Anthropic passes OpenAI in quarterly revenue, carried by enterprise and Claude Code

August 19, 2026. The Wall Street Journal reports that OpenAI grew second-quarter revenue 18% to $6.7 billion while its operating losses widened to $12.3 billion. The same day, Anthropic is reported to have more than doubled revenue to $11.6 billion and overtaken OpenAI for the first time, posting a small operating profit.

For anyone buying AI in the enterprise, that is a shift bigger than a ranking: the market is starting to reward enterprise and coding workloads, and to charge for consumer scale.

The score flips

The figures, sourced by the Wall Street Journal from internal documents at both companies, tell two opposite trajectories. OpenAI grew from $5.7 billion to $6.7 billion quarter over quarter — solid, but below what some investors expected — while its operating losses reached $12.3 billion. Anthropic, meanwhile, more than doubled revenue to $11.6 billion and booked a small operating profit.

The novelty is not the size, it is the order. A year ago, the idea that the company with the strongest consumer brand was no longer the revenue leader would have sounded implausible. It has now happened, and CNBC reads the crossover as a test: which companies and stocks benefit if enterprise AI spending keeps tilting toward Anthropic.

Where the money comes from

The useful detail is the source of the revenue. Anthropic has concentrated on enterprise and coding workloads, with Claude Code cited as one contributor. OpenAI carries the cost of a much broader consumer footprint — ChatGPT and mass distribution.

That is a difference of business model as much as of product. An enterprise workload, billed per developer or per seat by usage, produces margins and retention different from a consumer base that is expensive to serve at inference time. OpenAI told investors that growth accelerated in the third quarter, so one quarter does not settle the race — but it exposes the mechanics of it.

Claude Code deserves to be isolated in this story. It is Anthropic’s command-line coding agent, sold to development teams, and it is exactly the kind of product whose adoption is measured in paid seats rather than free users. If Anthropic’s revenue doubled while OpenAI’s losses deepened, it is partly because the coding agent plants itself in a workflow companies pay for willingly, where the consumer assistant monetizes with more friction.

What it changes for buyers

For a CISO or an SRE arbitrating an AI strategy, the flip has three concrete consequences.

First, model quality is no longer the only criterion. Revenue now rewards distribution, margins, developer loyalty, and the workflows companies actually pay for — not benchmark scores alone. Tooling decisions must therefore be made on integration into the development cycle, not on a leaderboard.

Second, code is the first category that pays. The fact that Claude Code is named outright as a contributor confirms what the development market already hinted: the coding assistant, wired into a repository and a CI pipeline, is the most directly monetizable AI workload today. It is also the most sensitive — a coding agent holds write access, which makes it an attack surface in its own right.

Third, vendor dependency is measured in money. When one actor concentrates its revenue on the enterprise and another subsidizes it through consumer scale, pricing and support trajectories diverge. An enterprise contract does not carry the same guarantees depending on whether it funds a profit or absorbs a loss.

A trajectory that reads over time

The crossover is not a quarterly accident: it extends a trajectory. Anthropic bet early on the enterprise and on code, with an offering billed by usage and wired into repositories, development environments, and pipelines. Claude Code, the command-line coding agent, is the product through which that bet monetizes: it plants itself in developers’ daily workflow, where renewal is a given and the purchase decision is immediate.

OpenAI took the opposite path. ChatGPT built a worldwide consumer brand, but that scale is paid for in inference costs and margins. The $12.3 billion in operating losses in the second quarter is the price of that footprint. The two companies sell models of comparable quality; what separates them now is the structure of their revenue, not the science of their models.

The numbers carry a scope nuance

A careful reader will note that revenues are not always comparable in scope. Specialized press points out that the two companies do not consolidate exactly the same activities, and that OpenAI told investors growth accelerated in the third quarter. The quarterly crossover does not close the race.

But the order of magnitude remains the headline: a company with no comparable consumer brand has passed in revenue the sector’s most heavily funded player. That is a signal about what the market actually pays for — enterprise and coding workloads — more than a final verdict on who “wins.” For a buyer, the lesson is operational: a vendor whose revenue comes from the workloads you run is a vendor whose roadmap will follow those workloads, not the reverse.

A signal that reaches beyond the two labs

The flip is not just about Anthropic and OpenAI. It reorders how the whole chain — customers, partners, infrastructure providers, integrators — prices AI. A market that rewards code and the enterprise is a market where integration, reliability, and support count as much as a model’s score. For teams building on top, that is good news: competition moves toward what they control — usage — rather than what they endure — the parameter race.

Over the longer term, the crossover also bends the incentives that shape the models themselves. A lab that earns its revenue from coding agents and enterprise workloads will steer its roadmap toward the behaviors those buyers value: determinism, tool reliability, auditability, and predictable pricing. A lab that earns from a consumer assistant will optimize for breadth and engagement. The models the market gets in two years are, to a real degree, a function of which customer base is paying for them today.

None of this makes the quarterly numbers destiny. OpenAI’s consumer footprint is a genuine asset — distribution at that scale is a moat, and it can convert users into paying workflows over time. Anthropic’s enterprise lead must survive the moment competitors bundle coding agents into their own suites. The durable lesson is narrower: revenue now concentrates where usage is paid for, and that is where the roadmap follows. For a buyer writing a multi-year AI contract today, that concentration is the number worth watching, more than any benchmark.

Verdict

If you buy AI for development, the signal is clear: the coding agent is becoming the economic center of gravity of the market, and Claude Code is the benchmark. Evaluate it on your own repository and your own CI, with guardrails on write access, rather than on public leaderboards.

If you arbitrate more broadly, remember the mechanics over the ranking: the market is ceasing to reward consumer scale and starting to reward usage-billed enterprise workloads. The competition no longer plays out only on models — it plays out on distribution, margins, and the workflows a company is willing to pay for.

References

• The Wall Street Journal — OpenAI’s Second-Quarter Sales Show Tepid Growth Compared With Anthropic, August 19, 2026
• CNBC — Stock winners and losers as Anthropic passes OpenAI as hottest AI upstart, August 19, 2026
• The Neuron — Everything That Happened in AI Today (Wednesday, August 19, 2026), August 20, 2026