at, and with net­worked fleets of them.↩︎

  20. is help­ful as usu­al, and a quick Google turns up a ran­dom lob­by­ing group giv­ing some num­bers (mul­ti­ply by 1000 to get megawat­t-hours):

    Ac­cord­ing to a 2008 Elec­tric Power Re­search In­sti­tute study, a con­ven­tional com­bined-cy­cle nat­ural gas plant costs about $1,000 per kilo­watt of ca­pac­ity con­struct­ed. A coal-fired plant costs more than $2,500 per KW hour to build. The cost of a new nu­clear plant is more than $4,000 per KW of ca­pac­i­ty. Wind gen­er­a­tion costs are about dou­ble nat­ural gas in­stal­la­tion costs. A new so­lar plant in Florida is pro­jected at about $6,600 per KW.

  21. It’s a lit­tle like try­ing to get Amer­i­can lib­er­tar­i­ans to en­gage in effec­tive col­lec­tive ac­tion like lob­by­ing or —if they were in­clined to be com­mu­nal and bow to hi­er­ar­chy, they’d not be lib­er­tar­i­ans but Re­pub­li­cans or some­thing!↩︎

  22. Less­Wrong is no ex­cep­tion. It al­ways amuses me that, as much as it gets called a cult, when you ac­tu­ally ask LW­ers what they be­lieve in sur­veys (2009, 2011), you find that the stereo­typ­i­cal Less­Wrong shib­bo­leth­s—Strong AI the­sis is true, AI will be ac­com­plished dur­ing the 21st cen­tu­ry, is the right eth­i­cal par­a­digm like lib­eral lib­er­tar­i­an­ism is the right po­lit­i­cal par­a­digm and athe­ism for re­li­gion, is more true than other quan­tum me­chan­i­cal in­ter­pre­ta­tions, is a good idea etc.—are ac­tu­ally held by bare ma­jori­ties and some­times pretty small mi­nori­ties. For ex­am­ple:

    • lib­er­tar­i­an­ism (32%) is rarer than lib­er­al­ism (35%)
    • con­se­quen­tial­ists are only 62% of re­spon­dents (a big sur­prise to me) and Many Worlders 56%
    • ‘prob­a­bil­ity cry­on­ics will work’ av­er­aged 21%; 4% of LW­ers are signed up, 36% op­posed, and 54% merely ‘con­sid­er­ing’ it
    • AI is feared less than pan­demics (26% vs 17%)
    • the me­dian Sin­gu­lar­ity is 2080 (not a Kurzweil­ian 2030–2040)
  23. This is an­other in­stance where de­tailed do­main knowl­edge would be help­ful. Shock waves dis­si­pate fast: ba­sic geom­e­try sug­gests they weaken as roughly the cube of the dis­tance be­cause the en­ergy is be­ing dis­si­pated over a spher­i­cal vol­ume. More pre­cise­ly, from an old nu­clear weapons text­book, I learned it was an in­ter­me­di­ate power be­tween in­verse square (a fixed en­ergy spread­ing over a 2D area) and in­verse cube (over 3D vol­ume) be­cause they hit bound­ary lay­ers and re­flect. Ei­ther way, a blast may need to be very close to a key part of a chip fab to mat­ter. This leads to some in­ter­est­ing trade­offs in tar­get­ing: So­viet nukes tended to have higher mega­ton­nage than Amer­i­can nukes, de­spite the lat­ter’s pre­sum­able tech­ni­cal su­pe­ri­or­i­ty, be­cause their tar­get­ing was in­ac­cu­rate and the higher mega­ton­nage made up for the in­ac­cu­racy via overkill; Amer­i­can mis­siles fa­vored mul­ti­ple war­heads on , be­cause the high ac­cu­racy meant that a tar­get could be brack­eted be­tween sev­eral small nukes and the con­verg­ing shock­waves much more effec­tive.↩︎

  24. in men­tions that heroin syn­the­sis was in­tro­duced to one re­gion by an old Chi­nese chemist with 7 dis­ci­ples—so many, pre­sum­ably, be­cause he had only one arm left and needed as­sis­tants.↩︎

  25. Note that this claim is dou­ble-edged: if it will be­come im­pos­si­ble in 2 gen­er­a­tions, then we can in­fer that it must be pos­si­ble ‘now’ (2003) at some very high price, and given Moore’s sec­ond law, may have been pos­si­ble at a rea­son­able price in the past (<1980s?).

