Skepticism about artificial intelligence has swept back through global financial markets. Forecasts emerged that the revenue hyperscalers earn from AI will fall far short of their capital investment, and on the same day, "AI credit risk" came to the fore as the financial burden on debt-reliant AI infrastructure companies drew attention. News that revenue at Anthropic, the operator of Claude, has more than tripled this year was not enough to quiet doubts that AI monetization remains a distant prospect.
Market research firm Gartner said in a report on the 14th that $3.5 trillion (about 4,713.8 trillion won) in AI infrastructure spending will be at risk of impairment by 2030. Will Sommer, a senior analyst at Gartner, said in the report that hyperscalers will spend about $1 trillion on AI infrastructure this year and invest an additional $6 trillion from next year through 2030. The revenue generated, however, will fall well short of that capital spending, the report projected. To cover their financing costs — a minimum return on invested capital of 12% that hyperscalers must deliver to investors — they would need to generate cumulative revenue of $8.5 trillion by 2030, but the actual figure will reach at most $3.6 trillion, according to the assessment.
Sommer said more than 80% of neocloud providers will cease operations by 2030. Unlike hyperscalers with diversified revenue streams, neoclouds — whose main business is AI computing and GPU-as-a-service cloud offerings — will absorb the full brunt of the shock.

CoreWeave, the largest neocloud provider, saw its shares fall 15.5% that day as the financial strain from its AI investments came into focus, while the yield on its corporate bonds jumped more than a percentage point to 15% (a 9.625% coupon maturing in 2032). The selloff followed remarks by CoreWeave co-founder Brannin McBee, who told the Financial Times the company would secure additional investment-grade ratings on chip-backed loans to draw in funding from insurers.
Gil Luria, an analyst at D.A. Davidson, said CoreWeave, which buys Nvidia chips on debt to build data centers, currently falls into the subprime tier with the greatest risk of losses, adding that the company is borrowing at 9% and earning 5%. He went so far as to say CoreWeave's share price could fall to zero. Bonds issued by CoreWeave fell below 80 cents on the dollar, and the probability of default within five years implied by credit default swap prices topped 50%. Tesla Chief Executive Elon Musk responded to the news on his social media account with a single word: "yikes."
As the AI-driven bad news rattled equity markets, 10-year government bond yields fell sharply during trading on the 15th in the United States (down 0.099 percentage points) and Japan (down 0.096 percentage points). Analysts said money that had left the sovereign bond market amid the surge in yields has returned alongside a flight to safety.
Even so, Nvidia appears set to capture the profits. JPMorgan said the free cash flow that five companies — Microsoft, Amazon, Meta, Google and Oracle — are burning on AI infrastructure investment is flowing straight to Nvidia.
Anthropic, which is preparing an initial public offering before year-end, posted third-quarter revenue of $16.8 billion, up 45% from the previous quarter, Reuters and other outlets reported, citing sources. That extended a steep run of growth after $4.7 billion in the first quarter and $11.6 billion in the second. The Financial Times noted that the company has turned a profit for two consecutive quarters on an adjusted basis, but said growth has slowed compared with early this year, when revenue was doubling. Bloomberg reported that market analysts are applying tougher standards to Anthropic on funding, cost of capital and returns.
There is also no shortage of "too big to fail" arguments that slowing AI investment now would trigger an immediate recession. Larry Fink, chief executive of BlackRock, Wall Street's largest asset manager, stressed that day that the United States would fall into recession if the buildout of AI infrastructure such as data centers were halted in any form.
