WHAT THE AI BUBBLE IS COVERING
AOC is right about the bubble. The frame she has chosen also limits what she can see and say.
Something has been visible for a while now, beneath the noise of markets and quarterly earnings and congressional hearings, to anyone paying attention at the level where things are actually moving. Not a secret. Not a conspiracy. A pattern — observable, documented, and for anyone willing to watch the behavior rather than the statements, not particularly hidden.
In November 2025, Representative Alexandria Ocasio-Cortez put a name to it. Speaking before a House subcommittee, she described what the numbers already showed: “We’re talking about a massive economic bubble. Depending on the exposure of that bubble, we could see 2008-style threats to economic stability.” She added that should the bubble pop, “we should not be entertaining a bailout of these corporations, as healthcare is being denied to everyday Americans, as SNAP and food assistance is being denied to everyday Americans.”¹
She is right about the bubble. She is right about the bailout. The analysis is real, and accurate, as far as the frame it operates from allows it to go.
That frame is part of what this piece is about.
OpenAI is projected to have $12 billion in revenue and an $8 billion operating loss for 2025. Its own forecasts show losses doubling to $17 billion in 2026, then doubling again to $35 billion in 2027.² Against more than $500 billion projected to be spent annually on AI infrastructure in 2026 and 2027, American consumers spend approximately $12 billion per year on AI services.³ A National Bureau of Economic Research study published in February 2026 found that despite 90% of firms reporting no measurable impact of AI on workplace productivity, executives continued to project AI-driven productivity gains — a gap between expectation and reality that the study compared to the productivity paradox of forty years ago.⁴ The product is not working at the scale the investment requires.
The bubble is real. What the product frame cannot see is not hidden from it — infrastructure was never a category it was built to measure.
What is being built — at $500 billion per year, financed by debt, drawing on the same water and power systems the ecological emergency is already pushing to their limits — is not primarily a product. It is infrastructure. And the question the product frame does not ask is: infrastructure for what?
The Physical Cost
The data centers now going up across the American landscape consume water at scales that belong in the same conversation as the aquifer depletion and hydrological disruption detailed in Part One of “The Ground Beneath the Emergency” series. Training one large AI model consumed 5.4 million liters of freshwater. Global AI-related water withdrawals are projected to reach 4.2 to 6.6 billion cubic meters annually by 2027 — four to six times the yearly water consumption of Denmark.⁵ Ireland’s existing data centers already exceed 20% of that country’s total national electricity demand. The US buildout is projected to consume 12% of all American electricity by 2028.⁶
Meta’s Hyperion campus in Louisiana has grown to nearly 4,000 acres — more than twice the size of New Orleans’ main airport.⁷ This is happening while the grid strains, while the aquifers drop, while the heatwave currently sweeping the US — as of this writing, July 2026 — is simultaneously testing the power infrastructure these facilities depend on and deepening the ecological emergency their construction accelerates.⁸
Will the investment return a profit? That is the only question a product frame is built to ask.
The question underneath it is: what does this infrastructure do when it is complete, regardless of whether it was ever profitable?
What It Is Actually For
On March 18, 2026, FBI Director Kash Patel confirmed to Congress that the FBI is buying Americans’ data from data brokers — including location histories — to track American citizens.⁹ AI facial recognition tools have been used to identify and punish protesters. AI-enabled analytics platforms consolidate government and commercial data on targets across what were once legally separate domains. The surveillance architecture that Part Two of “The Ground Beneath the Emergency” named as the infrastructure of triage — the apparatus being built to administer who is inside the wall and who is not, when the contractions arrive — runs on what is being built now, at a loss, in data centers competing for the water and electricity of a civilization already in stress.
The ruling class does not need the AI product companies to turn a profit. This is not a ledger anyone has shown us. It is what the behavior, read plainly, indicates. They need the infrastructure to work — and it works well enough, right now, for what it is actually being built to do.
This is not conspiracy. Conspiracies require secrecy. This is strategy — observable, documented, legible to anyone willing to watch the behavior rather than the statements. A decade and a half ago, a BBC drama series was portraying this exact calculus openly — in character dialogue, within the liberal establishment framing of British public broadcasting. The fictional container made the truth speakable that the same class wouldn’t say plainly elsewhere. That container is no longer necessary. The infrastructure is visible. The gap between what was speakable then in fiction and what is now plainly being built has closed.
The same ruling class that Part Two of this series shows building bunkers in New Zealand and Montana, acquiring private aquifer rights, consolidating farmland through institutional investment vehicles, and systematically dismantling the international cooperative frameworks that collective response to crisis would require — that class is also building, at extraordinary expense and without regard for quarterly returns, the technological architecture of population management.