    The NSA is a known ex­am­ple: among their many fa­cil­i­ties like the , they in their Army fa­cil­i­ties at (op­er­ated by Na­tional Semi­con­duc­tor Cor­po­ra­tion). There are many com­pelling ap­pli­ca­tions for an NSA chip fab to the point where it’s hard to guess what its pri­mary mis­sion might be:

    1. cus­tom fab­ri­ca­tion of rel­a­tively low tech but ex­tremely se­cu­ri­ty-sen­si­tive chips: per­haps en­cryp­tion chips for US gov­ern­ment use. (Nu­clear bomb com­po­nents have been sug­gested but would­n’t fall un­der their dual man­date of se­cur­ing US com­mu­ni­ca­tions and break­ing for­eign com­mu­ni­ca­tion­s.) Mak­ing cus­tom, lega­cy, and ob­scure chips is the offi­cial ex­pla­na­tion:

      The orig­i­nal idea was to ser­vice NSA’s own needs in the in­tel­li­gence are­na, par­tic­u­larly for old com­po­nents that ex­ec­u­tives of com­mer­cial semi­con­duc­tor firms are ei­ther un­able or un­will­ing to pro­duce [due to clo­sures of chip fabs, eg. SVTC]. Now the fa­cil­ity is ramp­ing up to ser­vice the needs of other fed­eral agen­cies and pro­grams, and is offer­ing a com­plete one-stop shop­ping cen­ter for de­sign, fab­ri­ca­tion, as­sem­bly, and test. This is ac­cord­ing to a sales pitch pre­sented by Le­land Miller, NSA’s mar­ket­ing di­rec­tor for mi­cro­elec­tron­ics, at the Oc­to­ber As­so­ci­a­tion of Old Crows (AOC) con­fer­ence in Vir­ginia Beach, Va. The man­u­fac­tur­ing process is bulk CMOS, us­ing 6-inch wafers, to fab­ri­cate fea­ture sizes of 1, 0.8, and 0.5 mi­crons (1000/800/500nm) in ei­ther two or three metal lay­ers. This is due to be up­graded to 0.22 mi­cron (200n­m), six-level metal tech­nol­ogy by July 2000. Typ­i­cal pro­duc­tion amounts to 1,000 wafer starts a mon­th, Miller says. Yields are com­pa­ra­ble to the in­dus­try av­er­age, he adds—around 15% for large die and 80–90% for small die. Out­put has risen sharply from 350,000 to 450,000 die over the past year, Miller notes. It was about 3,000 die when the fa­cil­ity be­gan at the 1-mi­cron lev­el. About 150 prod­uct de­signs are in the fab­ri­ca­tion process si­mul­ta­ne­ous­ly. The NSA fa­cil­ity has con­cen­trated on the agen­cy’s own needs, par­tic­u­larly on old 1-mi­cron prod­ucts, but has ser­viced other gov­ern­ment cus­tomer­s…In ad­di­tion to fab­ri­ca­tion and pack­ag­ing of pro­gram­ma­ble gate ar­rays in plas­tic mul­ti­chip mod­ules, flipchips, and ball grid ar­rays, the fa­cil­ity also can do cus­tom de­signs of ap­pli­ca­tion-spe­cific in­te­grated cir­cuits us­ing any start­ing point from block di­a­grams to fin­ished lay­outs. Typ­i­cal de­liv­ery times are 12 weeks from sub­mis­sion of a data­base tape, but spe­cial or­ders can be han­dled in three weeks. Func­tional and para­met­ric tests are con­ducted us­ing stan­dard com­mer­cial test equip­ment.

      This seems like the most likely ex­pla­na­tion: it’s much eas­ier for a small plant to keep up if it’s just mak­ing old out­dated chips.

