Anthropic Surpasses OpenAI in Revenue and Market Share

Article by Xiao Bing, Shenchao TechFlow

This could be the most thrilling AI revenge drama of the year.

The former giant in large models, OpenAI, has lost its dominant position. Anthropic, founded by a former OpenAI employee and six others, is eroding OpenAI’s leadership across metrics such as revenue, valuation, and enterprise market share.

The temperature difference in the secondary market is most apparent. Ken Smythe, founder of Next Round Capital, faces a pile of $600 million in applications to sell existing OpenAI shares, with six hedge funds and venture capital firms lined up to offload their holdings. This time last year, these shares would have been snapped up within days. Now? He’s scoured through hundreds of institutional investors—and still can’t find a single buyer.

At the same time, $2 billion in cash is waiting in line to purchase Anthropic shares.

On the on-chain derivatives platform Ventuals, Anthropic’s implied valuation briefly surpassed OpenAI’s at $863.6 billion versus $846.1 billion.

More telling is Goldman Sachs’ attitude: it no longer takes a performance fee when selling old shares of OpenAI to high-net-worth clients, effectively discounting them to push sales. But when selling stakes in Anthropic, it still charges a 15% to 20% carry—take it or leave it.

How has Anthropic, founded just five years ago, gradually surpassed its former employer, OpenAI?

The story begins in 2020.

That year, Dario Amodei was still the Vice President of Research at OpenAI, involved in building GPT-2 and GPT-3. Many versions circulated in Silicon Valley about why he left—some said it was because Microsoft’s investment changed OpenAI’s nature, while others cited fundamental disagreements over safety principles.

Dario himself discussed this issue on Lex Fridman’s podcast, roughly saying: Arguing over someone else’s vision is extremely inefficient; instead of trying to change others, take the people you trust and go do what you want to do.

In 2021, Dario left with his sister Daniela and five other core researchers from OpenAI to found Anthropic.

Sam Altman probably didn’t pay much attention at the time. Back then, OpenAI was at its peak, and a few researchers leaving wasn’t a big deal.

But during the height of the "boardroom coup" in November 2023, OpenAI’s board approached Dario, asking if he would be willing to replace Altman as CEO and merge the two companies.

Dario declined—he didn’t want the CEO position at OpenAI; he wanted to build an entirely new system from scratch according to his own logic.

From 2021 to 2024, Anthropic appeared almost invisible to the outside world.

When ChatGPT exploded globally at the end of 2022, Claude was still in internal testing. The Anthropic team felt its safety standards weren’t met and weren’t in a rush to launch. While competitors were scrambling to acquire users and dominate headlines, Dario and his team were focused on perfecting a training method called "Constitutional AI," which enables the model to self-regulate according to a predefined set of constitutional principles.

At the time, many felt Anthropic was being overly cautious—the market window is so narrow, if you don’t act, someone else will.

But looking back now, Anthropic made an extremely important choice during this “invisible” period: from day one, it focused its efforts on APIs and enterprise customers, almost entirely neglecting consumer-facing product promotion.

When Claude first launched in 2023, its popularity among end users was vastly inferior to ChatGPT’s, and most ordinary users weren’t even aware it existed.

Dario’s logic is roughly this: consumer attention comes quickly and fades just as fast; signed business contracts are what truly represent real money.

This assessment seemed conservative at the time but was proven correct by 2026. Of course, both narratives—that Anthropic deliberately chose the enterprise path or was forced into B2B due to inability to compete with ChatGPT in the consumer market—may contain elements of truth.

By early 2025, Anthropic’s annualized revenue had quietly climbed to $1 billion, a figure that attracted little attention at the time, given that OpenAI was already in the tens of billions—no one anticipated what was about to happen.

Numbers speak for themselves.

Anthropic's annualized recurring revenue (ARR): $1 billion on January 2025, $9 billion by year-end, $14 billion in February 2026, $19 billion in March, and surpassed $30 billion in early April.

OpenAI同期：2025年约130亿，到2026年4月约250亿。

Anthropic grew 30 times in 15 months, going from an order of magnitude behind OpenAI to surpassing it by 20%. OpenAI’s own growth has also been substantial, but when compared to Anthropic’s, it becomes a contrast between steady growth and exponential explosion.

The biggest structural difference here is that over 80% of OpenAI’s revenue comes from consumer subscriptions to ChatGPT. While 900 million weekly active users sound impressive, the paid conversion rate is only around 5%, meaning the remaining 95% are using the computing power for free.

