Wall Street is financing AI’s boom by borrowing against profits not yet earned. The rally is real. So is the strain beneath it.

Nvidia rose 2.4% Monday, and Wall Street behaved as if AI’s most difficult question—how the enormous buildout would be financed—had been answered by demand itself. It had not. The S&P 500 rose 0.7%, recovering last week’s losses and coming within 1.8% of its record set earlier this month. The Nasdaq added 0.8%, the Dow gained 0.3%, and Nvidia supplied the strongest lift, beginning to repair the damage from three consecutive losing weeks—its longest such streak since June.

AI’s tape has been unusually sensitive to corporate developments. Technology shares rose last week on SpaceX, Anthropic and Nvidia news, only for Nvidia to enter this week nursing that losing streak. Tuesday’s follow-through showed the same selectivity.

Asia embraced the signal with characteristic force. South Korea’s Kospi jumped more than 4% and traded 3.6% higher at 5,667.99 by midday. SK Hynix surged 7%, Samsung Electronics gained 4.8%, and Japan’s Nikkei 225 added 2.2% to 63,542.77. Advantest rose more than 10%, SoftBank Group climbed 7%, Taiwan’s Taiex added 0.5%, and Australia’s S&P/ASX 200 gained 0.3%.

Not everyone followed. Hong Kong’s Hang Seng surrendered its early advance and slipped 0.2% to 23,174.44, while the Shanghai Composite edged down 0.1% to 3,692.58. Traders were watching for announcements from a closed-door high-level meeting, or plenum, of China’s ruling Communist Party. The regional picture was nevertheless clear: AI hardware had pulled global risk appetite back from the edge.

But the more consequential development is what the boom now costs. Investors are no longer merely debating whether AI companies can sell more software. They are confronting the hardware, data centres and debt required to keep the expansion moving. AI-related shares remained unsettled because the boom has become extraordinarily expensive to fund, and expensive borrowing could strain the sector’s most leveraged ambitions. Higher borrowing costs increase the interest burden attached to data-centre construction and chip purchases, forcing investors to demand more revenue and profit from every dollar of AI spending.

Meta outlined plans to raise $35 billion through a stock sale to help finance its AI investments. Nvidia, Advanced Micro Devices and Broadcom have indicated that they may buy a total of $5 billion of Meta stock in the offering.

That is more than a side detail. Nvidia, AMD and Broadcom would be helping finance the customer whose capital purchases their own products. The same dollars circle the loop: the chipmakers back the balance sheet, and that balance sheet buys the chips. Every turn of the circle raises the cost of getting the return wrong.

Anthropic disclosed how much money it was losing while preparing to sell stock in the publicly traded US market for the first time. SpaceX, which also operates xAI, announced a roughly $40 billion deal with an entity managed by Apollo Global Management to lease AI computing power. The AI trade is no longer companies casually buying software with cash from existing operations. It is a vast construction effort requiring new equity, major contracts and increasingly confident assumptions about future demand.

Here is the engineering and financing substance, which is less ambiguous than the corporate slogan: a chip order moves capital; a computing lease multiplies it; a stock sale spreads risk. None of those things is a return until the workload and cash arrive. The instruments say what is happening; the adjective “AI” does not.

Then name who is carrying the load, because the financing runs through two channels and they bill different people. Equity is doing most of the work: Meta is selling stock, and Anthropic is preparing its first public-market sale. Existing shareholders can be diluted, and new shareholders are putting fresh capital at risk, so the bill can keep climbing. Stock sold before the profits arrive is borrowed capital too—borrowed from shareholders against a future that has not been built yet.

Strictly speaking, equity is not a loan. That distinction matters to accountants and rather less to the shareholders whose claims on future earnings have just been diluted. Credit is doing the rest: the data centres are being built with borrowed money, and that debt comes due whether or not the promised productivity arrives. New shareholders and bondholders are both being asked to fund a payoff that does not exist yet.

Every leap of the AI boom now arrives with a financing footnote. Investors can tolerate enormous spending when the spending is attached to visible revenue and rising profits. To be fair, none of this proves that the investments will fail, and financing construction before every customer has arrived is not itself evidence of fraud. The trouble is that the largest commitments are arriving before that attachment is visible, while the market treats the financing itself as evidence of future demand.

Debt does not care whether the model gets smarter. It cares about the date and the payment.

The bond market is sharpening the pressure. The reported 10-year Treasury market yield slipped to 4.98% from 5.00% late Friday, while the two-year yield rose to 4.71% from 4.69%. The two-year rate more closely tracks expectations for overnight interest rates, and investors widely expected the Federal Reserve to leave the federal funds rate unchanged when its meeting ended Wednesday. The Fed had raised the rate the previous month for the first time in three years, seeking to slow an inflation rate that remained uncomfortably high. Borrowing is not becoming cheap enough to make a capital-intensive boom effortless.

With the Fed widely expected to hold, Wednesday’s earnings—not the rate decision—become the primary test of whether AI spending can earn its cost of capital. Weak results would expose the cost of the buildout; only a surprise hike would compound the pressure.

Oil provided some relief. Brent crude fell 0.8% to $109.82 a barrel, and the US benchmark dropped 0.8% to $100.93. It is true that lower energy prices eased one source of inflation pressure. It was not enough to erase the Fed’s dilemma. The dollar rose to 157.65 Japanese yen from 157.50, while the euro slipped to $1.1173 from $1.1176.

The rally was not universal either. Brazil’s Bovespa plunged roughly 14% after President Luiz Inácio Lula da Silva won reelection in a result that surprised investors. Markets feared that Lula could increase government spending and intensify pressure on Brazilian debt. The plunge demonstrated that political shocks still have the power to overwhelm a favourable global backdrop. Capital was concentrating around one trade, not embracing risk everywhere.

Even so, oil’s retreat and Asia’s advance fit the pattern captured in Asian stocks’ rebound after the technology selloff: risk appetite remains responsive to the AI trade, while capital retreats from markets suddenly burdened by political or fiscal risk.

That makes this week’s earnings reports an audit rather than a ritual. Alphabet and Microsoft report Wednesday, with Amazon following Thursday. Roughly one stock in every three in the S&P 500 is scheduled to report. The decisive question will not be whether the companies mention AI. It will be whether AI spending starts producing the profits and productivity promised on the other side of the capital now being raised and contracted—Meta’s $35 billion, SpaceX’s $40 billion. The period is a substantial verdict on corporate profitability, not merely a test of a few technology names.

The useful discipline is a plain reconciliation: capital committed, debt due, revenue contracted and cash returned. An earnings report that cannot bridge those four items cannot convert repetition of “AI” into proof.

If Alphabet, Microsoft and Amazon demonstrate that AI spending is generating cash and productivity, then the enormous capital raises become a bridge to a larger economy. If they do not, Wall Street will discover that it has valued the financing pipeline more carefully than the returns flowing through it. Nvidia can keep rising. Wednesday is when the receipts begin to arrive.

Nvidia opened the door. Asia charged through. But the market has not delivered a verdict of permanent victory. It has voted for AI while asking who will pay and whether the earnings can service the debt. For now, the rally is running on borrowed capital—and borrowed capital always demands proof. Credit, results and the Fed still control the lock.