    2. cus­tom fab­ri­ca­tion of high­ly-spe­cial­ized chips for use in code-break­ing and sur­veil­lance: well-de­signed ASICs can offer over com­mod­ity CPUs or GPUs.

      More in­trigu­ing is the ob­ser­va­tion by David Honig that one pos­si­bil­ity is that the NSA is en­gaged in (men­tioned in one NSA job ad): e-beam is far too slow for profitable com­mer­cial chip pro­duc­tion, but it has the ma­jor ad­van­tage that it can pro­duce smaller chip fea­tures. For com­par­ison, the Wikipedia ar­ti­cle says e-beam sys­tems have worked at “~20nm since at least 1990” (with the state of the art be­ing <10nm) while In­tel only be­gan com­mer­cially sell­ing chips 22 years later in April 2012.

      There would be lit­tle rea­son for the NSA to pro­duce its own CPUs and GPUs with e-beam, and there are plenty of com­mod­ity ASIC mak­ers if the NSA sim­ply wanted a highly par­al­lel Bit­coin-style chip for brute-forc­ing SHA hash­es, since in both cases out­sourc­ing would be cheaper (one sim­ply buys a lot of SHA-cracking ASICs since the prob­lem is em­bar­rass­ingly par­al­lel); see for ex­am­ple, NSA’s IBM de­signed/built Wind­sor­Green/Wind­sor­Blue su­per­com­puter. But there is some­thing that ex­pen­sive tiny e-beam-made chips could be su­pe­rior at: run­ning se­r­ial prob­lem­s—where one can­not sim­ply par­al­lelize over many CPUs/GPUs/ASICs—faster than any­one else can. What NSA prob­lems might ben­e­fit from a few cus­tom ex­pen­sive very fast se­r­ial chips, I do not know; noth­ing in the Snow­den leaks I’ve seen has jumped out as a can­di­date.

    3. re­verse-engi­neer­ing for­eign-pro­duced chips: ei­ther to try to break their sys­tems, or look for back­doors in Amer­i­can-used chips (or per­haps for back­doors which can then be ex­ploited by the NSA).

    4. test out back­doors of their own on sim­ple test­bed chips, ei­ther for use in com­mer­cial plants’ chips or to test out back­door-tests for #3

  26. On a per­sonal note, my grand­fa­ther served in Ko­rea dur­ing the Ko­rean War. He was never in any dan­ger be­cause he was de­ployed well away from the front lines. We don’t know what he did, but he in­sists it is clas­si­fied, and hinted it in­volved nu­clear ma­te­r­i­al. (He spent much of his later ca­reer in­volved in gas masks.) This would be con­sis­tent with the dur­ing the Ko­rean War.↩︎

  27. pg 413–414 of :

    Though the United States and the USSR con­tin­ued to de­velop nu­clear tech­nol­ogy at a break­neck pace, they be­gan, how­ever hyp­o­crit­i­cal­ly, to pay homage to nu­clear dis­ar­ma­ment in con­fer­ences and state­ments. At the same time a grass­roots move­ment be­gan to stig­ma­tize the weapons. Demon­stra­tions and pe­ti­tions at­tracted mil­lions of cit­i­zens, to­gether with pub­lic fig­ures such as Li­nus Paul­ing, Bertrand Rus­sell, and Al­bert Schweitzer. The mount­ing pres­sure helped nudge the su­per­pow­ers to a mora­to­rium and then a ban on at­mos­pheric nu­clear test­ing, and then to a string of arm­s-con­trol agree­ments. The Cuban Mis­sile Cri­sis in 1962 was a tip­ping point. Lyn­don John­son cap­i­tal­ized on the change to de­mo­nize Gold­wa­ter in the Daisy ad and called at­ten­tion to the cat­e­gor­i­cal bound­ary in a 1964 pub­lic state­ment: “Make no mis­take. There is no such thing as a con­ven­tional nu­clear weapon. For nine­teen per­il-filled years no na­tion has loosed the atom against an­oth­er. To do so now is a po­lit­i­cal de­ci­sion of the high­est or­der.”205

    As the world’s luck held out, and the two nu­clear-free decades grew to three and four and five and six, the taboo fed on it­self in the run­away process by which norms be­come com­mon knowl­edge. The use of nu­clear weapons was un­think­able be­cause every­one knew it was un­think­able, and every­one knew that every­one knew it. The fact that wars both large (Viet­nam) and small (Falk­lands) were not de­terred by the in­creas­ingly in­effec­tual nu­clear threat was a small price to pay for the in­defi­nite post­pone­ment of Ar­maged­don.

