In February 2026, at an AI summit in New Delhi, Indian Prime Minister Modi attempted to bring the assembled tech leaders together for a photo symbolizing unity. While others joined hands, the two individuals standing on either side of him—Sam Altman and Dario Amodei—merely touched elbows awkwardly before quickly withdrawing their arms. This subtle gesture almost perfectly encapsulates one of the most dramatic relationships in today's AI landscape: sharing a common origin yet growing apart; united in mission yet fiercely opposed.

To understand the present state of OpenAI and Anthropic, one must first look to their past.

01. Departing from the Same Office

In 2015, OpenAI was founded as a nonprofit research lab with the lofty mission of "developing artificial intelligence for the benefit of humanity." Elon Musk recruited Altman to assemble a team primarily to counter DeepMind, which had been acquired by Google. At the time, OpenAI was a utopia for idealists.

It was during this period that Dario Amodei joined OpenAI. A researcher with a background in physics and biophysics, he soon took charge of the model safety team and led the development of GPT-2 and GPT-3. Those were truly entrepreneurial days—he and Greg Brockman would stay up late training AI agents to solve game problems, discussing both technology and the future.

The rift emerged in late 2020. Microsoft's investment propelled OpenAI toward commercialization, while Dario Amodei's concerns grew over the neglect of AI safety. In his view, OpenAI's trajectory had deviated from its original commitment to "safely developing AGI." In 2021, he left with his sister Daniela and several core researchers to co-found Anthropic.

Thus, the most dramatic rivalry in the AI landscape officially began. One is the industry leader that spawned ChatGPT; the other, a challenger founded by former core team members. Both are headquartered in San Francisco, both are dedicated to developing large language models, and both claim to prioritize safety in advancing AGI—on the surface, they resemble two branches sprouting from the same tree. Yet today, five years later, these branches have grown in starkly different directions.

02. Two Revenue Structures, Two Ways of Life

If anything most vividly reflects the divergence between the two companies, it is where their money comes from.

OpenAI has pursued a mass-market subscription product route. ChatGPT has surpassed 800 million weekly active users, with over 65% of its revenue coming from subscriptions in the first quarter of 2026. This follows a typical consumer internet growth logic: relying on scale, branding, and user engagement. However, as reported by Reuters and other media outlets citing internal financial data, OpenAI is expected to incur losses of up to $14 billion in 2026—three times its 2025 losses—and does not anticipate profitability until 2029.

Anthropic has taken a different path. About 80% of its revenue comes from usage-based API services, with subscriptions contributing just 15%. Individual user subscriptions account for a mere 5%, compared to roughly 40% for OpenAI. In other words, Anthropic is a thoroughgoing B2B company.

The results of this divergence are striking. While OpenAI remained mired in losses in the third quarter of 2026, Anthropic was projected to exceed $1 billion in GAAP profit in the same quarter, crossing the breakeven line first, according to analyst reports from SemiAnalysis and others. By April 2026, Anthropic's annualized operating revenue had surpassed $47 billion, compared to OpenAI's approximately $25 billion for the same period, as disclosed by Anthropic and reported by multiple financial media outlets. Even more dramatically, in May 2026, Anthropic completed a $65 billion Series H funding round, reaching a post-money valuation of $965 billion—surpassing OpenAI's $852 billion valuation.

One generates profits, the other burns cash; one achieves profitability first, the other still struggles to turn a profit. The same track ( track can be translated as "track" or "field") has produced vastly different financial trajectories. But if you think this is merely a choice of business strategy, you underestimate the depth of this divergence.

03. Enterprise vs. Consumer: A Victory of Market Logic

Anthropic's overtaking of OpenAI hinges most critically on the enterprise sector.

In April 2026, Ramp's latest AI Index Report revealed that Anthropic's workplace adoption rate reached 34.4%, surpassing OpenAI's 32.3% for the first time. Over the past 12 months, Anthropic's enterprise paid adoption rate soared from 9% to 34.4%, nearly quadrupling, while OpenAI's enterprise adoption rate inched up from 32% to just 32.3%, virtually stagnant.

Even more shocking was new procurement—in roughly 70% of direct head-to-head competitions, Claude secured the contract, not ChatGPT. Among new AI purchases, 65% of enterprises chose Anthropic, while only 32% selected OpenAI.

The engine behind this comeback was a product: Claude Code. This command-line programming tool, publicly released in May 2025, has become Anthropic's fastest-growing product in history. By February 2026, its annualized revenue had surpassed $2.5 billion, capturing 54% of the AI programming tool market, according to internal Anthropic data cited by multiple tech media outlets. More visibly, 4% of all public GitHub code submissions were completed by Claude Code. Currently, Claude Code accounts for over 7% of GitHub's daily code submissions.

