The Scale, The Plan, and The People — No One's Happy

# The Scale, The Plan, and The People — No One's Happy Source: [https://nooneshappy.com/article/the-scale-the-plan-and-the-people/](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/) May 20, 2026My previous post documented what happens to people and teams when output is decoupled from understanding\. That was the micro picture: what the pattern looks like from inside the room where the work is supposed to happen\. The response was larger than I expected\. A considerable number of readers wrote in to corroborate the pattern from inside their own organizations\. Because of this, I am drafting guidelines on recommended organizational AI use, downstream of those conversations, which will appear separately and soon\. If you’d like to contribute please email\. This article takes up the macro question\. If the pattern is as visible as the feedback suggests, if every consulting firm keeps utilizing agentic tools and workflows that produce subpar output, if every enterprise keeps buying tools that deliver little to no measurable productivity gain, the question worth asking is structural\. Why does a system that produces so many failures of this shape keep being built, sold, and scaled? It requires considering various facets of the current landscape: what the people closest to the architecture have begun to say about what these systems can and cannot do; the scale of the infrastructure bet being made on their behalf; the financial structure that locks each organization into continuing the spend; why if opinions are becoming more modest, spend is not following\. ## The age of scaling, in retrospect The people who built the current generation of AI systems have begun saying, in public, that the approach that got them here will not get them further\. Ilya Sutskever, the OpenAI co\-founder and former chief scientist whose work in the 2010s established the scaling hypothesis the entire industry runs on, told Dwarkesh Patel in late 2025 that the age of scaling — which he dated with characteristic precision as 2020 through 2025 — is over\. “Scaling the pre\-training is essentially tapped out,” he said\. “If you just multiply the scale by 100x now, you won’t get a qualitative change in capability\.” \[[1](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-1)\] The recipe, on his account, is finished; what comes next requires research\. Demis Hassabis, who runs Google DeepMind, put the window to AGI at five to ten years in October 2025, and in the same month publicly corrected an OpenAI researcher who had overclaimed results on a math benchmark \[[2](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-2)\] — a small, in\-room professional discipline that does not happen unless he has stopped believing the field can afford the inflation\. These are not fringe critics\. Yann LeCun left Meta in November 2025 to serve as Executive Chairman of Advanced Machine Intelligence Labs, a company built on the premise that the dominant architecture is a dead end\. He raised $1\.03 billion against that claim in March 2026, and has said publicly that nobody in their right mind will be using large language models of the current type within three to five years\. \[[3](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-3)\] Andrej Karpathy, who helped build OpenAI, said in October 2025 that current agents “just don’t work,” that there is “some over\-prediction going on in the industry,” and that anything resembling general intelligence is roughly a decade away\. \[[4](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-4)\] The frontier labs have begun adjusting their timelines\. Dario Amodei at Anthropic predicted powerful AI “as early as 2026” in*Machines of Loving Grace*, published in October 2024\. \[[5](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-5)\] By Davos in January 2026 he was calling for AGI\-level capability “in two years” — meaning 2028\. By his Dwarkesh interview the following month he was saying “I don’t believe we’re basically at AGI” and openly acknowledging that if his revenue forecast was off by a year, “there’s no force on earth, there’s no hedge on earth that could stop me from going bankrupt\.” \[[6](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-6)\] That shift may well represent genuine calibration rather than retreat — an honest revision as results come in\. But the capital expenditure commitments that now underpin the industry were made against the 2024 timeline, not the 2028 one\. The estimates have moved\. The money has increased\. ## The numbers The San Francisco Federal Reserve’s Economic Letter in February 2026 reviewed the macro literature and rendered the consensus in central\-bank register: “While GenAI and related applications are useful, they are not the innovation that spurs broad\-based reorganization of the economy\.” \[[7](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-7)\] Goldman Sachs’ chief economist Jan Hatzius, surveying the same terrain, concluded that AI had contributed “basically zero” to U\.S\. economic growth in 2025 and observed that “FOMO, not ROI, is driving hyperscaler capex\.” \[[8](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-8)\] This shows simply that the technology is real, its uses are real, and at the scale of the spend, the productivity it returns is not what the spend requires\. The mismatch becomes specific in the revenue math\. J\.P\. Morgan, modeling what the buildout would need to earn to clear a ten percent return on current capex, arrived at roughly six hundred fifty billion dollars per year in AI\-sector revenue\. The current run\-rate, by the most permissive count, is about twenty\-five billion\. The gap is twenty\-six\-fold; closing it would require something like $34\.72 per month, in perpetuity, from every iPhone user\. \[[9](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-9)\] The standard rebuttal is that inference costs are falling rapidly — by some estimates, ten\-fold per year — and that cheaper compute will unlock new use cases, driving revenue through volume even if per\-unit prices decline\. This is the Jevons paradox applied to tokens: make them cheap enough and demand expands to fill the gap\. But the argument has a problem\. The infrastructure being built is physical — data centers, substations, cooling systems, two\-year transformer orders — and physical infrastructure does not deflate\. A GPU generation may improve efficiency; the concrete and copper it sits in does not\. And if the Jevons paradox does hold, if falling costs drive proportionally more consumption, then total revenue stays roughly flat: more tokens at lower prices producing the same aggregate spend\. The gap doesn’t close; it just gets busier\. Meanwhile, as the next section documents, a meaningful share of the current consumption that looks like demand is waste, friction, and architectural limitation — and cost deflation applied to waste is still waste\. DeepSeek’s January 2025 release demonstrated a fundamental architectural update to models that can dramatically reduce GPU requirements while maintaining competitive capability, \[[10](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-10)\] which suggests the buildout is being sized against a hardware curve that research, not spending, may be the faster approach\. Beneath the central\-bank reading and the revenue math, the same conclusion appears, in shorter form, from every other vantage point