Anthropic Overtakes OpenAI in Private-Market Valuation as Chinese AI Offensive Fuels Overvaluation Warnings

Private-market valuation surpasses $1.2 trillion
AI model market increasingly shifts toward Chinese providers
Widening gap between corporate valuation and market competitiveness

U.S. artificial intelligence startup Anthropic has overtaken OpenAI in private-market valuation. The surge reflects intense investor demand driven by expectations of further gains following an initial public offering, but critics caution that the valuation may not reflect the company’s underlying worth because a substantial share of the transactions is conducted indirectly through special purpose vehicles (SPVs). In the AI model market itself, the rapid rise of lower-cost Chinese models is eroding the usage share of Anthropic’s Claude family. Anthropic has responded by expanding free token allocations for startups in a bid to lock in customers, but soaring computing costs are increasing the pressure to secure both price competitiveness and profitability.

Scarcity Frenzy Even Before the IPO

According to a July 12 report by U.S. business publication Business Insider, Anthropic shares are trading on the private-market platform Caplight at an implied valuation of $1.2 trillion. That is substantially higher than the $965 billion valuation the company secured in its Series H funding round in May. On Caplight, OpenAI is valued at $908 billion, placing Anthropic ahead of its rival. “Anthropic is the most sought-after company in the history of the venture secondary market,” Caplight CEO Javier Avalos said.

Glenn Anderson, CEO of private-market brokerage Rainmaker Securities, also said Anthropic shares were trading at a $1.2 trillion valuation. “Even at prices that high, virtually no one is willing to sell,” he said. “If I could fill all the buy orders on my books, I would be on a beach instead of doing this interview.” Some investors have reportedly gone so far as to offer their homes in exchange for Anthropic shares, underscoring the extraordinary buying frenzy.

Market observers attribute the disappearance of sellers to mounting speculation that Anthropic’s IPO is imminent, which has fueled expectations of further gains after the listing. The company’s recent surge in revenue has also contributed to its elevated valuation. Anthropic projects second-quarter revenue of $10.9 billion, up 130% from the previous quarter, and expects to post its first quarterly operating profit since its founding. Matt Murphy, a partner at early Anthropic investor Menlo Ventures, nevertheless pushed back against the private-market valuation. He described prices formed in the secondary market as “a noisy signal,” arguing that they merely reflect temporarily overheated investor sentiment.

Risks of Indirect SPV Investment, Including High Fees, Lack of Voting Rights and Potential Invalidation

Concerns are also mounting because a substantial portion of these private-market transactions takes the form of indirect investments through SPVs. An SPV is an investment vehicle that pools capital from multiple investors into a single legal entity to acquire shares in a privately held company. Investors gain exposure to movements in Anthropic’s share price by purchasing interests in the SPV, but the SPV itself becomes the legal shareholder. Only the SPV is listed on Anthropic’s shareholder register, while individual investors receive dividends or sale proceeds under contracts signed with the SPV manager. The structure is widely used in late-stage startup transactions because it allows issuers to limit the proliferation of shareholders while enabling smaller investors to gain access to scarce private-company shares.

The rights investors obtain, however, are determined by their contracts with the SPV. Voting rights, authority to dispose of shares and the timing of equity distributions following an IPO are all controlled by the SPV manager. In structures involving multiple layers of SPVs, fees can accumulate and ownership rights can become increasingly complex, making it difficult to assume that private-market prices will translate directly into actual investor returns. Issuer approval is another critical variable. Anthropic recently reiterated that it would not recognize share transfers or transactions involving share-related rights unless they had received board approval. The U.S. Securities and Exchange Commission likewise notes that transfers of private-company shares may be restricted under issuance agreements and securities laws. This means investors may be left holding only contractual rights if a transaction is completed without the issuer’s approval.

That does not mean the technological infrastructure needed to facilitate private-share transactions is entirely absent. Last month, Citigroup unveiled a service that issues and directly custodies tokenized depositary receipts backed by shares in private companies. Under the structure, Citi holds the underlying shares and issues corresponding digital depositary receipts, which investors can trade through existing wealth-management platforms or digital-market infrastructure. Recording ownership transfers on a blockchain can shorten settlement and title-transfer processes while creating scope for clearer ownership records and fee structures than conventional SPV transactions involving multiple intermediaries. Because such transactions do not take place on an official exchange, however, institutional challenges involving issuer approval, investor-rights protection and securities regulation remain unresolved.

Rising Loss Burden as Demand Shifts Toward Chinese AI Models and Competition for Customers Intensifies

Even if the transactional infrastructure is improved, whether Anthropic’s private-market valuation can be supported by its actual business performance remains a separate question. Bloomberg’s analysis of developer usage data from AI model aggregator OpenRouter found that U.S. AI companies’ share of token processing had fallen to 20% by the end of last month. Chinese AI companies, meanwhile, expanded their share to 80%, signaling a shift in market leadership. Given that U.S. providers held a 74% share in the same month a year earlier, developers’ criteria for selecting models appear to have changed dramatically in just one year.

Anthropic has not been immune to this shift. According to OpenRouter data, the Claude model family’s share of token usage fell from about 29% last year to 13.3% at the end of last month. Over the same period, Chinese AI models including DeepSeek, Alibaba’s Qwen, Zhipu AI’s GLM and Moonshot AI’s Kimi rapidly increased their usage and strengthened their presence in the developer ecosystem. While orders to acquire Anthropic shares are piling up in the private market, demand in the model market itself is dispersing toward Chinese providers competing aggressively on price.

As customer attrition becomes increasingly visible, Anthropic is substantially expanding its provision of free tokens to promising startups. According to The Wall Street Journal, Anthropic recently raised the maximum amount of free application programming interface credits offered to companies participating in Y Combinator, the world’s largest startup accelerator, from $30,000 to $500,000. No equity-related conditions were attached. The company expanded its support package immediately after OpenAI offered credits worth as much as $2 million to 169 Y Combinator startups. The strategy appears designed to encourage startups to adopt Claude from the earliest stages of development and thereby secure them as long-term customers.

The importance of capturing startup customers early stems from the development architecture of generative AI services. Once a company builds an application around a particular model, its prompts, data-processing systems, agent configurations and internal evaluation criteria are all tailored to that model. Switching to another model later requires the company to recalibrate its existing systems and validate their performance and reliability, generating substantial transition costs. Anthropic aims to integrate Claude deeply into developers’ services during the free-use period and convert them into paying customers by applying standard API rates once their credits are exhausted. Anthropic’s startup support program is structured so that standard pricing takes effect automatically, without a separate migration process, after the free credits expire.

Expanding free token allocations, however, may increase usage while simultaneously weighing on near-term profitability. Because top-performing models consume vast computing resources during inference, every increase in free calls raises the semiconductor and data-center costs Anthropic must absorb. According to U.S. semiconductor research firm SemiAnalysis, some of the token capacity available to subscribers paying $200 per month would cost as much as $14,000 if priced at standard API rates. With Chinese providers competing through lower token prices and open models, Anthropic faces the challenge of striking a sustainable balance between price competitiveness and profitability.