Alexandria Ocasio-Cortez is among the most clear-eyed members of Congress, and what she named keeps attention exactly where the infrastructure needs it kept: on the investment cycle, the earnings reports, the congressional hearing. Not because anyone told her to look there. Because that is what product economics is built to see, and nothing else.
The bubble concern is not merely insufficient. It is useful to what it appears to critique.
The bubble may be real and the project may be succeeding at the same time. No frame built to price a product can hold both truths at once.
What Survives the Correction
When the correction comes — and it will come; the mathematics of $500 billion spent against $12 billion returned cannot hold indefinitely — the consumer-facing products will take the visible losses. Valuations will correct. Some companies will fail. What will not be dismantled is the data center network, the surveillance architecture, the biometric databases, the autonomous targeting systems, the identification apparatus already integrated into federal agencies across this administration. These do not require quarterly earnings to justify themselves. They require power, water, and political will — and the administration building them has all three, regardless of what the market does.
The correction will actually serve the consolidation. When the overbuilt sector deflates, the players with government contracts, whose infrastructure is already woven into federal systems, survive. The rest are acquired or dissolved. Data and architecture concentrate further — less distributed, more controlled, held by fewer hands. And the population absorbing the economic damage of the correction — pension funds exposed to AI valuations, workers displaced by the automation the infrastructure enables, communities whose water and power bills rose to serve data centers that then went dark — will have less capacity to resist what the surviving infrastructure is turned toward.
Jeff Bezos described the logic from inside it, without quite naming what it meant: “This is an industrial bubble. Investments will be wasted, companies will fail, but society will ultimately inherit a powerful infrastructure.”¹⁰ He offered this as reassurance.
Read that again.
Society will inherit the infrastructure. The question the reassurance does not ask — the question no frame built around price and profit could ever ask — is: which society, in whose hands, toward what ends, at what cost to the water and the power and the living systems already failing under the weight of what this civilization has already taken.
The bubble is real. So is what it is covering.
What the bubble is covering — the ecological, political, and contemplative reality beneath it — is examined in full in The Ground Beneath the Emergency, a three-part series.
Endnotes
¹ Rep. Alexandria Ocasio-Cortez, remarks as delivered, House Subcommittee on Oversight and Investigations, Committee on Energy and Commerce, November 18, 2025. Full transcript: ocasio-cortez.house.gov. Covered by Common Dreams, November 19, 2025. commondreams.org
² OpenAI financial projections reported in The Guardian (November 2025) and GMO Viewpoints (2026). — OpenAI projected $12B revenue against $8B operating loss for 2025; losses forecast to double annually through 2027.
³ Multiple sources on AI capital expenditure vs. revenue, including nadcab.com analysis (2026) and CBS News (June 2026). — $500B+ annually projected for AI infrastructure 2026–2027; US consumer AI revenue approximately $12B annually.
⁴ Yotzov et al., “Firm Data on AI,” National Bureau of Economic Research Working Paper 34836, February 2026. — 90% of firms report no measurable AI impact on productivity despite executive projections of significant gains.
⁵ Li et al., “Making AI Less Thirsty,” cited in multiple subsequent analyses including arxiv.org (2026). — GPT-3 training consumed 5.4M liters total including electricity-related water use; global AI water withdrawals projected at 4.2–6.6 billion cubic meters annually by 2027.
⁶ Lawrence Berkeley National Laboratory, US data center energy projections; Consumer Reports, March 2026. consumerreports.org — Data centers projected to consume 12% of US electricity by 2028; Ireland’s existing centers already exceed 20% of national demand.
⁷ Fortune, “Meta is quietly expanding its $10 billion Hyperion AI data center,” February 4, 2026. fortune.com; Bloomberg, “Meta Is Transforming Rural Louisiana With a $200 Billion Data Center,” May 18, 2026. bloomberg.com — Combined land purchases bring the Richland Parish campus to nearly 4,000 acres, more than twice the size of Louis Armstrong International Airport in New Orleans.
⁸ Al Jazeera, “US heatwave raises alarms over AI data centre energy demands,” July 3, 2026. aljazeera.com — Current heatwave straining power grid and water supplies simultaneously with AI data center expansion demands.
⁹ The Conversation, “US government ramps up mass surveillance with help of AI tech, data brokers — and your apps and devices,” June 2026. theconversation.com — FBI Director Kash Patel confirmed March 18, 2026 to Congress that the FBI purchases Americans’ location data from data brokers.
¹⁰ Jeff Bezos, quoted in multiple analyses of the AI bubble including nadcab.com analysis, 2026.
~ John Fridinger
Summer, 2026
Talent, OR