with money in the trade: - **Tencent**, May 2026: its GPUs only pay for themselves when running personalized advertising\. \[[11](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-11)\] General\-purpose inference, the use case the buildout is sized for, the seller itself calls burning money\. - **Bank of America Fund Manager Survey**, late 2025 through early 2026: an AI bubble the single largest tail risk cited by global investors; a net twenty percent of managers calling firms overinvested, a reading not registered since August 2005\. Bloomberg’s December 2025 sweep reached the same conclusion: fifty\-four percent of fund managers already considered AI\-related stocks to be in bubble territory\. \[[12](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-12)\] The buyers of the trade agreeing with the bank math\. - **Gartner**, June 2025: more than forty percent of agentic AI projects will be cancelled by the end of 2027 — escalating costs, unclear value, inadequate risk controls\. \[[13](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-13)\] The analyst that named the category now projecting its retreat\. Two banks, a hyperscaler, the analyst that defined the category, and the fund managers whose job is to be priced into the trade — all arriving at the same conclusion\. The buildout assumes a revenue line that does not exist and that none of the production\-side studies suggest is in the process of forming\. ## Consumption The volume is real\. ServiceNow and Uber, reported by Axios in May 2026, exhausted their entire annual AI token budgets midway through the year and renegotiated their contracts\. \[[14](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-14)\] Anthropic spent the spring of 2026 throttling third\-party programmatic access to Claude on the grounds that the volume of agent traffic was outpacing what its subscription structure could absorb — the lab whose business model requires tokens to flow rationing token flow\. \[[15](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-15)\] This is the strongest evidence the bulls have: demand is growing, and growing fast\. But a meaningful share of what is being reported as demand is friction, waste, and compensation for architectural limitations\. The inflation has several sources, some operational and some structural — and the structural ones reframe the demand curve the entire buildout is sized against\. Current critiques, most prominently from Yann LeCun’s JEPA and H\-JEPA framework, argue that today’s dominant architectures lack several capabilities required for general intelligence\. The central missing component is a learned world model: a system that builds and predicts structured representations of how the world evolves, rather than predicting the next token\. From this core gap follow related limitations — the absence of robust long\-term memory for persistent state across time, limited ability for goal\-directed planning over internal representations rather than surface text, and weak uncertainty estimation when operating outside training distributions\. \[[3](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-3)\] The evidence is not only theoretical\. Even the flagship products — ChatGPT, Claude, Gemini — are not simply LLM calls\. They are complex engineering systems that use the model’s output to select and orchestrate non\-LLM tooling: search engines, code interpreters, calculators, file systems, external APIs\. The LLM produces text; the application routes that text through tools that do the actual work\. If the architecture genuinely understood the world — if it could plan, verify, and reason about consequences — the scaffolding would not be necessary\. The engineering effort required to make the product useful is itself a measure of what the model cannot do alone\. The practical consequence is that users compensate for these limitations with tokens\. They write longer prompts to constrain outputs that drift\. They make multiple attempts at the same task because the model cannot reliably plan across steps\. They build elaborate scaffolding — retrieval systems, verification loops, chain\-of\-thought prompting — to approximate capabilities the architecture does not natively possess\. Each workaround consumes tokens\. Each token is counted as demand\. This means a portion of the consumption curve is not demand for the product — it is demand for the product to be something it is not yet\. The spend rises precisely because agentic AI is not the AGI people want it to be\. Layered on top of these architectural limitations, the operational waste compounds: - **Runaway agent loops and billing failures\.**There was the recent AWS Bedrock incident, where a developer was charged a thirty\-thousand\-dollar overnight invoice because of an agent loop\. \[[16](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-16)\] Personally I have watched senior engineers accidentally spend thousands of dollars in tokens due to a process that didn’t exit, or agentic tooling error\. - **Architecture inefficiency\.**Teams are using the wrong models for the wrong tasks, building agent\-to\-agent and tool\-integration architectures using protocols like MCP and A2A that they do not yet understand well enough to implement efficiently\. - **Speed\.**The pressure to ship something labeled AI\-powered is not waiting for the tooling to mature\. I have watched this firsthand: systems designed by teams moving faster than their understanding of the protocols, routing tokens through unnecessary round\-trips and redundant context windows, burning on integration plumbing that a simpler design would have avoided entirely\. - **Developer tool churn\.**Code\-generation tools like Cursor and Copilot produce suggestions that get rejected, regenerated, rejected again, each cycle billing tokens for output that is never shipped\. The ratio of tokens consumed to tokens deployed in production is high, and nobody is publishing it\. Pre\-AI, my teams regularly developed and reused systems with similar functionalities across projects; it is not as though every engagement required building from scratch\. The tool has replaced reuse with regeneration, and the token cost of regeneration is counted as productivity, and soon I think there will be a regression to well tested manually built solutions due to the quality decrease\. - **Model\-side cost inflation\.