OpenAI’s Ask: Be Our Banker
What OpenAI’s guarantee request tells us about the merger of state power and private ambition.
Picture this: a windowless conference room, government badge lanyards tangled with venture-capital Patagonia vests, a draft term sheet on the table stamped PARTIAL GUARANTEE. The tech side talks velocity risk. The agency lawyer circles loan-to-value. Someone murmurs national security and the room exhales.
If the bets pay off, the private equity stack gets rich. If they don’t, taxpayers pay.
If the business succeeds, profits belong to the AI oligarchs; if it fails, the taxpayer is liable.
OpenAI’s CFO, Sarah Friar, asked Washington for a federal backstop on the eye-watering buildout of chips and data centers. After blowback, the company tried a reframe: not a bailout, just support for industrial capacity for technology leadership. The financial mechanics didn’t change. A guarantee slices the interest rate, expands leverage, and shifts default risk from Microsoft’s friends and Wall Street’s syndicates onto the public.
A subsidy. Non-dilutive, invisible on the cap table, and very real on the federal balance sheet.
The AI race now runs on debt and electricity. The tab? Roughly $3 trillion globally to erect hyperscale data centers and buy specialized chips. Capital that ages like milk, not wine. GPUs go obsolete in two to three years. Data-center layouts designed around today’s models can be stranded by tomorrow’s architectures.
Private creditors price that velocity risk. A federal guarantee socializes it.
Oracle is looking at $300 billion in projected deals over six years. Broadcom $350 billion. Microsoft $250 billion. These are vendor contracts for building infrastructure that might be technologically irrelevant before the loans mature.
Guarantees raise borrowing capacity, lower rates, and inflate valuations by juicing the net present value of future cash flows. OpenAI says margins look very healthy absent aggressive spend. The guarantee lets them keep spending at hyperscale while taxpayers swallow the tail risks of obsolescence and market whiplash.
Talk all you want about innovation. This is financial engineering with a national-security wrapper.
And it’s not just one company. The government is already entwined with the AI stack. Exclusive procurement. Formal collaborations. Preferred-vendor pipelines that naturally entrench incumbents. Guarantees available to a few strategic partners become a moat against smaller rivals who don’t have a hotline to OMB or a joint announcement with a cabinet official.
It’s corporatism: state power underwriting private dominance.
Let’s give the strongest argument its due.
China runs a tight, state-led industrial strategy across the compute stack. If the U.S. dithers, it loses. Guarantees are faster than grants, leaner than nationalization, and cheaper than direct spending because they activate private credit markets. The Congressional Budget Office itself says partial guarantees can share risk with lenders and lower budget costs.
The U.S. currently holds a commanding lead in compute. The country controls about 75% of global AI capacitycompared to China’s 15%. Keeping that lead will take speed, not sermons. In this view, guarantees are a bridge to scale critical infrastructure now, before supply chains tighten and geopolitics turn uglier.
Three problems, none small.
Moral hazard on steroids. The guarantee structure severs risk from consequence. If the AI buildout stalls or tech pivots render racks obsolete, the losses flow to Treasury, not term sheets. Solyndra secured a $535 million federal loan guarantee before filing for bankruptcy in 2011. Taxpayers ate hundreds of millions in losses. That fiasco was a rounding error compared to the trillion-dollar stakes here.
The assets then were specialized but tangible. Today’s are bleeding-edge GPU clusters that can be technologically devalued overnight.
Concentration as policy. Handing cheap, government-cushioned capital to a tiny club cements an oligopoly. The state becomes dependent on a few leveraged gatekeepers for national assets. If any of them wobble, the country’s strategic compute capacity wobbles with them.
Public risk, private upside. There’s no automatic public equity stake, no profit-share, no clawback. So yes, it’s a bailout. Pre-authorized. If Washington insists on underwriting the downside, taxpayers deserve the upside: equity, revenue participation, or guaranteed public-price access to compute.
Otherwise it’s just a transfer.
Out by a substation on the exurban fringe, a new hyperscale slab hums like a jet hangar. The utility upgraded miles of lines to feed it. Local officials carved out tax abatements. A federal letter of guarantee sits inside the project finance binder.
Two years in, a breakthrough in model efficiency torpedoes demand. Capacity correction, the bankers call it. The resale value of last-gen accelerators craters. The senior tranche is fine. Because you’re the senior tranche.
If Washington insists on guarantees, the bare minimum: take equity or revenue. No upside, no guarantee.
Open access mandates. Guaranteed, non-discriminatory compute for startups, universities, and independent researchers financed by the facility taxpayers just backstopped.
Hard risk-sharing. Partial guarantees with real private loss absorption. Credit-risk-transfer instruments to expose true pricing.
Transparent criteria. Define strategic asset in public, with published scores, not ad hoc vendor selection.
Or consider alternatives that don’t privatize the asset. Massively expand a publicly owned National Research Cloud. Use performance-based grants tied to measurable outputs rather than blanket credit wraps. Domestic jobs. Domestic components. Defense or medical benchmarks.
What’s missing should worry anyone paying attention.
The record is thin where it counts. No disclosed size of the guarantee ask. No program pathway. No guarantee percentage. No recourse or clawback terms. No binding public-access provisions. The widely cited compute-share numbers are estimates, not an audited baseline.
And the market-demand forecasts that supposedly justify a $3 trillion build? Heavy on projection, light on contracted offtake. If someone wants a balance sheet, they owe receipts. The receipts aren’t here.
This is supposedly about national security. Fine. Then be honest: this is industrial policy with a bailout valve. If losses get nationalized, the infrastructure should be nationalized too. Or at least the profits.
Otherwise, the fusion of state and capital hardens into a single organism that feeds on public risk and spits out private wealth. Beijing’s state-capital model doesn’t get beaten by becoming a worse, less accountable version of it.
If Congress sleepwalks into trillion-dollar guarantees without upside or guardrails, don’t call it strategy.
The next great transfer. From public to private, signed in the name of innovation, paid for in the currency of trust.