OpenAI Pushes Forward With $30 Billion Fundraising; Anthropic Prospectus Discloses Massive Compute Commitments

Recently, OpenAI and Anthropic have successively revealed progress on financing and going public.

OpenAI is advancing a financing round of about $30 billion, with a target valuation exceeding $1.4 trillion, seeking to lock in investor commitments before rival Anthropic's IPO, which is expected to launch in November.

According to disclosures, Abu Dhabi investment firm MGX and other UAE funds are discussing a combined investment of up to $10 billion, and BlackRock is also in talks to participate. The financing has not yet been completed, and the investor list and specific terms may still change. OpenAI's previous financing round, completed in March this year, was $122 billion at a valuation of about $852 billion.

Meanwhile, the draft IPO prospectus of Anthropic discloses that the company's revenue continues to grow, but at a slower pace, while computing power spending, accounting losses and long-term purchase commitments are equally staggering.

As key samples, the financial performance of the two leading model makers is driving a shift in capital market attitudes. Since October, AI-related stocks have been under sustained pressure, and valuations across the computing power supply chain have contracted. What the market worries about is not just short-term sentiment, but also slowing revenue growth, rising financing costs for AI infrastructure, and whether massive capital expenditure can be sustained.

The revenue test for OpenAI and Anthropic

OpenAI is accelerating financing both to replenish funds for continued expansion and to compete for the capital market's attention. As Anthropic's listing approaches, the revenue growth, business models and capital needs of the two companies will be more directly compared side by side.

OpenAI's revenue is still growing, but market expectations have already been adjusted. According to market sources, OpenAI's annualized revenue run rate was close to $50 billion as of the end of September, lower than the $70 billion widely cited by the market previously.

The difference is mainly based on different revenue accounting standards. The gross method adopted by Anthropic includes some revenue generated through partner channels such as Amazon AWS and Google Cloud in annualized revenue, while OpenAI only records the portion it actually receives, and the share taken by cloud providers is not included in revenue.

Anthropic's growth is also undergoing stricter scrutiny. According to disclosures, the company's revenue in the first half of 2026 reached $16.3 billion, compared with $1.1 billion in the same period last year; its net loss in the same period was $80.7 billion, of which about $72 billion came from accounting expenses arising from changes in the valuation of convertible notes and other financing instruments. Third-quarter revenue grew 45% quarter-on-quarter to $16.8 billion, but the growth rate slowed compared with previous quarters.

Xie Siyuan, managing director of Yijing Capital, told Yicai that leading edge in frontier technology and revenue scale are not mutually exclusive, and large enterprise customers tend to prioritize closed-source models with leading technology. Moreover, leading manufacturers are still reserving computing power for next-generation models, and the competition in capital investment is not over.

Therefore, the capital market is not ceasing to value technology, but demanding that technological advantages prove their commercial value. Model capabilities determine whether a company can enter high-value scenarios, while customer renewals, pricing power, gross margins and cash flow determine whether this advantage can be sustained.

Hundreds of billions of dollars in computing power commitments are both a foundation and a burden

Competition in frontier models is inseparable from computing power, but investment in computing power is turning from a growth engine into a long-term financial constraint.

The financial data disclosed this time show that Anthropic's existing computing power-related commitments have exceeded $600 billion and may increase further in the coming years. Its IPO draft also involves future computing power investment that may be as high as $750 billion.

Anthropic's spending on computing power and infrastructure in the first half of this year has approached $9 billion, exceeding the approximately $7.3 billion in related spending for the whole of 2025. The company has long-term procurement arrangements with suppliers such as Google Cloud, Amazon AWS and Microsoft, and some contracts have long terms and strict cancellation conditions. Long-term agreements help ensure computing power supply, but also mean that companies must continue to generate revenue for many years to cover the corresponding obligations.

Financing cost is another pressure. Anthropic is reportedly advancing a chip financing transaction, mainly to purchase about 1.5 gigawatts of Google's custom TPU (tensor processing unit) chips to expand its AI computing infrastructure. A Blackstone-related fund plans to invest $9 billion in debt, with a related yield of more than 10%.