**The models themselves are getting more expensive to use per request through two distinct mechanisms\. Anthropic’s latest flagship model ships a new tokenizer that encodes the same text into up to thirty\-five percent more tokens than its predecessor, inflating the bill without changing the output\. \[[17](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-17)\] Separately, higher default reasoning effort levels produce longer, more verbose responses — the user did not ask for more; the model simply produces it\. Multiply both effects across every API call, every agent loop, every enterprise integration, and the consumption figures rise independent of any change in what the work actually requires\. Ultimately the buyers have not learned to manage and the sellers have not learned to price, the two failures meeting in the middle and being reported, in the aggregate, as high token demand\. The buildout is being sized against consumption figures that include their own inefficiency\. If and when architectures improve — whether through the kind of foundational research Sutskever, LeCun, and others now advocate, or through breakthroughs that have not yet occurred — the token cost per unit of useful work will fall, not because inference got cheaper, but because the work required fewer compensatory tokens\. As teams mature, as the elongation documented in my previous post subsides, as architectures stabilize, and most importantly, as the companies funding AI realize when the value really exists to invest in the technology, the consumption curve may contract rather than grow\. The spending is being capitalized against an inflated number, not the number it will settle into\. I believe entire categories of agentic tooling and platforms will be abandoned within the next two years\. Companies will discover the cost of a single bad resolution at scale exceeds the labor savings\. Companies will revert to processes that keep humans in control — not because the technology is useless, but because the versions being deployed remove human judgment from workflows that require it\. ## Locked in Combined capital expenditure by the five companies building the infrastructure — Amazon, Alphabet, Meta, Microsoft, and Oracle — reached roughly $725 billion in 2026 \(approximately three\-quarters of which — roughly $544 billion — is directed at AI\-specific infrastructure: GPUs, servers, data center construction, power systems, and cooling\)\. \[[18](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-18)\] For context: the figure in 2022 was $162 billion; by 2024 it had reached $256 billion; by 2025, $448 billion\. Each year’s spend is committed against the next year’s larger spend, by companies whose capital intensity ratios — Meta at 54%, Oracle at 57%, Microsoft and Alphabet in the mid\-to\-high forties — would have been unthinkable for trillion\-dollar\-revenue firms in any prior decade\. \[[19](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-19)\] Over half a trillion dollars in annual AI capital expenditure is alarming on its own terms — no industry in history has spent at this rate\. But the deeper problem is where the money goes\. It does not reach the broader economy\. It circulates among the same handful of companies and the executives who run them\. The same three hyperscalers underwriting the bulk of the capital expenditure are the largest equity investors in the AI companies the capacity is being built for, and the financial structure connecting them has an interesting topology\. - Microsoft has committed approximately thirteen billion dollars to OpenAI; OpenAI has committed two hundred and fifty billion to Azure cloud spending\. - Microsoft records the Azure consumption as revenue, reports AI business at a thirty\-seven\-billion\-dollar annual run rate — a figure Om Malik, analyzing Microsoft’s most recent 10\-Q, traced directly to OpenAI’s spend flowing back through Azure — and uses the growth to justify its $192 billion 2026 capex commitment\. \[[20](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-20)\] - Alphabet has invested up to $40 billion in Anthropic and contracted a multi\-gigawatt TPU deal worth tens of billions in return\. \[[21](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-21)\] - Amazon has invested up to $33 billion in the same company, with over a hundred billion in commitments flowing back over ten years\. \[[22](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-22)\] > The investment generates losses, which generate infrastructure spending, which generates revenue, which absorbs the losses\. The economist Paul Krugman and the technologist Azeem Azhar have each described the arrangement as a “financial ouroboros” — what looks like booming revenue from external sales may be the same stock of money going in circles among a closed loop of companies\. \[[23](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-23)\] The circularity does not stop at the hyperscalers\. Where money circulates at this scale, intermediaries appear to position themselves inside the loop\. A layer of debt\-financed companies has emerged whose entire business is to buy GPUs with borrowed money and rent the compute back to the companies that funded them\. CoreWeave, the most prominent, has raised approximately $28 billion in combined equity and debt in the past twelve months\. It derives roughly two\-thirds of its revenue from Microsoft\. Its largest remaining contracts are with OpenAI and Meta — the same companies whose hyperscaler backers are, at one remove, the source of the capital CoreWeave borrowed to buy the hardware it rents to them\. In March 2026, CoreWeave closed an $8\.5 billion delayed\-draw term loan rated A3 by Moody’s, the first investment\-grade\-rated financing in history secured by GPU infrastructure and an associated customer contract\. \[[24](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-24)\] The rating, on inspection, is a judgment about Microsoft’s and Meta’s creditworthiness applied to debt whose underlying collateral is rapidly depreciating semiconductor hardware — replaced roughly every four to six years solely due to efficiency upgrades for newer models\. Because of its rating, pension funds, insurers, and money\-market vehicles can now hold this paper by mandate\. The risk has been intermediated; it has not been reduced\. The competitive pressure is not only internal\. In January 2025, when the Chinese lab DeepSeek released models that matched or approached frontier capability at a fraction of the compute cost, demonstrating that the relationship between spending and capability is far less linear than the buildout assumes\. \[[10](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-10)\] The effect was paradoxical: rather than prompting a reassessment of whether the scale of the American buildout is warranted, it intensified it\. The argument shifted from “we must spend to lead” to “we must spend or lose the lead to China,” converting a technology investment into a geopolitical obligation\. The Gulf states — Saudi Arabia, the UAE, and Qatar — are building parallel infrastructure stacks domestically and hold equity positions across all three major American frontier labs, so that unilateral exit by any US company is now a geopolitical concession as well as a financial one\. The Trump administration has treated the buildout as a national\-strategic project, effectively deferring to the tech sector’s own assessment of what the technology requires and what regulation it can tolerate\. Stargate launched at the White House with Altman and Ellison on stage, \[[25](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-25)\] with three executive orders removing AI regulation and streamlining data center permitting, \[[26](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-26)\] and a December 2025 order preempting state AI laws\. \[[27](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-27)\] The regulatory friction that might have slowed the spend has been removed\. The pattern has a precedent\. In the five years following the Telecommunications Act of 1996, telecom companies invested more than $500 billion — roughly a trillion in today’s dollars, including acquisitions — mostly financed with debt, into laying fiber\-optic cable, adding switches, and building out network capacity against demand projections that proved wildly unrealistic\. WorldCom’s CEO told investors that internet traffic was doubling every hundred days; in reality it was doubling roughly once a year\. \[[28](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-28)\] The