Anthropic Overtakes OpenAI: What the Numbers Mean for You

manager magazin summed it up: Anthropic is growing faster than any company in economic history and has surpassed OpenAI on many key metrics. That’s true. Still, it’s not the number that matters most to a business.

We use language models in our day-to-day operations—on our own hardware and through the major providers. So here are the numbers, what they reveal, and most importantly: what’s missing from the success story.

What Actually Happened

On May 28, 2026, Anthropic announced a funding round of $65 billion at a post-money valuation of $965 billion. The same press release contains the key sentence: run-rate revenue had crossed the $47 billion mark earlier that month.

Two months earlier, on March 31, OpenAI had closed a round of $122 billion at a valuation of $852 billion—and cited monthly revenue of around $2 billion. Annualized, that comes to about $24 billion.

The discrepancy is real. But it doesn’t tell the whole story.

The shift in the enterprise market is more pronounced than the revenue gap: Anthropic accounts for 40 percent of spending on language model interfaces, OpenAI for 27 percent, and Google for 21 percent. Three years ago, OpenAI held 50 percent.

Why: It’s All About Coding

The reason lies not in the valuation, but in a single application category. In the coding market, Anthropic holds an estimated 54 percent share compared to 21 percent for OpenAI—half a year ago, that figure was 42. The driver is Claude Code and the suite of tools surrounding it.

This aligns with what we’re seeing. For tasks consisting of a single answer, the top models are neck and neck. The difference emerges in long, unsupervised toolchains—when an agent must maintain direction, detect errors, and correct them over twenty steps. In these cases quality does not decline linearly but abruptly — and that is precisely where Anthropic has not merely held its lead but extended it.

More on this: Claude 5th Generation in Detail · GPT-5 from OpenAI · Agentic Coding in Practice

The Structure Behind It

The most revealing difference isn’t the size, but where the money comes from.

These are two different businesses. OpenAI sells a little to a lot of people, while Anthropic sells a lot to relatively few companies. Both can work; the metrics are simply difficult to compare. If you compare them side by side, you’re comparing a subscription business with an infrastructure business.

For you as a business, this means: The provider whose business depends on corporate customers has a structural interest in stability, contract compliance, and long model lifecycles. The provider whose business depends on end customers has an advantage in reach. This isn’t a moral judgment, but a statement about whose incentives align better with yours.

What the Success Story Doesn’t Say

This is where it gets really interesting for a business that’s planning ahead.

The computing costs are enormous and locked in for the long term. Anthropic has secured the entire capacity of a data center and is paying $1.25 billion per month through May 2029 for it. That’s not an amount you can easily cut back on in the short term if demand weakens.

Profitability is new and slim. The gross margin came from minus 94 percent in 2024 and sits somewhere between 40 and 60 percent today depending on the source; the target is 77 percent by 2028. For the second quarter of 2026, the company has forecast its first profitable quarter—with the explicit caveat that this may not hold true for the rest of the year due to upcoming computing costs.

Demand is concentrated. Roughly a third of conversations on Claude.ai come from a single occupational group — computer and mathematical occupations; on the API the share is considerably higher. That is both a strong position and a concentration risk. A better coding model from the competition would not hit a peripheral area, but the core.

The market is shifting toward cost-efficiency. The phase in which customers burned tokens just to see what was possible is over; efficiency is becoming a key selection criterion. This is good for customers but puts pressure on precisely the revenue streams that underpin these valuations.

And above all else is the interdependence: Major technology conglomerates hold stakes in both companies and are simultaneously their suppliers and customers. This makes the reported figures harder to interpret than they appear.

What a Business Should Take Away

Three things—and none of them is “switch providers.”

First: Choose based on the task, not market share. A 40 percent market share says nothing about whether the model correctly reads your delivery slips. For coding agents and long chains, the recommendation is currently clear-cut. For classification, extraction, and bulk processing, the better question is whether it even needs to be a top-of-the-line model—the price difference is a factor of seventy. We’ve broken this down in AI Costs per Task.

Second: Build against an interface, not against a provider. The landscape shifts noticeably every six to eight weeks. If you tie your application to a single provider, every switch costs you project time. A gateway in front of it turns that same switch into a single line of configuration—that’s exactly how we do it; see OmniRoute vs. LiteLLM.

Third: Expect prices to fluctuate in both directions. Prices per token have fallen over the years, and this trend is continuing. At the same time, both companies are sitting on obligations that must be paid. Anyone basing a calculation today on a single provider’s list price should be aware of how expensive a switch would be. See AI Price Decline 2026.

The Part Neither Set of Numbers Covers

Both providers are based in the U.S. For data that is not allowed to leave the premises, the question of market share is secondary to the question of where it is processed.

In practice, the answer is usually not an either/or decision: the top-of-the-line model for complex cases, and a model with open weights running on our own hardware for everything that occurs regularly and in large volumes or involves personal data. We operate both side by side. See also: Open vs. Closed Models · Sovereign AI and Data Sovereignty · EU AI Act and Local AI

Conclusion

Anthropic has overtaken OpenAI in the enterprise market, and the reason is a single application category in which they have recently extended their lead. This is noteworthy and relevant when selecting a coding agent.

For everything else, the following applies: The numbers describe a race between two providers, not a model’s suitability for your specific task. The more robust conclusion is unspectacular—keep the application interchangeable, choose based on the task at hand, and maintain an in-house option for sensitive data.

A structured overview of the entire field can be found in AI Models Compared 2026.

Sources

• Anthropic, Series H — funding round, valuation, run-rate revenue (2026-05-28)
• CNBC, OpenAI closes funding round at an $852 billion valuation (2026-03-31)
• Menlo Ventures, The State of Generative AI in the Enterprise — enterprise spend and coding market share (December 2025)
• TechCrunch, Anthropic will pay $1.25 billion per month for compute (2026-05-20)
• CNBC, Anthropic set to hit $10.9 billion in revenue during second quarter (2026-05-20)
• Anthropic, Economic Index — January 2026 — usage distribution by occupation
• CNBC, Big Tech's Anthropic and OpenAI stakes are distorting the corporate earnings picture (2026-08-03)