    …One hope­ful sign is that nu­clear pro­lif­er­a­tion has not pro­ceeded at the fu­ri­ous rate that every­one ex­pect­ed. In the 1960 pres­i­den­tial elec­tion de­bates, John F. Kennedy pre­dicted that by 1964 there might be “ten, fifteen, twenty” coun­tries with nu­clear weapons.206 The con­cern ac­cel­er­ated when China con­ducted its first nu­clear test in 1964, bring­ing the num­ber of na­tions in the nu­clear club to five in less than twenty years. Tom Lehrer cap­tured pop­u­lar fears of run­away nu­clear pro­lif­er­a­tion in his song “Who’s Next?” which ran through a list of coun­tries that he ex­pected would soon be­come nu­clear pow­ers (“Lux­em­bourg is next to go / And who knows? Maybe Monaco”).

    But the only coun­try that ful­filled his prophecy is Is­rael (“‘The Lord’s my shep­herd,’ says the Psalm / But just in case-we bet­ter get a bomb!”). Con­trary to ex­pert pre­dic­tions that Japan would “un­equiv­o­cally start on the process of ac­quir­ing nu­clear weapons” by 1980 and that a re­uni­fied Ger­many “will feel in­se­cure with­out nu­clear weapons,” nei­ther coun­try seems in­ter­ested in de­vel­op­ing them.207 And be­lieve it or not, since 1964 as many coun­tries have given up nu­clear weapons as have ac­quired them. Say what? While Is­rael, In­dia, Pak­istan, and North Ko­rea cur­rently have a nu­clear ca­pa­bil­i­ty, South Africa dis­man­tled its stash shortly be­fore the col­lapse of the apartheid regime in 1989, and Kaza­khstan, Ukraine, and Be­larus said “no thanks” to the ar­se­nals they in­her­ited from the de­funct So­viet Union. Al­so, be­lieve it or not, the num­ber of non­nu­clear na­tions that are pur­su­ing nu­clear weapons has plum­meted since the 1980s. Fig­ure 5-22, based on a tally by the po­lit­i­cal sci­en­tist Scott Sagan, charts the num­ber of non­nu­clear states in each year since 1945 that had pro­grams for de­vel­op­ing nu­clear weapons.

    The downslopes in the curve show that at var­i­ous times Al­ge­ria, Aus­tralia, Brazil, Egypt, Iraq, Libya, Ro­ma­nia, South Ko­rea, Switzer­land, Swe­den, Tai­wan, and Yu­goslavia have pur­sued nu­clear weapons but then thought the bet­ter of it-oc­ca­sion­ally through the per­sua­sion of an Is­raeli air strike, but more often by choice.

    Pinker fo­cuses on the nu­clear taboo, but can we ig­nore en­tirely the global re­stric­tions on nu­clear tech­nol­ogy trans­fer and uni­ver­sal state reg­u­la­tion of any­thing to do with nu­clear tech­nol­o­gy? The par­al­lels to semi­con­duc­tor chip tech­nol­ogy are clear.↩︎

  28. pg124–125, (em­pha­sis added):

    In the prob­lem of de­cod­ing, the most im­por­tant in­for­ma­tion which we can pos­sess is the knowl­edge that the mes­sage which we are read­ing is not gib­ber­ish. A com­mon method of dis­con­cert­ing code­break­ers is to mix in with the le­git­i­mate mes­sage a mes­sage that can­not be de­cod­ed; a non-sig­nifi­cant mes­sage, a mere as­sem­blage of char­ac­ters. In a sim­i­lar way, when we con­sider a prob­lem of na­ture such as that of atomic re­ac­tions and atomic ex­plo­sives, the largest sin­gle item of in­for­ma­tion which we can make pub­lic is that they ex­ist. Once a sci­en­tist at­tacks a prob­lem which he knows to have an an­swer, his en­tire at­ti­tude is changed. He is al­ready some 50% of his way to­ward that an­swer.