accounting structures that sustained the buildout bear a striking resemblance to the ones operating today\. Qwest sold indefeasible\-right\-of\-use fiber capacity to Global Crossing while Global Crossing sold capacity back — bilateral swaps that inflated reported revenue on both sides\. In a single quarter in 2001, reciprocal transactions accounted for thirty\-two percent of Global Crossing’s reported cash revenue; without them, the company’s adjusted EBITDA would have swung from positive four\-hundred\-seventy\-two million to negative forty\-three million\. \[[29](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-29)\] The SEC later found Qwest had fraudulently recognized more than $3\.8 billion in revenue through structures of this kind\. Employees at Qwest had a name for the practice\. They called it heroin, because “each quarter’s gap between real revenue and the projection required a larger dose to fill\.” \[[30](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-30)\] The structural parallels are not difficult to see\. The circular investment loops, the debt\-financed intermediaries, the demand projections set by the people selling the supply, the accounting structures that allow each participant to report the other’s spending as revenue, the impossibility of unilateral exit once the commitments are made\. What is different is the scale\. The telecom buildout’s peak annual capex, adjusted for inflation, was roughly $213 billion\. The AI buildout will direct approximately two and a half times that to AI\-specific infrastructure in 2026 alone\. The reason none of them can stop is that the investment, the revenue, and the justification for the next investment are the same transaction\. If Microsoft reduces its OpenAI commitment, it loses one of Azure’s largest customers, the AI revenue line that justifies $192 billion in capex, and the earnings growth that holds its stock price — all at once\. The same logic binds Alphabet and Amazon to Anthropic: the equity position and the cloud contract are not separate bets, they are the same bet, and unwinding one unwinds both\. Beneath them, the debt compounds the lock\. CoreWeave’s A3 rating, Meta’s SPV financing, Oracle’s bond offering — each is underwritten by the assumption that hyperscaler commitments continue\. If any one of them pulls back, the ratings reprice, potentially to junk in a single cycle, and the institutional investors holding the paper absorb the difference\. No single company can exit without triggering consequences for every other company in the chain\. The result is a structure in which everyone continues to spend because everyone else is spending, and the cost of being the first to stop exceeds the cost of continuing — even when continuing no longer makes sense\. Of the five, in my opinion, Meta is the one to watch for the first pullback\. Its capex\-to\-revenue ratio is the highest among the trillion\-dollar\-revenue hyperscalers; Barclays models a roughly ninety\-percent collapse in its free cash flow\. In the spring of 2026 it financed the same Louisiana data center twice in two weeks — first through a $30 billion off\-balance\-sheet SPV whose A\+ rating S&P called dependent on Meta’s residual\-value guarantee as the “linchpin,” then, fourteen days later, through $30 billion in corporate bonds resting on the same balance sheet that backs the guarantee\. \[[31](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-31)\] Meta is the only one of the five hyperscalers without a cloud computing business to monetize the infrastructure externally\. JPMorgan downgraded the stock in April 2026, citing a “challenging path” to capitalizing on AI spending\. \[[32](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-32)\] When asked on the Q1 earnings call about evidence of return on investment from $145 billion in annual AI capex, Zuckerberg called it “a very technical question\.” \[[33](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-33)\] The trigger, when it comes, will likely be a quarter in which ad revenue growth decelerates while capex guidance does not\. The fallout would not be contained to Meta’s stock price\. Its SPV financing, its bond issuance, and the pension and insurance portfolios holding both would reprice simultaneously\. ## The plan The Federal Reserve’s Spring 2026 Financial Stability Report elevated artificial intelligence from fifth to third among perceived threats to American financial stability, behind only geopolitical risk and ahead of inflation\. The Chicago Fed’s 2026 insights note described, in plain language, the tail risk to banks from commercial loans underwriting AI capital expenditure\. Vice Chair Bowman convened a Financial Stability Oversight Council roundtable on AI and systemic risk\. \[[34](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-34)\] The institutions whose job it is to identify systemic exposure have begun identifying it\. In November 2025, at the Wall Street Journal’s Tech Live event, OpenAI’s chief financial officer Sarah Friar was asked how the company intended to finance its chip and data center commitments\. Her answer was specific: OpenAI was looking for an ecosystem of banks, private equity, and a federal “backstop” or “guarantee” that could lower financing costs and increase the amount of debt the company could take on\. The interviewer pressed: a federal backstop for chip investment? Friar confirmed\. This was the moment someone accidentally spoke the truth\. The retraction arrived within twenty\-four hours, from three directions\. Friar herself, on LinkedIn: “I used the word ‘backstop’ and it muddied the point\.” Sam Altman, on X: “We do not have or want government guarantees for OpenAI datacenters\.” David Sacks, Trump’s AI and crypto policy czar: “There will be no federal bailout for AI\.” \[[35](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-35)\] The CFO of the company at the financial center of the buildout, asked in public how it would pay for commitments at the scale it has signed, gave the answer the structure of the deal requires\. The retraction is the more informative half of the exchange\. You do not need three people to deny something in twenty\-four hours unless the thing they are denying was the plan from the start\. She has since been excluded from key financial meetings, her absence described as “notable and awkward\.” \[[36](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-36)\] The person who said the quiet part out loud was not corrected\. She was sidelined\. The precedent she was reaching for is not hypothetical\. - In 2008, the federal government authorized $700 billion to stabilize the financial system\. The banks were made whole; the Congressional Oversight Panel later found no evidence the funds were used to prevent foreclosures\. Millions of homeowners lost their houses while the institutions that sold them the mortgages were recapitalized with public money as “too big to fail”\. - In 2020, the Federal Reserve bought corporate bonds directly off the open market for the first time, extending the assumption of public rescue from banks to the corporate credit market\. - In 2023, the FDIC waived its own statutory deposit limit overnight when Silicon Valley Bank collapsed, protecting deposits that were legally uninsured\. Each intervention was described as exceptional\. Each established that institutions whose failure would propagate will be caught — and that the people