Anthropic, on the other hand, generates 80% of its revenue from enterprise customers and API calls.

Corporate income and consumer income are entirely different categories.

Enterprise contracts are difficult to switch once signed, involve high switching costs with prolonged use, result in high renewal rates, and see increasing amounts year over year.

Consumers can cancel their subscriptions at any time, and a new release can cause a wave of churn.

In trading terms, one is a long-duration asset, and the other is a short-duration asset.

Let’s look at a few specific figures. By April 2026, Anthropic had surpassed 1,000 enterprise customers with annual fees exceeding $1 million, doubling in just two months. Eight of the Fortune 10 companies use Claude. In the most critical赛道 of code generation, Claude captured 42% to 54% of the global market share, while OpenAI held only 21%. Ramp’s enterprise spending data shows Anthropic’s share of corporate AI expenditures surged from 10% at the beginning of 2025 to over 65% by February 2026.

Do these numbers mean OpenAI is done? Not necessarily. But they do highlight one thing: the perceived first-mover advantages—brand, user base, ecosystem—that everyone assumed were untouchable a year ago have had little impact in the enterprise market; enterprise purchasing decisions follow a different logic.

The catalyst for Anthropic's revenue explosion is a product called Claude Code.

Launched in May 2025, it reached an annualized revenue of $1 billion by November, and surpassed $2.5 billion by February 2026. A product growing from zero to $2.5 billion in nine months.

Reviewing SaaS industry history, no faster case has been found. Cursor reached $500 million in over a year; GitHub Copilot took even longer.

What exactly distinguishes Claude Code from previous AI programming tools?

Simply put, GitHub Copilot helps you complete the next line as you write code—you’re still the one doing the work. Claude Code, on the other hand, is when you tell it, “I need a user login module,” and it writes the code, creates files, runs tests, and submits changes—all on its own—while you just watch.

This difference may sound like merely a matter of degree, but it’s actually a paradigm shift—one is a “better tool,” the other is a “colleague who replaces your work.”

Anthropic's internal data better illustrates the point.

Boris Cherny, head of Claude Code, said he now writes 100% of his daily code using Claude Code, and the entire engineering team generates 70% to 90% of its code with it. Ninety percent of Claude Code’s own codebase was written by itself.

In February 2026, Pragmatic Engineer conducted a survey of 15,000 developers, and Claude Code ranked first among the most popular AI coding tools. By early 2026, 4% of public commits on GitHub were generated by Claude Code, with expectations to exceed 20% by year-end.

The success of Claude Code reveals a reality many in the AI industry are reluctant to acknowledge: the commercial ceiling for chatbots as a category may be inherently low. What truly drives enterprises to invest heavily are AI tools that integrate into workflows and replace specific job functions.

ChatGPT opened the door to AI, but whether you turn left or right after entering determines who can turn users into revenue. Anthropic turned right, entering the realm of enterprise production.

In January 2026, Anthropic released Cowork, extending the same concept from developers to all white-collar roles. Something built by four engineers in ten days had most of its code written by Claude Code itself.

Since Claude Cowork's launch, approximately $2 trillion in market value has been erased from the global SaaS sector.

The products and strategies are obvious differences, but the real key lies in the people.

First, look at OpenAI: from 2024 to 2025, the company experienced a systematic exodus of senior executives.

Co-founder and Chief Scientist Ilya Sutskever left to found Safe Superintelligence. CTO Mira Murati left to found Thinking Machines Lab. Co-founders John Schulman and Jan Leike, head of the Superalignment team, went to Anthropic.

Chief Research Officer Bob McGrew left, Vice President of Research Barret Zoph left, and co-founder and president Greg Brockman went on an extended leave. In the summer of 2025, at least seven researchers were poached by Meta’s Superintelligence Lab.

Of OpenAI’s original 11 co-founders, only Sam Altman and researcher Wojciech Zaremba remained full-time by the end of 2025. A former employee told Fortune: “OpenAI without Ilya is a different company; OpenAI without Greg is a very different company.”

Anthropic presents a different picture.

All seven co-founders—Dario Amodei, Daniela Amodei, Jared Kaplan, Jack Clark, Sam McCandlish, Ben Mann, and Tom Brown—are still with the company, and there has not been a single public executive-level departure in its five-year history.

The contrast is so striking that it’s worth asking: What exactly did Anthropic do to keep people around?