    In view of this, it is per­fectly fair to say that the one se­cret con­cern­ing the atomic bomb which might have been kept and which was given to the pub­lic and to all po­ten­tial en­e­mies with­out the least in­hi­bi­tion, was that of the pos­si­bil­ity on its con­struc­tion. Take a prob­lem of this im­por­tance and as­sure the sci­en­tific world that it has an an­swer; then both the in­tel­lec­tual abil­ity of the sci­en­tists and the ex­ist­ing lab­o­ra­tory fa­cil­i­ties are so widely dis­trib­uted that the qua­si­-in­de­pen­dent re­al­iza­tion of the task will be a mat­ter of merely a few years any­where in the world.

  29. pg 39–40:

    To put it at its most el­e­men­tary, while ob­serv­ing oth­ers rid­ing bi­cy­cles does not en­able one to learn the skills of the cy­clist, it nev­er­the­less shows that cy­cling is pos­si­ble. Know­ing that older broth­ers or sis­ters have learned to ride can en­cour­age younger sib­lings not to con­clude from early fail­ures that the task is im­pos­si­bly hard.

    …The con­fi­dence—in­deed over­con­fi­dence—of wartime An­glo-Amer­i­can physi­cists (in­clud­ing Con­ti­nen­tal refugees) in the ease of de­vel­op­ment of a nu­clear weapon does not seem to have been widely shared by their French, Ger­man, or So­viet col­leagues, and the gov­ern­ments of the last two coun­tries were un­con­vinced prior to 1945 that the task was fea­si­ble enough to be worth the kind of re­sources the Amer­i­cans de­voted to it (see, e.g., Hol­loway 1981; Gold­schmidt 1984, p. 24).24 Trin­i­ty, Hi­roshi­ma, and Na­gasaki were dra­matic demon­stra­tions that the task was not im­pos­si­bly hard, and this proof (as well, of course, as the per­ceived threat to the So­viet Union) ex­plains the sud­den shift in the USSR in 1945 from a mod­est re­search effort to an al­l-out, top-pri­or­ity pro­gram (Hol­loway 1981).

    As we have seen, the British test ex­plo­sion in 1952, al­though no threat to France, con­tributed to the lat­ter’s weapons pro­gram by sug­gest­ing that de­vel­op­ing an atomic bomb was eas­ier than had pre­vi­ously been as­sumed. Like­wise, the Chi­nese ex­plo­sion in 1964 showed other de­vel­op­ing coun­tries that the atomic bomb was not nec­es­sar­ily the pre­serve solely of the highly in­dus­tri­al­ized world. Fur­ther­more, pro­found ques­tions over the fea­si­bil­ity of early hy­dro­gen bomb de­signs helped de­lay the Amer­i­can move from an atomic to a hy­dro­gen bomb (Bethe 1982). By con­trast, all sub­se­quent hy­dro­gen bomb pro­grams could pro­ceed with con­fi­dence in the ba­sic achiev­abil­ity of their goal, and, in words used in an­other con­text by a group of weapons de­sign­ers (Mark et al. 1987, p. 64), “The mere fact of know­ing [some­thing] is pos­si­ble, even with­out know­ing ex­actly how, [can] fo­cus … at­ten­tion and efforts.”

