NBD Commentary: OpenAI Cuts Valuation and Funding as the U.S. AI Supply-Side Narrative Reaches a Turning Point

Article body

Du Hengfeng, Commentator of National Business Daily

On October 31, according to multiple media reports, OpenAI lowered the target valuation of its new round of financing from $1.4 trillion to $1.25 trillion, a valuation reduction of $150 billion, or about 10.7%; the financing size was also cut from $30 billion to $20 billion, a direct reduction of one-third. For a company regarded as a "bellwether of the AI era," this is not only a concession in financing terms, but also represents a profound shift in the growth narrative of the US AI industry.

This round of expansion in the AI industry carries a distinct feature of "supply creating demand." Whether out of extreme optimism about the prospects or fear of being left behind by the times, US tech giants and frontier AI labs have all made "stacking computing power" their core goal. The six hyperscale cloud providers' AI capital expenditures for the two years from 2026 to 2027 total as much as $1.73 trillion, with planned spending for 2026 alone approaching $1 trillion, comparable in scale to the US annual defense budget.

AI labs have pushed expansion to the extreme. In March this year, OpenAI completed $122 billion in financing at a valuation of $852 billion. Over the past 15 months, the company has signed long-term computing power and data center agreements with a nominal total of more than $1.4 trillion, about 35 times its current annualized revenue. The computing power investment plan disclosed in Anthropic's prospectus is similarly enormous: over the next decade, its committed cloud computing and computing power spending will reach as much as $518 billion, about 80% of which is non-cancelable. The company also plans to build its own data centers, and its debt-to-asset ratio is already close to 80%.

As the absolute core of AI computing hardware, Nvidia has adopted another strategy. By directly injecting capital into computing service providers such as CoreWeave, it has not only helped spawn a large number of "neocloud" companies, but also, by investing in upstream companies such as Marvell Technology, bound the expansion of the industrial chain upstream and downstream into its own "growth flywheel." After recently announcing its subscription to Meta's private placement shares, Nvidia also applied this model to the Magnificent Seven for the first time.

At this point, the entire US AI industry has formed a self-reinforcing cycle: raise valuations, obtain financing, expand hardware or computing power procurement, improve model capabilities, create more revenue, and then raise valuations again. But this cycle is built on continuous external financing. Now, with the financing environment tightening across the board, the industry's expansion logic has undergone a qualitative change.

The first source of pressure comes from the debt risk of "neocloud" companies. Represented by Nvidia-backed CoreWeave, which has expanded through high leverage, its debt has ballooned from about $15 billion to $52 billion, the yield on its unsecured bonds has approached 15%, and the market has even forecast a 63% probability of default within five years. Its stock price has retreated more than 67% from its historical high, dragging shares of neocloud companies such as Nebius and IREN down by as much as more than 40%. Musk also issued a warning, saying that more than 200 "neocloud" companies will experience guppy-like deaths and mergers, and "some will survive, but not many."

The second source of pressure comes from the adjustment in the secondary market. The Nasdaq 100 index fell 6.65% in October, the worst monthly performance since March 2025. Free cash flow at Meta, Google and Amazon turned negative, and S&P lowered its rating outlook on Meta to negative. Tech giants had originally relied on large-scale bond issuance and secondary stock offerings to sustain capital expenditures, but as of the close on October 30, the stock prices of Meta, Alphabet and Intel continued to decline, and the pricing of the three companies' secondary offerings was already higher than their current secondary-market prices, making subsequent equity financing by them or other companies more difficult.

The third source of pressure comes from Anthropic's upcoming IPO. As OpenAI's most direct competitor, Anthropic's listing valuation has been revised down all the way from $2 trillion to $1.5 trillion. When even the highest-quality assets in the sector must accept harsh pricing in the secondary market, the valuation and pricing logic of the primary market has begun to loosen across the board. Morgan Stanley previously pitched OpenAI to private wealth clients at a valuation of $1.4 trillion, but market feedback was ultimately lukewarm, with no investors actively scrambling for allocations.

With financing becoming a hard constraint, the self-cycle on the supply side is difficult to sustain. OpenAI's move this time to "cut valuation and cut financing" shows that the old model of continuously raising valuations, taking money to expand, and then raising valuations again has reached its end. The market no longer pays for "stories," but instead asks the simplest business question: can astronomical investment actually bring reasonable returns?

Obviously, the answer to this question is no longer the headlong advance of infrastructure and computing power, but the quality of the demand side, that is, whether application scenarios can truly be implemented and whether the quality of revenue and cash flow can withstand scrutiny. McKinsey research shows that nearly 80% of companies deploying AI have failed to achieve an increase in net profit, and 95% of generative AI pilots have brought no direct financial return. Goldman Sachs estimates that for the six hyperscale cloud providers to achieve a 15% return on about $1.73 trillion in computing power investment from 2026 to 2027, they would need to generate a cumulative total of about $1.42 trillion in revenue between 2028 and 2030.

(Source: National Business Daily) Middle of the article Embedded advertisement