America Is Betting the Debt on AI Chips

For the first time, the main engine of American economic growth is not what households buy. It is what companies spend on data centers. In 2025, AI data center spending's contribution to GDP growth passed consumer spending, with Microsoft, Google, Amazon, and Meta forecasting a combined $364 billion in capital spending ^[1]. Renaissance Macro Research says the shift staved off a recession ^[1]. That fact reframes the whole economy. The question is what is holding it up, and at what cost. The answer is a bet, and the people making it have named it out loud. Elon Musk, who runs the government's cost-cutting effort, has been explicit about the stakes.

We are 1,000% going to go bankrupt as a country and fail as a country, without AI and robots. — Elon Musk

Goldman Sachs CEO David Solomon made the same case from the other side of the table, pointing to a national debt that has climbed from $7 trillion to $38 trillion since the financial crisis ^[2].

If we continue on the current trajectory and don't raise the level of growth, there will be a bill. — David M. Solomon

His answer is not austerity. It is growth, and the path he describes runs through AI infrastructure. The administration calls its version "Run It Hot": tariffs to raise revenue, a budget-neutral tax bill, and pressure on the Federal Reserve to keep rates low by installing a dovish chair ^[3]. Look at what each lever actually does, and they all point the same direction. A sovereign wealth fund, funded by tariff agreements, with a portion designated for AI chip projects ^[4]. An executive order opening 401k plans, covering roughly 90 million Americans, to a $12 trillion market of private equity and private credit, the same asset classes funding data centers ^[5]. Federal nuclear-weapons land at Idaho National Lab, Oak Ridge, and Savannah River converted to AI hubs under leases that cost the developers nothing upfront ^[6]. No single document ties these together as one plan. But each one moves capital toward AI compute. Wall Street is building the other half. Nvidia and six firms, BlackRock, Goldman Sachs, KKR, Apollo, Blackstone, and Brookfield, have signed agreements to mobilize more than $500 billion in outside capital for AI compute, with Nvidia potentially guaranteeing a quarter of the collateral based on the residual value of the chips ^[7]. Jensen Huang said it directly.

This is really the first time that technology chips have become an investable asset class. — Jensen Huang

So the bet is now a structure, and three cracks run through it. First, the returns are not arriving. MIT's Project NANDA reports that 95% of enterprise generative AI pilots have shown no measurable result, with $30 to $40 billion invested and no profit-and-loss impact ^[8]. OpenAI missed its internal revenue and user-growth targets in April, and its CFO warned the company may struggle to fund $600 billion in committed compute ^[9]. Second, the growth is the debt. The credit intensity of GDP has reached a 70-year high: the country now needs $3.73 of debt for every $1 of growth ^[10]. That is not growth paying down the debt. That is the debt being counted as growth. Third, the consumer is not being paid. Household debt hit a record $19.9 trillion while the savings rate fell to 2.6%, and spending is being sustained by rising stock prices, not wages ^[10]. Société Générale's Albert Edwards put the image on it.

This makes the economy all the more vulnerable should investors doubt the pot of gold at the end of the AI rainbow. Watch this debt-laden space. — Albert Edwards

Then there is the fiscal irony, and it is the sharpest turn in the whole structure. Yale's Budget Lab found that if AI shifts income from labor to capital, the revenue gains could be halved by 2030, because capital is taxed at lower rates than wages ^[11]. The growth strategy could worsen the deficit it was designed to cure. None of this means the administration is trying to replace the bond market. It is trying to hold it together. The national debt passed $40 trillion in August, months ahead of forecasts, and the sell-off pushed 30-year Treasury yields to 5.34% ^[12]. Treasury Secretary Scott Bessent is doubling buybacks of long-term bonds to provide liquidity support ^[12].

As Treasury Secretary, my job is to be the nation’s top bond salesman. — Scott Bessent

But here is the circularity. The whole growth strategy depends on global capital flowing toward AI compute rather than government bonds, and the two are now explicitly competing for the same pools of money ^[13]. AI capital spending and Treasury issuance are bidding against each other. The bond market, the stock market, and the consumer economy have become one structure. The consumer spends because AI stocks are up. The government borrows because it is betting AI will grow the debt away. Wall Street lends against chips because it believes the same thing. If investors stop believing in the AI growth thesis, all three crack at once, because they are the same bet.