In early 2026, Forbes estimated that each of the seven co-founders held approximately 1.8% of the shares, with minimal variation. At a $380 billion valuation, each person’s stake was worth about $6.8 billion. This nearly equal equity structure differs sharply from Silicon Valley norms, where the CEO typically holds the largest share, with other founders receiving progressively smaller portions. Equal ownership at least eliminates one of the most common sources of friction within founding teams: perceptions of unfairness.

Equity is merely superficial; what deserves greater attention is Dario Amodei’s time investment in management.

He said on the Dwarkesh Podcast that he spends about one-third to 40% of his time “ensuring Anthropic’s culture is good.” For a CEO of an AI company, this proportion is unusually high. As the company has grown to 2,500 employees, he can no longer be involved in every technical and product decision, so he has chosen to focus his energy on more “leverageable” efforts: keeping everyone aligned.

How exactly do I do it?

He holds a company-wide meeting every two weeks, internally called “DVQ”—Dario Vision Quest. The name was coined by employees; Dario once considered changing it because it sounded like a hallucinogenic experience. For each meeting, he prepares a three- to four-page document and delivers a one-hour presentation to the entire company, covering topics ranging from product strategy to geopolitics and major trends in the AI industry. Most employees attend in person or remotely.

On a more everyday level, Anthropic has a Slack culture of “notebook channels.” Every employee, including Dario himself, maintains a public Slack channel where they regularly share their thoughts, work progress, and even their confusions.

Growth lead Amol Avasare compared it on Lenny's Podcast to an “internal Twitter feed,” where you can jump into any channel—whether from the research team or any other department—to see what they’re thinking. Dario encourages employees to “argue directly with him.”

In his Fortune interview, he said: “My goal is to build a reputation for telling companies the truth—calling out problems directly and avoiding ‘corpo speak’—that defensive, politically correct corporate language. If you hire people you trust, you can communicate without any filters.”

This “anti-public relations” internal communication style stands in stark contrast to OpenAI, where during its board crisis at the end of 2023, internal communication was so fragmented that even the CTO was unsure of what had happened.

Anthropic’s cultural filtering begins at the hiring stage. Every candidate, regardless of the position they’re applying for, must go through a standardized “cultural interview.” Only employees who have been with the company for at least 30 days and completed multi-stage cultural training are eligible to serve as cultural interviewers. The logic is this: cultural transmission is too important to entrust to someone who hasn’t yet fully understood what the company’s culture truly is.

According to reports, one question in the cultural interview was: If Anthropic decides not to release the model due to inability to guarantee security, and your equity becomes worthless as a result, would you be willing to accept that?

This is not a rhetorical question; candidates who cannot answer this question correctly will not be hired, regardless of their technical skills.

One additional detail: At Anthropic, all technical roles—from new hires to founding executives—use the same title: “Member of Technical Staff.” There are no hierarchical distinctions such as “Senior,” “Principal,” or “Distinguished.” Employees internally refer to one another as “ants” (derived from the abbreviation of Anthropic).

The company even hired a full-time philosopher, Amanda Askell, whose job is to shape Claude’s moral reasoning framework. She told Time: “Sometimes it feels like you have a 6-year-old child, and you’re teaching it what kindness is—but by the time it’s 15, it will be smarter than you in every way.”

Daniela Amodei's role in this system is often underestimated.

Dario is the technical visionary and external spokesperson, while Daniela oversees execution, culture, talent, and operational infrastructure. Reports indicate that the executive teams for research, product, sales, and operations all report directly to her. She has a clear preference in hiring: individuals who are strong communicators, emotionally intelligent, kind, curious, and eager to help others. In an industry dominated by technical founders, this emphasis on “soft skills” is uncommon.

All seven founders of Anthropic have pledged to donate 80% of their wealth, and nearly 30 Anthropic employees have signed up to attend the 2026 EA (Effective Altruism) conference in San Francisco—more than double the combined attendance from OpenAI, Google DeepMind, xAI, and Meta’s superintelligence lab.

The core asset of an AI company is the human mind. Code can be copied, computing power can be purchased, but a researcher’s intuition and judgment cannot be taken away.

When your chief scientist, CTO, and chief research officer leave one after another over a two-year period, what you lose cannot be measured by funding amount. Anthropic’s stability in talent may be its most difficult-to-replicate advantage.

Every victory is a victory of values.

At this point, it’s only fair to say a few words in defense of OpenAI.