on the other side of the transaction will not\. The financial system now prices AI infrastructure debt on this assumption\. The plan, to the extent that anyone in the room would use the word, is a bailout\. No one has voted on it\. No legislation has been passed\. The assumption holds because the precedent holds\. ## The floor So what happened when telco collapsed between 2000\-2002? It erased more than two trillion dollars in market capitalization\. Twenty\-three telecom companies filed for bankruptcy, led by WorldCom — then the largest bankruptcy in American history\. Bond investors recovered just over twenty cents on the dollar\. But the losses did not stay on Wall Street\. The FCC chairman testified to the Senate Commerce Committee that the industry owed one trillion dollars, much of which would never be repaid\. \[[37](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-37)\] Public pension funds bore billions in direct losses: California’s public employees’ retirement system lost $565 million on WorldCom alone; New York’s state retirement system, $300 million; Texas, $277 million\. Tens of thousands of workers lost their jobs and their retirement savings simultaneously\. WorldCom employees whose 401\(k\) plans were loaded with company stock watched both disappear\. State and local governments, forced to cover pension shortfalls, cut public services to make up the difference\. Fiber\-optic cable, once laid, has a shelf life measured in decades\. When demand eventually caught up — and it did — the dark fiber lit up and became the backbone of the modern internet\. The investors who overpaid were not made whole, but the physical asset retained value\. Semiconductors will not\. GPUs depreciate on a cycle of roughly six years, driven not by wear but by architectural obsolescence; each new generation renders the prior one uneconomical to operate\. The data centers being built today will house hardware that is outdated before the demand the buildout assumes has had time to materialize\. When the correction comes — and I believe it will be the largest in the history of the technology industry — the assets at the center of it will not be waiting patiently underground for the world to catch up\. They will be waste\. The AI buildout will reach the same people the telecom collapse reached — through more channels: - **Retirement defaults\.**The ten largest S&P 500 companies — the majority of which are the same hyperscalers and AI\-infrastructure companies driving the buildout — now account for roughly 40% of the index’s total weight while contributing 32% of earnings\. Under the Pension Protection Act, auto\-enrollment routes employees into target\-date funds that track these indexes by default\. About forty cents of every default 401\(k\) dollar flows into these AI\-exposed companies\. Households are price\-insensitive buyers of bubble exposure by design\. \[[38](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-38)\] - **State tax subsidies\.**Virginia’s data center sales tax exemption cost $1\.6 billion in fiscal 2025 — the state projected $1\.54 million when the program was created\. Wisconsin, $2 billion\. \[[39](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-39)\] Texas will lose $3\.2 billion over the next biennium; it projected $180 million\. Georgia expects $2\.5 billion this year\. Illinois reached $1 billion before the governor suspended the program\. The subsidies produce roughly one permanent job per million dollars of public cost\. \[[40](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-40)\] - **Mark\-to\-market accounting\.**In Q1 2026, more than half of Amazon’s quarterly profit came from marking up the value of its Anthropic stake — not from selling products or cloud services but from updating the estimated value of an investment\. Alphabet reported $28\.7 billion of its $62\.6 billion quarterly profit from the same source\. Each revaluation inflates earnings, lifts the company’s index weight, and reweights the 401\(k\) that buys more of the inflated stock\. No cash changes hands\. \[[41](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-41)\] Neither Anthropic nor OpenAI has gone public — retail investors absorb the risk through their retirement accounts but could not have captured the upside\. Columbia Business School’s Robert Willens, on the structure: “They’re able to control or influence the value of one of their own assets… and one that they’re able to mark to market by engaging in business transactions with that entity\.” \[[41](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-41)\] - **Investment\-grade debt\.**The debt underpinning the buildout — CoreWeave’s A3\-rated term loan, Meta’s off\-balance\-sheet SPVs, Oracle’s $25 billion bond — is rated on the strength of the hyperscalers’ creditworthiness, not the underlying assets\. That rating is what allows pension funds and insurers to hold the paper by mandate\. Anthropic is now in discussions to raise at least $30 billion at a valuation exceeding $900 billion — and neither Anthropic nor OpenAI has gone public\. \[[42](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-42)\] The companies whose explosive growth is driving the index gains, the mark\-to\-market profits, and the entire capex justification have never been available for ordinary investors to buy directly\. Retail investors absorb the risk of the bubble through their retirement accounts but could not have participated in the upside even if they had wanted to\. The growth was captured privately\. The exposure is distributed publicly\. It has always been a move available to those who build complex systems to use the complexity itself as leverage — to secure investment, financing, and time from people who do not understand what is being built\. The railroad barons did it with federal land grants\. The defense contractors did it with cost\-plus procurement\. The telecom executives did it with demand projections they knew were false\. The pattern is not new\. What is new are the lack of regulation, and the modern financial surfaces through which the cost is distributed\. The companies and people distributing it are personally enriched at every step — equity in the labs they fund, stock in the companies they run, fees on the debt they structure, carry on the funds they manage\. The buildout does not need to succeed for the people building it to profit\. It needs only to continue\. There is an implicit guarantee that none of the people who made the bet will be the ones who pay for it\. The answer is ordinary people\. The same people the buildout was sold as replacing will be the ones who pay for it\. --- ## Epilogue The costs that do not appear on any balance sheet are borne by the communities\. In neighborhoods near data centers, electricity prices have risen as much as 267% over the past five years\. \[[43](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-43)\] Seventy\-eight percent of Americans say they are concerned that new data centers will raise their energy bills\. \[[44](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-44)\] A Carnegie Mellon economist, studying approximately 2,800 operational data centers, estimated $25 billion per year in hidden health and environmental damage\. \[[45](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-45)\] Residents in Virginia describe the sound from nearby facilities as an “internal organ vibration” and say their communities are “becoming an industrial wasteland\.” \[[46](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-46)\] Two\-thirds of data centers under development since 2022 are in water\-stressed areas, each large facility consuming as much water as a town of ten to fifty thousand people\. \[[47](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-47)\] The communities absorbing these costs were not consulted, did not benefit, and in most cases were not informed until the construction was underway\. Between April and June of 2025, twenty data center proposals worth $98 billion were blocked or delayed by local opposition — two\-thirds of all projects tracked in that period\. \[[48](https://nooneshappy.com/article/the-scale-the-plan-and-the-people/#ref-48)\] The industry’s response has not been to slow down\. It has been to secure federal preemption of local objections\. The time to raise these questions is before the correction, not after\. That is the lesson of every bubble that came before this one, and it demonstrates that this is the plan, not a case of memory failure\. --- ## References 1\.