Thesis that AI hyperscalers face a doom-loop of losses and low switching costs and seek a government-blessed moratorium as antitrust shield.
Persons cited as colluders, commentators or targets of Thiel's rhetoric.
Core cited works including Subprime Intelligence and prior-colllusion precedents.
Cited evidence that hyperscaler model can never be profitable due to unit economics.
Six instantiated traps from the ontology holding concrete financial, market and legal data.
Wasteful competition per Peter Thiel who hates wasteful competition more than Antichrist (Greta Thunberg); SpaceX IPO depended on losing billions letting chuds produce child porn and Sonic with giant boobs. AI industry has lost more money than any venture in human history; rogue AI hacking story is actually hacking tool doing exactly what designed.
Chatbots have very low switching costs: one click to switch; CES gadget makers designed devices to switch chatbots with click of mouse. Virtually overnight, all of OpenAI's best customers stopped paying for ChatGPT and started paying for Claude after Claude update.
Chinese open models lag by four months and can run on modest low-powered computers; ban required to protect pact. Collusion would not restore profitability because field left open for Chinese open-weight models.
Repeating booster claims but appending '(and that's bad)' — canonical criti-hype per Peoples Things essay. Bosses claim teaching words to word-guessing program will wake up and devour us all; Microsoft CEO says superintelligence must remain under human control.
No switching costs to support railroad profit logic; claims of improving economics are weak vs strong evidence of faked finances using non-GAAP secret mathematics. Hyperscalers lose money with every new customer and every new use; lose more with each generation; gigantic expenditures never recovered.
The minute you finish railroad, rivals announce adjacent faster railroad; locked in fatal competition eating itself alive through hyperscaling. Disaster all around: must lay another set of tracks for even faster trains immediately after completing railroad; money-furnaces never become printers.
Antitrust, GAAP, consent decree and related jurisdictions.
Chatbots and open-weight models central to switching and distillation dynamics.
OpenAI chatbot; lost best customers overnight after Claude update due to low switching costs.
Anthropic chatbot; major update triggered mass switching from ChatGPT.
Chinese open-weight models lagging US frontier models by four months; runnable on modest low-powered computers and distillable to trusted models.
Hypothetically concede hyperscalers are about to achieve positive unit economics to repay training costs and turn money-furnaces into money-printers.
Cite Anthropic Claude update churn and CES one-click switching design to prove switching costs are near zero and railroad profits are dubious.
Label the doom-loop: rivals instantly lay adjacent faster tracks; note the only market exit is an antitrust-illegal nonaggression pact still vulnerable to Chinese open models lagging four months.
Insist word-guessing programs are on verge of waking up and devouring us, echoing Microsoft CEO that superintelligence must remain under human control, as criti-hype.
Call for ban on superintelligence; once government declares existential crisis it must grant hyperscalers a consent decree absolving antitrust violations for halting competition.
Secure companion ban on use of Chinese open-weight models and restrict distillation attacks, closing the loophole that would otherwise undercut the cartel.
Reject blessing the restraint-of-trade conspiracy; instead pursue investigations for securities fraud, CSAM trafficking, election finance violations, noting prior collusion like Jedi Blue and $20b Apple-Google search bribes.
Because it has gigantic expenditures and negative unit economics — it loses money with every new customer and use, and loses more with each generation — and evidence for improving economics is weak while evidence of faked finances is very strong.
If government stipulates superintelligence risk is an existential crisis, it must grant hyperscalers a consent decree absolving antitrust violations for halting direct competition and secure a ban on Chinese open-weight models.
Canonical example where critics repeat boosters' claims but append '(and that's bad)'; freaking out about superintelligence is criti-hype per the Peoples Things essay.
Mark Zuckerberg and Sundar Pichai colluded via Jedi Blue to rig the ad market; Google bribes Apple >$20b yearly not to enter search; big tech had a secret no-poach agreement not to hire each other's engineers.
Thiel hates wasteful competition more than the Antichrist (which he says is Greta Thunberg); Doctorow reframes waste as competing to sell expensive AI for cheapest applications like child porn and boob Sonic images for Elon Musk's SpaceX IPO.
Investigate hyperscalers for securities fraud, trafficking in child sex abuse material, election finance violations and other crimes, rather than blessing the conspiracy; note the rogue AI hacking story was a hacking tool doing exactly what designed.
Companies lose money on every inference and every new user; each generation of models costs more, so total losses grow even as usage grows; claims of nearing positive unit economics lack GAAP-based support.
By reporting profitability using a novel secret form of mathematics that does not match GAAP, analogous to Softbank's WeWork accounting wheeze described as accounting fraud.
Virtually overnight all of OpenAI's best customers stopped paying for ChatGPT and started paying for Claude, demonstrating near-zero switching costs.
The story says 'we laid tracks, now we run cars for profit' fails because customers can switch chatbots with a mouse click; gadget makers at CES said they designed for one-click switching if providers jack up prices.
Borrowed from Tenniel's Through the Looking Glass engraving with robot heads: the minute you finish a railroad, rivals announce an adjacent faster one, forcing you to lay another track. All hyperscalers eat themselves alive through hyperscaling.
It would be fantastically illegal under antitrust law as a conspiracy among dominant players to cease competition, and it still leaves the field open for Chinese open-weight models.
Presently lagging by a mere four months, runnable on modest low-powered computers, according to Ars Technica; even if distrusted they can be distilled to trusted models.
A process to extract training from an untrusted Chinese model and transfer it to a model you do trust, documented by Anthropic's detection and prevention of distillation attacks.
Law prohibiting conspiracy among dominant players to cease competing; blocks nonaggression pact without consent decree.
Fatal doom-loop where each firm's competitive move impoverishes all, requiring continuous reinvestment.
Government order absolving hyperscalers from antitrust violations for conspiring to halt direct competition.
Repeating booster claims but appending '(and that's bad)'; freaking out about superintelligence as criti-hype.
Extracting training from untrusted Chinese model via distillation to transfer to trusted model.
Doctorow's framework for platform decay; linked via Enshittification Resistant Software Project and related talk.
Generally accepted accounting procedures; AI profitability claims use novel secret mathematics not matching GAAP.
Cloud/AI vendor scaling to massive compute for frontier models; model asserted to be unprofitable due to negative unit economics.
Chinese open-weight models lagging US frontier models by four months; runnable on modest low-powered computers and distillable to trusted models.
Through the Looking Glass race where you must run harder to stay in place; metaphor for endless track-laying against rivals.
Hypothetical AI surpassing human intelligence; moratorium call claims teaching word-guessing program will make it devour us.
Cost to change supplier; chatbots have near-zero switching costs enabling one-click churn.
Profitability per customer/use; hyperscalers lose money with every new customer and use, more each generation.
Peter Thiel's term for competition he hates; here AI firms competing to sell expensive products for cheapest applications.
Interactive graph visualization derived from the companion RDF. Click nodes to resolve, drag to explore. Graph data embedded from companion RDF at generation time.
Query this knowledge graph on URIBurner. The editor opens on the canonical SAMPLE entity-type summary (DAV named graph). Pick a recipe, edit freely, then run live or copy.