  30. tech­ni­cal re­port M-85/11 “Lit­er­a­ture Sur­vey of Un­der­ground Con­struc­tion Meth­ods for Ap­pli­ca­tion to Hard­ened Fa­cil­i­ties”:

    An eval­u­a­tion of a nu­clear power plant con­cept [79] re­vealed that lo­cat­ing the fa­cil­ity un­der­ground with a cut and cover tech­nique would be 11% more ex­pen­sive than an above-ground plan. The in­creased cost was at­trib­uted to di­rect con­struc­tion costs be­ing 70% high­er, the need for spe­cial equip­ment for ven­ti­la­tion and other func­tions, and the ad­di­tional time re­quired to build the un­der­ground struc­ture. More costs are in­curred from hard­en­ing un­der­ground tun­nels to re­sist blasts or seis­mic loads. A de­sign cost study [85] es­ti­mates that hard­en­ing a tun­nel to re­sist a seis­mic load of 0.5 g would in­crease con­struc­tion costs by 35%…Un­der­ground sub­way sta­tion con­struc­tion costs were also com­pared to show that an un­der­ground sta­tion us­ing a tun­neled earth ex­ca­va­tion tech­nique with an 85-ft (25.5-m) over­bur­den would cost about 25% more than one con­structed by cut and cover meth­ods with a 20-ft (6-m) earth cov­er, an 47% more than a cut and cover sta­tion with a 6-ft (1.8-m) cov­er…Ref­er­ence 77 com­pares the cost of sit­ing a nu­clear power plant un­der­ground. The in­ves­ti­ga­tion found that a cut and cover buried fa­cil­ity would cost 14–25% more and a mined rock plant 10–18% more than a sur­face power plant. A sec­ond re­port [86] states that costs for sit­ing a nu­clear power plant un­der­ground in rock are about 25% more. Ref­er­ence 80 ex­am­ines the costs of un­der­ground homes and large pub­lic build­ings. Based on life-cy­cle cost fig­ures of 5 case stud­ies: ‘It does ap­pear clear, how­ev­er, that the use of earth­-shel­ter­ing does not in­crease con­struc­tion costs in any no­table way, and may in fact rep­re­sent a de­crease in some cases’ [80]. An ex­am­ple earth­-shel­tered house is cited as cost­ing 28% more to con­struct, but 12–20% less to own and op­er­ate over the 30-year life of the home.

  31. Lieber­man 2003:

    The Chi­nese gov­ern­ment is suc­cess­fully us­ing tax sub­si­dies (see be­low) to at­tract for­eign cap­i­tal from semi­con­duc­tor firms seek­ing ac­cess to what is ex­pected to be one of the world’s largest mar­kets. This strat­e­gy, which is sim­i­lar to that em­ployed by the Eu­ro­pean Union in early 1990s, is a means of in­duc­ing sub­stan­tial in­flows of di­rect in­vest­ment by pri­vate firms. In­deed, much of the fund­ing is Tai­wane­se, dri­ven by the tax in­cen­tives and their need for mar­ket ac­cess, es­pe­cially for com­mod­ity prod­ucts such as DRAMs. The strat­egy does not rely on cheaper labor, as that is a small el­e­ment in semi­con­duc­tor pro­duc­tion.

    The Chi­nese are, how­ev­er, able to in­creas­ingly draw on sub­stan­tially larger pools of tech­ni­cally trained la­bor as com­pared to the U.S., from the large co­horts of do­mes­tic en­gi­neer­ing grad­u­ates.17 Im­por­tant­ly, the out­put of Chi­nese uni­ver­si­ties is sup­ple­mented by large num­bers of en­gi­neers trained at U.S. uni­ver­si­ties and mid-ca­reer pro­fes­sion­als who are offered sub­stan­tial in­cen­tives to re­turn to work in Chi­na. These in­cen­tives for sci­en­tists and en­gi­neers, which in­clude sub­stan­tial tax ben­e­fits, world-class liv­ing fa­cil­i­ties, ex­ten­sive stock op­tions taxed at par val­ue, and other ameni­ties, are prov­ing effec­tive in at­tract­ing ex­pa­tri­ate la­bor. For ex­am­ple, the Chi­nese cen­tral gov­ern­ment has un­der­taken in­di­rect sub­si­dies in the form of a sub­stan­tial re­bate on the val­ue-added tax (VAT) charged on Chi­ne­se-made chip­s.19 While many be­lieve this is an il­le­gal sub­sidy un­der GATT trade rules, the im­pact of the sub­sidy on the growth of the in­dus­try may well be ir­re­versible be­fore-and if-any trade ac­tion is tak­en. There are a va­ri­ety of other doc­u­mented mea­sures adopted by the Chi­nese gov­ern­ment.20…Cur­rently the Chi­nese gov­ern­ment is pro­vid­ing a 14% re­bate on VAT to cus­tomers who buy Chi­ne­se-made semi­con­duc­tor chips, es­sen­tially pro­vid­ing a large sub­sidy of their do­mes­tic in­dus­try in clear vi­o­la­tion of GATT rules.14 Thus, U.S.-made chips would pay a 17% VAT, and Chi­ne­se-made chips would pay a 3% VAT. Given the tight price com­pe­ti­tion of chips and the grow­ing im­por­tance of the Chi­nese chip mar­ket, this is a very [im­por­tant] step to­wards end­ing U.S. pro­duc­tion. [This VAT ap­par­ently was re­pealed.]