[Ilya Sutskever — We’re moving from the age of scaling to the age of research](https://www.dwarkesh.com/p/ilya-sutskever-2)\(Dwarkesh Patel, November 25, 2025\)\. Sutskever dates the age of scaling as roughly 2020–2025, argues another 100x of pre\-training would not produce a qualitative capability change, and says the field is returning to a research era\. 2\.[AGI Still Years Away Despite Tech Leaders’ Bold Promises](https://medium.com/@cognidownunder/agi-still-years-away-despite-tech-leaders-bold-promises-for-2026-146c9780af65)\(Cogni Down Under, October 21, 2025\), and Gary Marcus,[“The last few months have been devastating”](https://garymarcus.substack.com/p/the-last-few-months-have-been-devastating)\(October 18, 2025\)\. Hassabis put the AGI window at five to ten years in October 2025 and publicly corrected OpenAI’s Sébastien Bubeck on overstated math\-benchmark claims that month\. 3\.[Yann LeCun Says Nobody in Their Right Mind Will Use LLMs And GenAI Within 5 Years](https://wonderfulengineering.com/yann-lecun-says-nobody-in-their-right-mind-will-use-llms-and-genai-within-5-years/)\(Wonderful Engineering, 2025\) and[Yann LeCun Left Meta, Raised $1B & Says AI Is Wrong](https://kersai.com/yann-lecun-ami-labs-world-models-llms-wrong-complete-guide-2026/)\(Kersai, 2026\)\. LeCun left Meta in November 2025 to serve as Executive Chairman of Advanced Machine Intelligence Labs \(Paris\); the company closed a $1\.03B seed at a $3\.5B pre\-money in March 2026\. LeCun’s JEPA/H\-JEPA framework identifies world modeling, working memory, hierarchical planning, and calibrated uncertainty as cognitive capabilities current architectures lack\. 4\.[Andrej Karpathy on OpenAI, the AI bubble, and the Dwarkesh Patel interview](https://fortune.com/2025/10/21/andrej-karpathy-openai-ai-bubble-pop-dwarkesh-patel-interview/)\(Fortune, October 21, 2025\); see also[Simon Willison’s notes on the same interview](https://simonwillison.net/2025/Oct/18/agi-is-still-a-decade-away/)\. Karpathy says current agents “just don’t work,” that the industry is “making too big of a jump,” and that anything resembling AGI is roughly a decade away\. 5\.[Machines of Loving Grace](https://www.darioamodei.com/essay/machines-of-loving-grace)\(Dario Amodei, October 2024\)\. The essay places “powerful AI” “as early as 2026” before exploring 5–10 year transformation scenarios\. 6\.[Dario Amodei in conversation with Dwarkesh Patel, February 2026](https://www.dwarkesh.com/p/dario-amodei-2)\. Amodei: “I don’t believe we’re basically at AGI”; on revenue sensitivity, “there’s no force on earth, there’s no hedge on earth that could stop me from going bankrupt” if his forecast is off by a year\. 7\.[AI Moment: Possibilities, Productivity, Policy](https://www.frbsf.org/research-and-insights/publications/economic-letter/2026/02/ai-moment-possibilities-productivity-policy/)\(Federal Reserve Bank of San Francisco*Economic Letter*, February 2026\)\. “While GenAI and related applications are useful, they are not the innovation that spurs broad\-based reorganization of the economy\.” 8\.[AI boosted US economy by “basically zero” in 2025, says Goldman Sachs chief economist](https://www.tomshardware.com/tech-industry/artificial-intelligence/ai-boosted-us-economy-by-basically-zero-in-2025-says-goldman-sachs-chief-economist-we-think-theres-been-a-lot-of-misreporting-of-the-impact-that-ai-investment-had-on-gdp-growth)\(Tom’s Hardware, 2026\), and[Goldman finds no meaningful relationship between AI and productivity](https://fortune.com/2026/03/03/goldman-earnings-ai-anxiety-no-meaningful-impact-productivity-economy-30-percent-in-2-areas/)\(Fortune, March 3, 2026\)\. Hatzius attributes the buildout to FOMO rather than measured return on investment\. 9\.[J\.P\. Morgan calls out AI spend, says $650B in annual revenue required to deliver a mere 10% return on AI buildout](https://www.tomshardware.com/tech-industry/artificial-intelligence/usd650-billion-in-annual-revenue-required-to-deliver-10-percent-return-on-ai-buildout-investment-j-p-morgan-claims-equivalent-to-usd35-payment-from-every-iphone-user-or-usd180-from-every-netflix-subscriber-in-perpetuity)\(Tom’s Hardware, 2026\)\. $650B/yr is equivalent to ~$35/month per iPhone user or ~$180/month per Netflix subscriber, in perpetuity\. 10\.[DeepSeek’s AI Breakthrough: Matching OpenAI at a Fraction of the Cost](https://www.nytimes.com/2025/01/27/technology/deepseek-china-ai-openai.html)\(New York Times, January 27, 2025\);[DeepSeek shakes AI industry with low\-cost models](https://www.reuters.com/technology/artificial-intelligence/deepseek-sets-new-benchmark-ai-efficiency-2025-01-20/)\(Reuters, January 2025\)\. DeepSeek’s R1 model matched frontier capability at a reported training cost orders of magnitude below US competitors, triggering a broad reassessment of the relationship between spending and capability\. 11\.[Tencent admits GPUs only pay for themselves when powering personalized ads](https://www.theregister.com/off-prem/2026/05/14/tencent-admits-gpus-only-pay-for-themselves-when-powering-personalized-ads/5240150)\(The Register, May 14, 2026\)\. Hyperscaler confirmation that general\-purpose AI inference is unprofitable; only ad\-targeted use cases clear the cost of compute\. 12\.[For the first time in 20 years, AI bubble fears have fund managers saying companies are overdoing it](https://fortune.com/2025/11/18/fund-managers-fear-ai-bubble-say-companies-overinvested/)\(Fortune on the BofA Fund Manager Survey, November 18, 2025\) and[Wall Street Sees an AI Bubble Forming and Is Gaming What Pops It](https://www.bloomberg.com/news/articles/2025-12-14/wall-street-sees-an-ai-bubble-forming-and-is-gaming-what-pops-it)\(Bloomberg, December 14, 2025\)\. AI bubble cited as \#1 tail risk by 45% of fund managers at peak; net 20% of managers say firms are overinvesting — first such reading since August 2005\. Bloomberg’s parallel survey: 54% of fund managers calling AI\-related stocks “bubble territory\.” 13\.[Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027](https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027)\(Gartner press release, June 25, 2025\)\. Cancellations attributed to escalating costs, unclear business value, and inadequate risk controls\. 14\.[Anthropic tightens Claude limits as OpenAI courts agent users](https://www.axios.com/2026/05/14/anthropic-claude-price-openai-tokens)\(Axios, May 14, 2026\)\. ServiceNow and Uber burned through their entire annual AI token budgets before mid\-year; Axios: “All\-you\-can\-eat AI subscriptions may not survive the agent era\.” 