SILICON VALLEY — A funding squeeze touched off by Meta is blocking the path of Big Tech companies racing to invest in artificial intelligence. Their spending has grown so quickly that they have exhausted the options available to them — internal cash, corporate bonds and circular financing. Yet the companies say demand remains overwhelming and that they cannot stop raising money.
Combined capital spending at five companies — Microsoft, Alphabet, Amazon, Meta and Oracle — is projected to total $945.55 billion in January next year, roughly six times the level of January 2023, Reuters reported on the 22nd. Their free cash flow over the same period is expected to swing to a deficit of $2.87 billion from a surplus of $223.11 billion.
What chilled the market was the speed of Meta's spending. The company disclosed that day that it has committed nearly $825 billion, or about 1,100 trillion won, to AI infrastructure.
Of that, $355.7 billion consists of operating and finance leases that are not reflected on its balance sheet, including rent on data centers and energy infrastructure through 2036. Separate non-cancelable short- and long-term contracts, such as third-party cloud computing services, account for more than half of the total at $454.5 billion. Adding up to $14.7 billion in mandatory cloud capacity purchases over five years brings the commitments to $825 billion. That is $125 billion more than the $700 billion disclosed at the end of July, an increase over just three months.
Immediately after the announcement, the premium on credit default swaps tied to Meta — a gauge of the risk that the company defaults — jumped 8.9 basis points from the previous session to 119.3 basis points. A basis point is one-hundredth of a percentage point. Credit default swaps are derivatives sold as protection against corporate default, and a higher premium signals greater risk of non-payment. The reversal came just a day after Meta posted record third-quarter revenue on the strength of its AI agent Muse.
To raise money quickly, Meta plans to issue non-voting common shares, known as Class C stock, in January next year. The company will distribute one new share for each share held by existing shareholders, then issue additional shares of the same class after a set period and sell them into the market.
Anthropic, which has faced scrutiny over its losses, also appears set to press ahead with its initial public offering as planned. Reuters reported that Anthropic plans to file its registration statement as early as the end of this month and carry out the IPO in the middle of next month.
Anthropic spent $9 billion in the first half of this year alone on computing capacity and infrastructure, and contracts it must pay for computing capacity over as long as the next decade total $450 billion. Including additional negotiations now underway, that spending would swell to as much as $750 billion, or about 992.78 trillion won.
OpenAI, which is preparing an IPO next month, also appears likely to proceed on schedule even after lowering its valuation target to $1.25 trillion from $1.4 trillion. The Financial Times reported the change, saying it followed complaints from some investors that the target was too high.
Concern about Big Tech is spreading across the market, but that is becoming the very justification for the rush to raise capital. Meta said the market environment is growing more difficult because of large-scale AI fundraising by companies including Alphabet, SpaceX, Anthropic and OpenAI, and stressed that it needs to secure money ahead of its rivals.
Nvidia Chief Executive Jensen Huang made a similar argument that day at the company's GTC 2026 event in Berlin, saying the obstacle to expanding AI investment is funding rather than weak demand. He also defended the so-called neocloud providers at the center of investor concern — CoreWeave, Nebius and IREN — saying demand for their services is growing so explosively that it runs off the chart.
"What is holding back the growth of neocloud companies is not demand but working capital," he said. "This problem will be solved, because what fundamentally matters is the value these companies provide."
Industry watchers read Big Tech's moves as an effort to draw more money from private markets through investment banks while counting on financial support from the U.S. government.
The administration of President Donald Trump issued an executive order in July last year setting out loan guarantees, lending, subsidies and tax benefits, and long-term government purchase agreements as measures to support data centers. The industry wants that support expanded further.
The order covered not only data centers meeting certain requirements but also related energy, manufacturing and infrastructure projects. As more Big Tech companies procure their own electricity to run AI data centers, analysts say aggressive use of these programs by the U.S. government could give the companies breathing room in their fundraising.