The huge sums of money pouring into artificial intelligence are starting to draw closer scrutiny, as new disclosures highlight just how large the losses and borrowing behind the boom have become.
Anthropic's draft prospectus for a planned initial public offering reportedly shows a first-half net loss of $80.7 billion, with much of that figure tied to noncash accounting charges rather than cash actually going out the door. Even so, a loss of that size underscores how much capital frontier AI labs are burning through as they race to build and train ever larger models, all while needing to raise even more money to keep up. By contrast, SpaceX, Elon Musk's rocket and satellite company, reported $14.6 billion in quarterly revenue, with its AI division now generating more sales than its long-standing satellite internet business, a sign of how central AI has become even to companies not traditionally seen as AI players.
The strain is also showing up in credit markets. Much of the AI data center boom has been funded not just by equity investment but by debt, and that debt is increasingly being repriced by creditors who are reassessing the risk. CoreWeave, the Nvidia-backed AI cloud computing provider, had roughly $4 billion of its notes trading below 80 cents on the dollar in the week ending October 16, a level typically associated with financially distressed companies. Borrowing for data centers across Asia has approached $29 billion, and in the four weeks leading up to October 16, three data center developers backed by Nvidia, SoftBank and Blackstone delayed or shelved planned initial public offerings, suggesting investors are growing more selective about which AI infrastructure bets they are willing to fund.
The caution has reached beyond the biggest U.S. players. In Australia, data center operator NextDC has lost more than 40% of its market value following delays tied to securing enough electricity for its facilities, a reminder that the physical constraints of power supply, not just chips or capital, are becoming a real bottleneck for the industry's growth plans. Meanwhile, Nvidia is pushing in the opposite direction, introducing modular AI computing systems built more like shipping containers than conventional data centers, an approach designed to let customers deploy computing capacity faster.
Taken together, these developments suggest the AI investment boom is entering a more discerning phase, where headline revenue growth and ambitious plans are being weighed more carefully against actual cash losses, debt loads and real-world constraints like electricity supply.
Watch for Anthropic's formal IPO filing and pricing, and for whether more data center developers delay or cancel public offerings as credit investors grow more cautious.
Why did Anthropic lose $80.7 billion?
Much of the loss reported in its draft IPO prospectus is tied to noncash accounting charges rather than cash spending, but it still reflects the enormous cost of building and training frontier AI models.
Why are data center IPOs being delayed?
Investors and creditors are growing more cautious about the debt used to fund AI data centers, and some projects also face delays tied to securing enough electricity.