虎嗅

Anthropic is mass-producing its own Zhang Leis and Xu Xins

原文:Anthropic正在批量制造自己的张磊徐新

Summary of Key Points

In 2023, Spark Capital invested $75 million in Anthropic, an AI company that had virtually no revenue or mature products at the time. Today, this investment has nearly returned a hundredfold (about $7 billion). By focusing on the corporate market rather than competing with OpenAI for consumer users, Anthropic has become a player at the infrastructure level of AI technology, and its valuation once surpassed that of OpenAI ($96.5 billion vs $85.2 billion). Different investors—VC firms, tech giants, and even the bankrupt FTX—have drawn different conclusions from this investment: VCs bet on growth, giants on building an ecosystem, while FTX made a lucky discovery. However, book value does not equate to actual profit. Anthropic’s upcoming IPO will be the test that determines its true worth.

I. What Did Spark Capital Bet On with Its $75 Million? – Not Short-Term Traffic, but on AI as an Essential Infrastructure

At the beginning of 2023, when ChatGPT was just becoming popular and OpenAI was the leader in the AI world, backed by Microsoft with a strong brand and technology, Anthropic only had a top-tier team and the Claude model, yet it had not proven its profitability. Yasmin Razavi from Spark Capital took the risk of investing $75 million, believing that large-scale AI models are not just ordinary software but the “infrastructure” of the future technology industry—essential resources that businesses cannot do without, much like water, electricity, and coal.

She argued that even if OpenAI was ahead, the infrastructure market could accommodate multiple players, just as the internet era had both Alibaba and JD.com, or the mobile era had both WeChat and TikTok. Therefore, she focused on whether Anthropic could become part of the next generation of essential technologies, rather than its current revenue situation. Later events proved her right: Anthropic established a foothold with corporate clients, with investments from Amazon and Google, validating her judgment.

II. How Did Anthropic Reverse the Situation? – Not Competing with OpenAI for Consumers, but Focusing on Corporate Needs

While OpenAI became globally popular with ChatGPT among consumers, Anthropic chose a different approach: it focused on the corporate market, developing AI tools that meet businesses’ essential needs.

For example, its Claude model was used for coding, managing company knowledge, and conducting research analyses—areas where companies are willing to spend money. By October 2025, it had 300,000 corporate clients, with the number of high-paying customers (over $100,000 per year) increasing sevenfold. It also secured partnerships with Amazon and Google: Amazon invested $4 billion to use Anthropic’s services on its AWS platform, and Google invested $2 billion in TPU chips for AI training. These investments not only provided funding but also solved the critical issue of computational power, enabling Anthropic to continuously improve its models.

As a result, Anthropic went from being a follower of OpenAI to a leader in enterprise AI, with a valuation that surpassed OpenAI’s.

III. Different Investors’ Approaches: VCs Bet on Growth, Giants on Building Ecosystems, FTX Made a Lucky Find

Anthropics rise has benefited various investors, but their motivations varied:

  • VCs (Venture Capital): They bet on companies growing from small startups to giants.

Spark Capital led the Series C round and has now earned nearly a hundredfold on its investment. Menlo Ventures initially hesitated but later made the largest investment in the company’s history during the Series D round, earning $14 billion. Their strategy is similar to investing in companies like JD.com or Meituan during the early internet era: acquiring shares at low prices in hopes of long-term growth.

  • Tech Giants (Amazon, Google): They focus on building ecosystems.

Amazon’s investment not only allows them to benefit from Anthropic’s growth but also gives them access to its AI services through AWS. Google, meanwhile, developed its own Gemini model while also investing in Anthropic and providing TPU chips. This strategy aims to both gain a share of the AI market and use Anthropic’s technology to strengthen its cloud business.

  • FTX’s Legacy: A Lucky Turn in a Changing Era

FTX invested $500 million in Anthropic in 2021 during the crypto boom. Although FTX later went bankrupt, its investment in Anthropic proved profitable as the AI industry surged. The FTX team sold its shares for $1.3 billion, reflecting how the AI revolution provided a last-minute opportunity despite the crypto bubble’s collapse.

IV. Book Value of $7 Billion Does Not Equal Real Profit: IPO Is the Test

Spark Capital’s $7 billion is just book value; to convert it into cash, Anthropic needs to go public (IPO) or be acquired.

Anthropic has already submitted its IPO documents and could list as early as this fall. The question remains: Can its current valuation of $96.5 billion be supported by actual performance?

Anthropic predicts its first profit in the second quarter of 2026 (revenue of $1.09 billion, profit of $559 million), with external analysts being even more optimistic (profit exceeding $1 billion by the third quarter). However, the market is becoming more cautious. AI stocks have risen excessively, and investors are now more concerned about the actual profitability of these companies. If Anthropic meets its targets after going public, its book value will turn into real cash; otherwise, its valuation may fall.

This is a common issue for all AI companies: capital markets offer high valuations, but ultimately, performance is what determines their true worth.

In Conclusion

Anthropic’s story serves as a textbook example for venture capital in the AI era. By betting on the future amidst uncertainty and choosing the right direction (infrastructure) and a differentiated approach (corporate market), one can achieve substantial returns. However, whether these investments will pay off depends on actual performance after the IPO. After all, no matter how large the bubble, it must be supported by real results.

(End of Article)