Even for fast-growing leading model companies, obtaining the funds needed for expansion requires paying a not insignificant risk premium. Interest rates above market standards reflect that investors have become more cautious about the concentrated financing needs of AI companies in the short term and are demanding higher risk compensation.

In response, Xie Siyuan said that model makers are still investing continuously to strive for next-generation technological advantages. He believes that if global models stopped commercial expansion and investment in Scaling Law, the profit situation of leading companies would quickly become apparent. But the premise for this view to hold is that continuous investment can ultimately bring enough incremental revenue and competitive advantage. If model prices fall, customer demand slows or financing costs rise, long-term computing power commitments may also turn from competitive barriers into financial burdens.

Chinese model makers face similar choices. Xie Siyuan said that Chinese models have a relatively clear domestic market, but need to further expand capital expenditure, and Alibaba and ByteDance are carrying out related expansion. For Chinese companies, the key is not the scale of investment, but whether they can form an effective closed loop among model capabilities, inference costs, product ecosystems and local customer demand.

At the same time, a clearer model stratification may emerge in the market: highly complex tasks will be contested by frontier closed-source models, while cost-sensitive, high-frequency invocation scenarios are better suited to lightweight models and open-source models.

Xie Siyuan summarized this as "users flowing to mid-tier models, but profits flowing to top-tier models." This judgment means that the competitiveness of Chinese models should not be measured only by low prices, but also by whether the open-source ecosystem, deployment flexibility and large-scale service capabilities can be converted into stable revenue.

The capital market gives harsh feedback

Since October, AI-related sectors have come under pressure, and the market has begun to reassess the payback period of computing power investment.

As of the close on October 20 local time, BlackRock's semiconductor-themed ETF SOXX had fallen about 16.15% from $596.44 on October 2. The market is worried about slowing revenue growth at leading model vendors, rising financing costs for AI infrastructure, and whether huge capital expenditures can be sustained.

Ken Griffin, founder and chief executive of U.S. hedge fund Citadel, said publicly on October 20 that investors are almost certainly in an AI bubble. What worries him is not that AI has no value, but that companies' ability to convert AI into productivity gains varies widely. The market is re-examining the bills of OpenAI and Anthropic to confirm whether the investment is truly bringing returns.

He also believes that a bubble does not mean the underlying technology lacks long-term value. Major economic and technological changes are often accompanied by speculative fever, and companies that can survive the bubble cycle still have a chance to become long-term winners in the future.

This is also the core divergence facing the computing power industry chain. Large cloud service providers with diversified businesses and cash flow can diversify AI investment risks to a certain extent; new cloud vendors that rely on external financing to buy chips and then rent out computing power are more vulnerable to customer concentration, debt costs and equipment depreciation.

The primary market is also no longer dominated by optimistic expectations. Anthropic is seeking a valuation of up to about $2 trillion, but valuation expectations from the capital market have already been revised downward. Independent research firm New Constructs even gave a low-case valuation of about $144 billion and questioned the growth and profitability assumptions required for a higher valuation.

The firm judged that to support a $2 trillion valuation, Anthropic would need to achieve $358 billion in after-tax operating profit in 2035, equivalent to twice Nvidia's relevant figure over the past 12 months and 2.6 times Apple's; revenue would need to reach $3.6 trillion, which is 1.4 times the combined revenue of Microsoft, Apple, Alphabet, Amazon, Nvidia, Meta and Oracle.

In the long run, Xie Siyuan believes that the computing power growth trend will not stop in the next five years, but whether the growth comes from real demand or from companies betting in advance on future demand also needs to be further clarified. Computing power procurement contracts, chip orders and data center construction can all show the scale of investment, but they cannot automatically prove that end users are willing to keep paying at a sufficiently high price.

The key to this competition is changing. In the past, ample funding could help companies train larger models; now, huge investment itself has become an operating variable that must be tested. Model vendors must prove that additional computing power can be converted into higher-quality revenue, computing power suppliers must prove that long-term contracts are executable, and investors must distinguish between the long-term value of technology and current asset prices.

A trillion-dollar valuation does not mean the market has confirmed the return on AI investment. As OpenAI's financing and Anthropic's listing approach, the capital market is demanding that leading model companies prove that their performance growth is not only fast enough, but also sufficient to support the computing power commitments they have already made.