    …There are a va­ri­ety of other doc­u­mented mea­sures adopted by the Chi­nese gov­ern­ment.20 The de­vel­op­ment of spe­cial gov­ern­ment funded in­dus­trial parks, the low costs of build­ing con­struc­tion in China as com­pared to the U.S., and their ap­par­ent dis­in­ter­est in the ex­pen­sive pol­lu­tion con­trols re­quired of fab­ri­ca­tion fa­cil­i­ties in the U.S. all rep­re­sent fur­ther hid­den sub­si­dies. The ag­gre­gate effect of these in­di­vid­ual “sub­si­dies” may be only a few tens of per­cent­age points of de­crease (lit­er­al­ly, only 20–30%) in the man­u­fac­tur­ing costs of the chips, but in such a cost-driven in­dus­try, this differ­ence ap­pears to play an im­por­tant role in dri­ving the en­tire off­shore mi­gra­tion process for these crit­i­cal com­po­nents. Es­sen­tial­ly, these ac­tions re­flect a strate­gic de­ci­sion and rep­re­sent a con­certed effort by the Chi­nese gov­ern­ment to cap­ture the ben­e­fits of this en­abling, high­-tech in­dus­try, and thereby threat­en­ing to be a mo­nop­oly sup­plier and thus in con­trol of pric­ing and sup­ply.

  32. Brown & Lin­den 2005:

    From 1984 to 2004, the share of semi­con­duc­tor sales in Asia, in­clud­ing Japan, has risen from 38 to 63% of the world to­tal.10

    …The shift of ca­pac­ity from Japan and the United States to the rest of Asia (pri­mar­ily South Ko­rea and Tai­wan) is strik­ing. Japan and the United States ac­counted for 80% of fab ca­pac­ity in 1980, but only 49% of ca­pac­ity in 2001…In 2001, ap­prox­i­mately one-third of U.S.-owned ca­pac­ity was lo­cated off­shore as shown in Ta­ble 5. The off­shore fabs were pri­mar­ily in Japan and Eu­rope, which re­flects the rise of joint ven­tures to share risk as the cost of fabs in­creased. Con­verse­ly, about 22% of the fab ca­pac­ity lo­cated in North Amer­ica was owned by com­pa­nies based in other re­gions (not shown).

    …The prospects vary greatly by the in­sti­tu­tional en­vi­ron­ment in each lo­ca­tion. As dis­cussed above, Tai­wan’s fa­b­less sec­tor, which did not arise as an out­growth of U.S. de­sign off­shoring, is nearly a gen­er­a­tion be­hind U.S. ri­vals in terms of in­no­v­a­tive prod­ucts. For now, lo­cal firms in In­dia have gen­er­ally avoided the fa­b­less mod­el, but in China there is a small but in­creas­ing num­ber of fa­b­less firms tar­get­ing world mar­kets. Al­though for now Chi­nese firms lack ex­pe­ri­enced en­gi­neers and man­agers and are be be­hind Tai­wan in their de­vel­op­ment of in­no­v­a­tive prod­ucts, this will grad­u­ally shift in the years ahead. It is too early to know where this process will end.