Anthropic's revenue has surpassed OpenAI's, and sentiment in the secondary market is shifting. But OpenAI has not collapsed—it just completed a $122 billion funding round, with participants including Amazon, NVIDIA, SoftBank, and Microsoft. ChatGPT still has 900 million weekly active users.

In consumers' minds, "AI" and "ChatGPT" are nearly synonymous, but OpenAI does have some structural issues, and these problems erupted simultaneously in 2026.

Financial pressure is the most direct.

OpenAI is projected to lose $14 billion in 2026. Cumulative losses from 2023 to 2028 could reach as high as $44 billion. HSBC analysts believe profitability will not arrive before 2030. The Wall Street Journal estimates that by 2030, OpenAI’s annual training costs will reach $125 billion, while Anthropic’s will be around $30 billion during the same period. Given that both companies are training cutting-edge models, this fourfold cost difference requires explanation. Part of the reason is OpenAI’s more aggressive investment in computing infrastructure, and part may stem from efficiency issues. Capital markets are clearly concerned about this gap: Anthropic is expected to achieve positive cash flow by 2027, while OpenAI has pushed its break-even point to 2030.

There have also been some issues at the product level.

Sora shut down in March 2026. The video generation tool reportedly cost $15 million per day to operate, while generating only $2.1 million in total revenue. The shutdown also damaged OpenAI’s partnership with Disney, causing a rumored $1 billion investment deal to fall through. OpenAI’s new Head of AGI Deployment, Fidji Simo, told employees bluntly that the company “cannot afford to be distracted by side projects.”

Then there’s advertising. In February 2026, OpenAI introduced ads into the free and Go versions of ChatGPT. While this in itself isn’t major news—many products operate on ad-based models—it stands out given OpenAI’s history, because Sam Altman had explicitly stated in 2024 that advertising was a “last resort” and that combining ads with AI made him “uniquely uneasy.” Just 15 months later, that “uniquely uneasy” stance gave way to an official rollout. With only 5% of its 900 million users paying, this number forced his hand.

Corporate governance was even more complex. The restructuring from a nonprofit to a for-profit entity took nearly a year. It involved Elon Musk’s lawsuit, a joint open letter of opposition from former employees, a public letter signed by Nobel laureates, and investigations by the attorneys general of California and Delaware. The restructuring was finally completed in October 2025, with the nonprofit foundation retaining 26% of the shares and control. Critics argue that this arrangement is merely symbolic.

Individually, these issues aren’t fatal. Together, they paint a troubling picture: a company that once led the industry’s imagination is now dominated by headlines of internal governance battles, product shutdowns, and advertisements.

Anthropic’s momentum is truly strong—revenue has surpassed competitors, it’s being embraced by secondary markets, and the Pentagon incident has delivered global free PR. But there’s one thing to remember: if you had asked any industry analyst at the end of 2023 whether OpenAI could be surpassed, 99% would have said it was impossible. The fact that consensus shifted so rapidly should make us cautious about today’s new consensus.

A few high-probability certainties: Anthropic has made the right move by focusing on enterprise; its revenue model, with 80% coming from enterprise, is significantly healthier than ChatGPT’s consumer-focused approach, as clearly supported by financial data. Claude Code represents a genuine product breakthrough—achieving $2.5 billion in annual recurring revenue in just nine months speaks volumes.

But there are just as many uncertainties. OpenAI boasts 900 million weekly active users and the world’s strongest AI brand recognition. If it finds an effective consumer monetization strategy—even just increasing its paid conversion rate from 5% to 10%—the entire narrative would need to be rewritten. One characteristic of the AI industry that makes forecasting so risky is that a single major model breakthrough could completely reset the landscape.

The capital flows in the secondary market do indicate a certain direction, but the secondary market has also favored WeWork.

A more cautious conclusion is that Anthropic’s approach has been validated in the first round of AI commercialization, while OpenAI’s approach is now being questioned. But it’s still too early to say who has won—this battle is only halfway through.

When Dario Amodei left OpenAI in 2021 with six others, few could have anticipated the situation we see today. A researcher出身于安全领域的他，在一个所有人都在竞相提速的行业中，凭借更少的资金和更强的自我约束，将昔日雇主逼至不得不向投资者撰写备忘录以解释其竞争力的境地。

The most interesting thing about this story is that it still doesn't have an ending.

Disclaimer: This article does not constitute investment advice. The valuation data mentioned herein is sourced from secondary market trading platforms and public reports, and may differ from actual trading prices.