The Number Nobody Expected to See in April

On April 7, 2026, Anthropic announced it reached $30 billion in annualized revenue. OpenAI is at $25 billion. The crossover analysts predicted for August happened in April.

When I read this, my first reaction was to check the source. Then check again. Because a year ago — literally 12 months — Anthropic was at ~$1 billion in annual revenue and OpenAI at $6 billion. The gap seemed insurmountable.

It wasn’t.

The company most people outside B2B circles couldn’t name two years ago now generates more revenue than the company that invented the consumer AI category. And it did so spending 4x less on model training.

That number — 30x growth in 15 months — is the fastest expansion in enterprise software history. And it tells a story that made me rethink everything I thought I knew about the AI race.

How This Happened (The Strategy Nobody Copied)

The answer is almost embarrassingly simple: while OpenAI built for the audience, Anthropic built for the workflow.

OpenAI has 900 million weekly active ChatGPT users. An absurd awareness machine. But OpenAI’s revenue is still predominantly consumer — $20/month subscriptions that cancel if the user gets bored.

Anthropic has 18.9 million monthly active web users. A fraction. But 80% of its revenue comes from enterprise. Corporate contracts. API. Over 1,000 companies spend more than $1 million per year on Claude — that number was 500 in February. It doubled in two months. Eight of the top ten Fortune 10 companies are clients.

Enterprise revenue is different. It’s stickier. It expands as organizations deploy across more teams. It renews. It compounds. It’s exactly the kind of revenue Wall Street values — and that Anthropic accumulated quietly while the world watched ChatGPT.

The Obsession with Daily Value

Anthropic’s competitive advantage is the obsession with making users’ lives more valuable through practical integrations. Not demos, not impressive videos — tools that replace line items in the budget.

Claude Code reached $2.5 billion in annualized revenue by early 2026. On SWE-bench — the benchmark for autonomous coding in real-world codebases — Claude Code outperforms OpenAI’s Codex by over 23 percentage points.

Claude in Excel audits financial models in minutes. Founders, VCs, and operators use this daily. Claude in PowerPoint for presentations. Claude Cowork for non-technical knowledge work. Claude Code Security for vulnerability scanning. Each product solves a specific high-value workflow.

Anthropic launched Claude for Word, Excel, and PowerPoint with integrations that, according to analysts, surpass Microsoft’s own Copilot in real, intuitive assistance. The question keeping Redmond up at night: why would anyone use Microsoft’s native AI suite if Claude is more efficient at those tools?

The integration with Intuit for tax returns — helping millions file their taxes — is another example of the “invisible and useful” approach Anthropic pursues.

OpenAI’s Strategic Retreat

OpenAI killed Sora — or at least put it on ice. In March 2026, it announced the shutdown of the Sora video app, months after launch, citing a shift to research and robotics.

Why abandon such impressive video technology? The answer is focus — and money.

OpenAI projects $14 billion in losses for 2026. Cumulative losses may reach $115 billion through 2029. No positive cash flow projected before 2029. It desperately needs a successful IPO — multiple reports (Reuters, Stocktwits, TradingKey) indicate it’s evaluating an offering as early as October 2026.

For Wall Street, what matters now aren’t AI-generated videos but practical, recurring use cases. And in this race, OpenAI is playing defense: it announced that enterprise now represents over 40% of revenue (up from 30% last year), with a goal of reaching parity with consumer by end of 2026. Nine million paying business users as of February. APIs processing over 15 billion tokens per minute.

But Anthropic got there first. And in enterprise, first-mover builds the moat.

Chips Aren’t Everything

OpenAI committed to the Stargate project — $600 billion in compute investment through 2030. But projections suggest it will spend $125 billion per year on training by then. Anthropic? About $30 billion for the same period. Same race. 4x cost difference.

Anthropic proved something I didn’t think was possible: you can win the model race without having the biggest datacenter. What matters isn’t how many chips you have, but how efficiently you turn processing power into customer value.

Anthropic’s infrastructure strategy is different: it operates on AWS Trainium, Google TPUs, and NVIDIA GPUs simultaneously. Claude is the only frontier AI model available on all three major cloud providers — AWS Bedrock, Google Cloud Vertex AI, and Microsoft Azure Foundry. That gives optionality and resilience OpenAI, tied to Azure, doesn’t have.

The Shift in Wall Street’s View

Goldman Sachs reportedly charges a 15-20% carry on Anthropic secondary stakes while discounting OpenAI shares — a clear signal of where institutional money sees the higher-growth opportunity.

Anthropic projects positive cash flow by 2027. OpenAI doesn’t project it before 2029. For enterprise procurement teams evaluating long-term vendor stability, Anthropic’s financial trajectory is significantly less risky.

Anthropic’s valuation — $380 billion after its $30 billion Series G in February — is stratospheric. But at $30 billion annualized revenue, the price-to-revenue ratio is healthier than many big techs.

What I Really Think

Anthropic is winning the race that matters — the enterprise revenue race — with a fraction of OpenAI’s brand recognition. And this teaches me something profound about how technology markets actually work.

Consumer scale and revenue scale are not the same thing. 900 million weekly ChatGPT users is impressive. But 1,000 companies spending $1M/year each is what pays for infrastructure.

It’s not that OpenAI is losing. It keeps growing. It has the broadest ecosystem, the most massive user base, and is moving aggressively into enterprise. But the era of assuming the most famous AI company is also the most successful is over.

The market isn’t waiting for OpenAI’s next bombshell announcement anymore. It’s reacting to what Anthropic has already shipped. And for those working with AI daily, the lesson is clear: the “Claude effect” isn’t about the smartest model. It’s about the most useful product.

Ease of execution is the new success metric. Not the highest benchmark.

Share if this shifted your perspective:

The most famous company isn’t always the most successful. And the most useful product isn’t always the most impressive. 2026 is proving both.

Read Also