Why has programming capability become the decisive factor? The answer is simple: most enterprise AI procurement decisions originate not from CEOs making strategic declarations but from frontline engineers adopting tools themselves, recommending them to their teams, and ultimately driving company purchases. Anthropic has seized this "bottom-up" penetration logic, while OpenAI remains stuck in its habitual C-end brand warfare.

According to Anthropic and multiple third-party agencies, eight of the Fortune 10 have become paying customers, with over 1,000 enterprise clients spending more than $1 million annually. These enterprise clients' willingness to pay and loyalty far exceed those of C-end users. Anthropic's net dollar retention rate has reached 500%—meaning what? Its Q1 ARR was just $2 billion last year but surged to $30 billion this Q1, with $12 billion coming from repeat purchases by existing customers.

Meanwhile, OpenAI still maintains advantages in broader non-technical industries—but this edge is rapidly shrinking. Ramp's historical data reveals a pattern: early adopters' choices today become the entire market's choices in a few months.

04. Safety: A Slogan, Two Practices

If the divergence in business models represents tactical differences, the split in safety philosophies represents strategic ones—and it is precisely at this level that the two companies' transformation from collaborators to adversaries is best explained.

Anthropic's methodology is called "Constitutional AI." Simply put, it involves writing a "constitution" for the model, listing dozens of principles—"be helpful," "be honest," "be harmless"—and having the model continuously check its outputs against these principles during training. Anthropic hopes Claude acts benevolently out of identify with ( identify with can be translated as "alignment with" or "identification with") its values, not out of fear of punishment. In January 2026, Anthropic updated Claude's "constitution," explicitly prioritizing safety over usefulness.

OpenAI's approach is more flexible. It develops specific countermeasure guidelines for different threat categories, including identifying all potential abuse pathways before deployment and establishing protections for each. The difference was succinctly summarized by an industry observer: Anthropic insists on "control first, safety anchored," while OpenAI "accelerates full throttle."

This divergence came to the fore in 2026. Anthropic clashed with the U.S. government over Pentagon cooperation eligibility, while Sam Altman proudly announced that OpenAI had secured a classified project with the U.S. Department of Defense. Dario Amodei's internal criticisms of his former employer grew sharper, even comparing OpenAI and its rivals to tobacco companies that continue selling harmful products. Anthropic even rejected a $200 million deal due to safety requirements, resulting in its designation as a "supply chain risk" and suspension by relevant government agencies—a gap quickly filled by OpenAI.

More intriguingly, the two companies' hiring directions reflect their philosophies. OpenAI has approximately 670 open positions, concentrated in sovereign computing, public sector and defense collaboration, and trust-and-safety work for systemic risks. Anthropic lists over 400 roles, focusing on behavioral risks and CBRN (chemical, biological, radiological, nuclear) threat modeling. One is building "computing infrastructure," the other "trust infrastructure."

05. Two Paths, No Right or Wrong

Returning to the scene at the New Delhi summit—where Altman and Amodei did not shake hands but merely touched elbows—this detail stands out because it symbolizes more than personal animosity between two individuals. It represents a deeper split: over how quickly AI should be rolled out to the world, over how high a priority safety should take over commercial interests, and over how the creators of a technology that could rewrite humanity's fate should face the public.

Back when OpenAI was first founded, in a shared house on Delano Avenue in San Francisco, Greg Brockman and Dario Amodei debated this very issue. Brockman believed that since technology would change everyone's lives, it should be shared with everyone as soon as possible; Amodei argued that for a technology capable of rewriting order, it might not be appropriate to first unleash the most radical judgments on the public.

That debate changed nothing at the time. But looking back years later, it almost seems like a foreshadowing. The paths OpenAI and Anthropic later took were already faintly outlined in that moment.

Five years on, Anthropic has surpassed OpenAI in revenue, valuation, and enterprise adoption rates. But this does not imply right or wrong. OpenAI brought AI into millions of households with ChatGPT; Anthropic embedded AI into enterprises' core workflows with Claude Code. One expands breadth, the other deepens depth. One believes in "moving fast and breaking things," the other in "thinking clearly before acting."

On the AI track, there may never be a single correct answer. What is fascinating is that those who once believed AI would rewrite the world have ultimately practiced their shared faith in two starkly different ways. And those unbridgeable disagreements may be precisely what keeps this industry vibrant.