15\.[Anthropic cuts third\-party usage](https://www.axios.com/2026/04/06/anthropic-openclaw-subscription-openai)\(Axios, April 6, 2026\) and[AI demand is inflated, and only Anthropic is being realistic](https://www.cnbc.com/2026/04/17/ai-tokens-anthropic-openai-nvidia.html)\(CNBC, April 17, 2026\)\. Anthropic placed third\-party programmatic usage behind a separate credit meter in spring 2026\. 16\.[Bedrock and a hard place: Claude adventure leaves AWS user staring down $30K invoice](https://www.theregister.com/saas/2026/05/14/bedrock-and-a-hard-place-claude-adventure-leaves-aws-user-staring-down-30k-invoice/5238153)\(The Register, May 14, 2026\)\. Runaway agent loop produced an overnight $30K Bedrock invoice; Cost Anomaly Detection did not fire\. 17\.[Claude Opus 4\.7 Pricing](https://platform.claude.com/docs/en/about-claude/pricing)\(Anthropic, 2026\)\. Opus 4\.7 ships a new tokenizer that produces up to 35% more tokens from the same text at an unchanged headline rate; the model also defaults to a higher reasoning effort level, producing longer responses per request\. 18\.[Big Tech’s AI Spending to Reach $725 Billion in 2026](https://www.statista.com/chart/35046/capital-expenditure-of-meta-alphabet-amazon-and-microsoft/)\(Statista, 2026\),[Hyperscalers Hit $700 Billion in 2026](https://finance.yahoo.com/sectors/technology/articles/hyperscalers-hit-700-billion-2026-111243744.html)\(Yahoo Finance / 24/7 Wall St, 2026\), and[The Magnificent Capex: AI Infrastructure Spending and Who Actually Benefits](https://www.fergusonwellman.com/blog/2026/5/8/the-magnificent-capex-ai-infrastructure-spending-and-who-actually-benefits)\(Ferguson Wellman, May 2026\)\. Aggregate Big Five capex: ~$162B \(2022\) → ~$256B \(2024\) → ~$448B \(2025\) → ~$725B \(2026 confirmed at Q1 earnings\); approximately 75% directed at AI\-specific infrastructure\. 19\.[The Magnificent Capex](https://www.fergusonwellman.com/blog/2026/5/8/the-magnificent-capex-ai-infrastructure-spending-and-who-actually-benefits)\(Ferguson Wellman, May 2026\)\. 2026 capex as % of revenue: Meta ~54%, Microsoft ~47%, Alphabet ~46%; Amazon’s Q1 free cash flow collapsed to $1\.2B after a $59\.3B YoY surge in infrastructure spending\. 20\.[Microsoft calls for $190 billion in 2026 capital spending on soaring memory prices](https://www.cnbc.com/2026/04/29/microsoft-msft-q3-earnings-report-2026.html)\(CNBC, April 29, 2026\) and[What Microsoft’s 10\-Q Says About OpenAI](https://om.co/2026/05/01/what-microsofts-10-q-says-about-openai/)\(Om Malik, May 1, 2026\)\. Microsoft’s $37B AI run rate, traced to OpenAI’s Azure spend, justifies the company’s ~$190B 2026 capex plan; OpenAI’s $250B Azure commitment is the underlying flow\. 21\.[Google’s $40B Anthropic Bet](https://tech-insider.org/google-40-billion-anthropic-investment-tpu-compute-2026/)\(Tech Insider, April 25, 2026\)\. 22\.[Amazon commits up to $25B more to Anthropic; Anthropic to spend $100B\+ over 10 years on AWS](https://www.aboutamazon.com/news/aws/aws-anthropic-100-billion-trainium-2026)\(Amazon, April 20, 2026\)\. 23\.[Should we worry about AI’s circular deals?](https://www.noahpinion.blog/p/should-we-worry-about-ais-circular)\(Noah Smith, 2025\), discussing the “financial ouroboros” framing from Paul Krugman and Azeem Azhar; see also Paul Krugman,[Talking AI With Martin Wolf](https://paulkrugman.substack.com/p/talking-ai-with-martin-wolf)\. The Microsoft→OpenAI→Azure→Microsoft loop produces revenue at Microsoft from money it put into OpenAI in the first place\. 24\.[CoreWeave Takes As Much Financial Engineering As It Does Datacenter Design](https://www.nextplatform.com/cloud/2026/04/09/coreweave-takes-as-much-financial-engineering-as-it-does-datacenter-design/5215794)\(The Next Platform, April 9, 2026\);[CoreWeave closes $8\.5B DDTL 4\.0, rated A3 by Moody’s](https://www.sec.gov/Archives/edgar/data/0001769628/000176962825000033/ddtl30pressrelease-ex991x6.htm)\(CoreWeave 8\-K, March 31, 2026\)\. ~$28B raised in 12 months; ~67% of FY2024 revenue from Microsoft; first investment\-grade\-rated financing in history secured by GPU infrastructure\. 25\.[What to Know About “Stargate,” OpenAI’s New Venture Announced by Trump](https://time.com/7209167/stargate-openai-donald-trump/)\(TIME, January 22, 2025\)\. Trump, Altman, Son, and Ellison appeared at the White House to announce up to $500B of Stargate AI infrastructure spending\. 26\.[Accelerating Federal Permitting of Data Center Infrastructure](https://www.whitehouse.gov/presidential-actions/2025/07/accelerating-federal-permitting-of-data-center-infrastructure/)\(Executive Order, July 23, 2025\)\. Part of a three\-EO package directing federal agencies to streamline permitting, provide financial support, and unlock federal land for data center development\. 27\.[Eliminating State Law Obstruction of National Artificial Intelligence Policy](https://www.whitehouse.gov/presidential-actions/2025/12/eliminating-state-law-obstruction-of-national-artificial-intelligence-policy/)\(Executive Order, December 11, 2025\)\. Establishes a DOJ AI Litigation Task Force and conditions federal funding on state\-law compliance with a national AI policy framework\. 28\.[The Spreadsheet that Fueled the Telecom Boom — and Bust](https://apennings.com/technologies-of-meaning/the-spreadsheet-that-fueled-the-telecom-boom-and-bust/)\(Anthony J\. Pennings, PhD\) and[Did WorldCom Puff Up the Internet Too?](https://www.lightreading.com/business-management/did-worldcom-puff-up-the-internet-too-)\(Light Reading\)\. The “doubling every 100 days” claim originated with UUNet’s Tom Stluka’s best\-case model, was disavowed internally, and was nonetheless published in WorldCom’s 1998 annual report\. Real growth rates were 70–150% per year\. 29\.[Capacity Swaps by Global Crossing and Qwest: Sham Transactions Designed to Boost Revenues?](https://www.govinfo.gov/content/pkg/CHRG-107hhrg81961/html/CHRG-107hhrg81961.htm)\(U\.S\. House Energy and Commerce Subcommittee hearing, 2002\)\. In a single Q1 2001 quarter, reciprocal IRU swaps contributed ~32% of Global Crossing’s reported cash revenue; without them, adjusted EBITDA swung from positive $472M to roughly $\-43M\. 30\.[SEC v\. Qwest Communications International Inc\. — Litigation Release 18936](https://www.sec.gov/enforcement-litigation/litigation-releases/lr-18936)\(Securities and Exchange Commission\)\. Qwest fraudulently recognized over $3\.8B in revenue between 1999 and 2002 through non\-recurring IRU and equipment transactions; internally, the practice was likened to an “addiction” and the transactions called the company’s “heroin\.” 31\.[A Behind\-the\-Scenes Look at How Meta Raised $30 Billion for Its AI Data Center](https://www.tipranks.com/news/a-behind-the-scenes-look-at-how-meta-raised-30-billion-for-its-ai-data-center)\(TipRanks, 2026\) and[Blue Owl and Meta close record $30bn financing for AI data centre expansion in Louisiana](https://pe-insights.com/blue-owl-and-meta-close-record-30bn-financing-for-ai-data-centre-expansion-in-louisiana/)\(Private Equity Insights, 2026\)\. The Hyperion campus was financed first through the Beignet Investor SPV \(PIMCO $18B, BlackRock $3B, Blue Owl\-managed funds taking the rest\), rated A\+ on the strength of Meta’s residual\-value guarantee — described by S&P as the “linchpin” of the rating — then again via $30B in Meta corporate bonds two weeks later\. 