In just one year, the pattern of the AI industry has quietly changed.

OpenAI, which was once in the limelight in the investment circle, is now being fully overtaken by another AI company, Anthropic. Whether it is market share among enterprise customers, valuation in the secondary market, word-of-mouth in the venture capital circle, and popularity on social media, Anthropic has surpassed OpenAI in almost every aspect.

This trend was particularly evident at the recent HumanX AI conference in San Francisco. According to Business Insider, the venture capitalists and entrepreneurs present almost unanimously agreed that Anthropic is the most promising AI star in Silicon Valley at the moment.

The wind in Silicon Valley has changed: Anthropic has become the new favorite

This year's HumanX AI conference doubled in size from last year — about 6,700 people attended, and tickets exceeded $4,000 per person. But unlike last year's enthusiastic pursuit of OpenAI in Las Vegas, this year's Anthropic became the focus of the audience at the San Francisco venue.

Jared Quincy Davis, CEO of AI cloud platform Mithril, said: "Anthropic is gaining momentum. They focus on enterprise services, cutting-edge technology, and code generation, and actively avoid some consumer-grade applications, which is the right move. ”

Halfway through the conference, Anthropic also announced a new model, the Mythos. The company said that this model is too powerful and is temporarily closed to the public for cybersecurity reasons, and is only available to a small number of enterprises through a program called "Project Glasswing".

Tomasz Tunguz, a partner at venture capital firm Theory Ventures, commented: "Mythos is significant and the market is very excited. ”

On the other hand, OpenAI has almost no public platform. Participants' doubts about it mainly focused on two things: spending a lot of money to buy an online talk show TBPN, which is incomprehensible; The cooperation between CEO Sam Altman and the Pentagon has sparked controversy.

Andy Chen, a former partner at well-known venture capital firms Coatue and Kleiner Perkins, bluntly said: "Many people are dissatisfied with Altman's approach, and it is expected that OpenAI may lose a lot of talent in the future."

The wind direction of the secondary market has also changed

The capital market's reaction is the most direct - where the money flows shows who everyone is more optimistic about.

In early April this year, six institutional shareholders of OpenAI (including well-known venture capitalists and hedge funds) wanted to sell about $600 million of old OpenAI shares through the secondary trading platform Next Round Capital. But surprisingly, the platform contacted hundreds of potential buyers, but no one was willing to buy.

Ken Smythe, the founder of the platform, said helplessly: "We have hundreds of institutional resources, but we can't find one willing to take over. Instead, many buyers told us that they had $2 billion in cash in their hands and were waiting to buy Anthropic stock. ”

This trend of "abandoning OpenAI and chasing Anthropic" is equally evident on other equity trading platforms such as Augment and Hiive. Augment co-founder Adam Crawley said: "Everyone thinks that Anthropic's valuation will soon catch up with or even surpass OpenAI, so they want to get on the bus as soon as possible. ”

Valuation data also confirms this shift: OpenAI's valuation in the secondary market is about $765 billion, which is 90% off the previous round of financing (10% discount); Anthropic's valuation has reached $600 billion, more than 50% higher than the previous funding round; There are even rumors on social platforms that Anthropic's latest valuation in the private market has slightly outpaced OpenAI.

What is even more intriguing is the attitude of Wall Street investment banks: Morgan Stanley, Goldman Sachs, etc. began to actively promote OpenAI shares to high-net-worth customers, and did not charge any performance share - equivalent to "liquidation and selling"; However, for customers who want to invest in Anthropic, Goldman Sachs still charges a 15%–20% profit share as usual, indicating that they still consider it a "scarce high-quality asset".

｜B端市场：Anthropic已形成压倒性优势

收入数字背后，更深层的结构性变化正在企业市场发生。

1. 市场份额：Anthropic遥遥领先

在最重要的AI代码生成领域，Anthropic的Claude模型占全球42%–54%的份额，而OpenAI只有21%。在企业级AI智能体（agent） 市场，Anthropic占40%，OpenAI为27%。

2. New customers love Anthropic more

According to Ramp data, in March 2026, 65% of new companies purchasing AI services chose Anthropic, and only 32% chose OpenAI. By April 2026, Anthropic has more than 1,000 enterprise customers spending at least $1 million a year on it — a figure that has doubled in two months. Today, more than 80% of its revenue comes from API calls and enterprise customization services, indicating that it has truly reaped the dividends of the enterprise market.

3. The cost gap is huge and the profit prospects are clear

According to estimates by the Wall Street Journal, by 2030, OpenAI will burn $125 billion per year just by training models; And Anthropic only costs about $30 billion - less than a quarter of OpenAI. With high growth + low cost, Anthropic is expected to achieve positive cash flow (i.e., make more money than spend) by 2027; OpenAI still does not see a clear timeline for profit.

4. OpenAI's dilemma: The C-end is lively, but not profitable

OpenAI's only lead is the consumer market: ChatGPT has 900 million weekly active users. But more than 98% of them are free users, not only do not pay, but also consume a lot of computing power, bringing almost no income. In February this year, OpenAI tried to add ads to ChatGPT, which caused strong dissatisfaction and controversy among users.

OpenAI's counterattack: showing the computing power trump card, but it is difficult to hide the anxiety

In the face of increasing doubts from the outside world, OpenAI sent a confidential memo to shareholders this week, which was quickly leaked.

In this document, OpenAI publicly referred to Anthropic as its "biggest competitor" for the first time and tried to stabilize the military's morale with a core advantage - computing power infrastructure: by 2025, OpenAI will have 1.9 gigawatts (GW) of AI computing power; It is expected to exceed 10GW ("double-digit low") in 2026; By 2030, the goal is to reach about 30GW.

In contrast, OpenAI estimates: Anthropic will have only 1.4GW by the end of 2025;

By 2026 it will reach a maximum of 7–8GW. "Even by the most optimistic estimates, we are expanding significantly faster and the gap is widening," the document emphasized. ”

However, Anthropic is not without backhand - it has reached an agreement with Google and Broadcom to receive 5GW of next-generation TPU computing power support from 2027, and the shortcomings of computing power are expected to be alleviated in the future.

But what is really noteworthy is not the data itself, but the symbolism of this matter: a company that was once recognized as the absolute hegemon of the AI industry now has to write a letter to shareholders to explain that "we haven't lost yet", and even write the names of opponents into internal strategy documents - which in itself shows that OpenAI is feeling unprecedented pressure.

Bet on "hematopoietic ability", not "story bubble"

When Anthropic proved with 30 billion ARR that "companies are willing to pay for value" and OpenAI is still worried about the electricity bills of free users, we know that this race has moved from the laboratory to the board of directors.

For the primary market:

Prioritize the Anthropic ecosystem: its API-driven business model has verified PMF (product-market fit); Be cautious about OpenAI concept stocks: high valuations rely on IPO expectations, but the profit path is vague, and cash flow risks accumulate.

For the secondary market (if a future IPO):

Anthropic: High growth + high efficiency + strong B-end stickiness, suitable for long-term holding; OpenAI: There may be a narrative catalyst for computing power in the short term, but we need to be wary of the Davis double kill of "market capitalization and fundamentals divergence".

For the industrial chain:

Cloud vendors (AWS/Azure/GCP): Anthropic's multi-cloud strategy is beneficial to all three parties; Developer tools (e.g., Cursor, Replit): Deep integration with Claude will be standard; Network security: The Mythos model opens up a new track for AI attack and defense through Project Glasswing.