32\.[Meta Platforms gets a downgrade from JPMorgan on massive AI spending forecast](https://www.cnbc.com/2026/04/30/meta-platforms-gets-a-downgrade-from-jpmorgan-on-massive-ai-spending-forecast.html)\(CNBC, April 30, 2026\)\. JPMorgan downgraded Meta to neutral from overweight, slashing its price target to $725 from $825, citing a “challenging path” to capitalizing on AI\. 33\.[Meta is spending up to $145 billion this year on AI\. When asked about signs of ROI, Zuckerberg said ‘that’s a very technical question’](https://fortune.com/2026/04/29/meta-ai-spending-145-billion-zuckerberg-roi-technical-question/)\(Fortune, April 29, 2026\)\. 34\.[Federal Reserve Spring 2026 survey highlights geopolitical risks, AI concerns as top threats to financial stability](https://cryptobriefing.com/fed-spring-2026-geopolitical-ai-risks/)\(2026\), and[Speech by Vice Chair for Supervision Bowman on artificial intelligence in the financial system](https://www.federalreserve.gov/newsevents/speech/bowman20260501a.htm)\(Federal Reserve Board, May 1, 2026, delivered at the FSOC Artificial Intelligence Series Roundtable\)\. AI rose from fifth to third in perceived threats; private credit moved from ninth to fourth\. 35\.[OpenAI CFO Sarah Friar says company isn’t seeking government backstop, clarifying prior comment](https://www.cnbc.com/2025/11/06/openai-cfo-sarah-friar-says-company-is-not-seeking-government-backstop.html)\(CNBC, November 6, 2025\) and[OpenAI CFO walks back remarks about federal loan guarantees](https://www.theregister.com/2025/11/06/openai_cfo_walks_back_remarks/)\(The Register, November 6, 2025\)\. Friar at WSJ Tech Live raised a federal “backstop” or “guarantee” for AI chip financing; the retraction came within 24 hours from Friar, Altman, and David Sacks\. 36\.[OpenAI CFO Excluded From Investor Meetings Amid IPO and Spending Clash With Altman](https://thedeepdive.ca/openai-cfo-excluded-from-investor-meetings-amid-ipo-and-spending-clash-with-altman/)\(The Deep Dive, 2026\), summarizing The Information’s reporting that Friar has been left out of key financial meetings — in one case, an absence described as “notable and awkward\.” 37\.[The Great Telecom Implosion](https://www.princeton.edu/~starr/articles/articles02/Starr-TelecomImplosion-9-02.htm)\(Paul Starr, Princeton, 2002\) and[Telecoms crash](https://en.wikipedia.org/wiki/Telecoms_crash)\(overview\)\. Roughly $2T in telecom market capitalization erased between 2000 and 2002; 23 bankruptcies; bond investors recovered ~20 cents on the dollar; FCC Chairman Michael Powell testified the industry owed about a trillion dollars, much of which would never be repaid\. 38\.[Your 401\(k\) Is Propping Up the AI Bubble](https://www.promarket.org/2026/05/05/your-401k-is-propping-up-the-ai-bubble/)\(ProMarket / Stigler Center, May 5, 2026\)\. Top\-ten S&P 500 weight ~41% \(from 19% in 2015\) against ~32% of earnings; auto\-enrollment \+ target\-date defaults route ~40 cents of every default 401\(k\) dollar to those names; 61% of Vanguard plans use auto\-enrollment, 84% of participants hold target\-date funds\. 39\.[Cloudy With a Loss of Spending Control: How Data Centers Are Endangering State Budgets](https://goodjobsfirst.org/cloudy-with-a-loss-of-spending-control-how-data-centers-are-endangering-state-budgets/)\(Good Jobs First — LeRoy & Tarczynska, April 2025\) and[Many states don’t report losses from data center tax breaks, study says](https://stateline.org/2026/04/15/many-states-dont-report-losses-from-data-center-tax-breaks-study-says/)\(Stateline, April 15, 2026\)\. State\-by\-state foregone\-revenue figures, the uncapped/indefinite design of the exemptions, the “Dark 12” states that disclose nothing, and the Wisconsin $2B figure \(per WPR, April 2026\)\. 40\.[Many states don’t report losses from data center tax breaks, study says](https://stateline.org/2026/04/15/many-states-dont-report-losses-from-data-center-tax-breaks-study-says/)\(Stateline, April 15, 2026\)\. Taxpayers in subsidy\-offering states pay roughly $1M per permanent job created at a data center site\. 41\.[Half of Google’s and Amazon’s “blowout AI profits” came from a stake in Anthropic — not from their actual business](https://fortune.com/2026/04/30/google-amazon-ai-profits-anthropic-stake-bubble-earnings-2026/)\(Fortune, April 30, 2026\)\. Amazon Q1 2026: $16\.8B pre\-tax gain on Anthropic markup; Alphabet: $28\.7B of $62\.6B quarterly profit from the same\. Columbia Business School’s Robert Willens: “They’re able to control or influence the value of one of their own assets… and one that they’re able to mark to market by engaging in business transactions with that entity\.” 42\.[Anthropic in talks for $30B at $900B valuation](https://www.bloomberg.com/news/articles/2026-05-12/anthropic-30-billion-900-billion-valuation)\(Bloomberg, May 12, 2026\)\. The round is reportedly in final stages; would mark Anthropic’s last private round before a potential late\-2026 IPO\. 43\.[Data Center Power Demands Are Contributing to Higher Energy Bills](https://www.eesi.org/articles/view/data-center-power-demands-are-contributing-to-higher-energy-bills)\(Environmental and Energy Study Institute, 2025\)\. In areas with high concentrations of data centers, electricity prices have risen as much as 267% over the past five years\. 44\.[New Survey From Consumer Reports finds Majority of Households Strained by Energy Bills, Concerned over Data Centers’ Impact on Bills](https://advocacy.consumerreports.org/press_release/new-survey-from-consumer-reports-finds-majority-of-households-strained-by-energy-bills-concerned-over-data-centerss-impact-on-bills/)\(Consumer Reports, November 2025\)\. 78% of US adults are somewhat or very concerned that new data centers will raise their energy bills\. 45\.[Data centers cost the U\.S\. economy $25 billion a year in hidden health and environmental damage](https://fortune.com/2026/04/21/data-centers-environmental-health-costs-25-billion/)\(Fortune, April 21, 2026\), summarizing Nicholas Muller \(Carnegie Mellon\),[Measuring the Impact of Data Centers in the United States Economy](https://www.nber.org/papers/w35100)\(NBER Working Paper 35100\)\. Analysis of ~2,800 operational data centers; ~$3\.7B of the $25B directly tied to AI workloads; Virginia and Texas account for ~30% of the total\. 46\.[This “Health Earthquake” Is Hitting Virginia Residents as Data Centers Surge](https://www.usnews.com/news/national-news/articles/2026-04-28/living-in-hell-data-center-neighbors-grapple-with-noise-air-pollution)\(U\.S\. News & World Report, April 28, 2026\)\. A Prince William County resident describes the sound from a neighboring Google data\-center complex as an “internal organ vibration” that “literally rocks your core\.” 47\.[The AI Boom Is Draining Water From the Areas That Need It Most](https://www.bloomberg.com/graphics/2025-ai-impacts-data-centers-water-data/)\(Bloomberg, 2025\)\. Roughly two\-thirds of US data centers built or in development since 2022 sit in places already gripped by high water stress; a mid\-sized facility consumes as much water as a small town and the largest consume the daily equivalent of a city of fifty thousand\. 48\.[Data Center Watch Report Q2 2025 UPDATE](https://www.datacenterwatch.org/q22025)\(Data Center Watch, 2025\)\. In Q2 2025, twenty projects representing $98B in proposed investment were blocked or delayed amid local opposition — about two\-thirds of all protested projects tracked in that period\.