It seems that Sam Altman isn't having his best day. Between a 10,000-word New Yorker article questioning his integrity and the fact that Anthropic has just surpassed OpenAI's annual revenue, he's probably not sleeping much these days.

Here's where it gets interesting: Anthropic went from $1 billion in annual revenue at the beginning of 2024 to $30 billion now. OpenAI is at $25 billion. Yes, those numbers are ARR (recurring annual revenue), not cash in the bank, but Anthropic's calculation is almost identical to OpenAI's, so the comparison is valid.

What really catches my attention is the fundamental difference in how these two companies make money. OpenAI has 900 million weekly active users—ChatGPT is practically the largest consumer app in history. But here’s the problem: only 5-6% pay. The remaining 94% use everything for free. And yes, that’s a financial disaster. OpenAI is burning money massively by keeping ChatGPT basically free, and according to The Information, they expect losses of $14 billion in 2026.

Anthropic chose a completely different path. About 80% of its revenue comes from businesses. Two years ago, only 12 companies paid more than a million a year. Now there are over 1,000, and that number doubled in less than two months. Eight of the ten largest Fortune 500 companies are clients. The average revenue per active monthly user for Anthropic is $211. For OpenAI, it’s $25 per weekly active user. Even with different calculation methods, the difference is staggering.

This reflects two completely different mindsets. OpenAI thought like a consumer internet company: attract massively, monetize later. Facebook, Google, TikTok followed that logic. Anthropic thought like an enterprise software company: go straight to those who can pay. Salesforce, Oracle, SAP did the same.

OpenAI’s problem is that AI inference costs are extremely high. Those 900 million free users are not an asset; they’re a liability. They need to convert them before running out of money. Anthropic doesn’t have that problem because it started by selling directly to companies.

What’s fascinating is that Claude Code (a prototype that a TypeAI engineer wrote in five days) already generates $2.5 billion in annual revenue. 4% of all public commits on GitHub globally are from Claude Code, and that doubles every month. That’s how you see the monetization difference.

Now, about infrastructure: Anthropic just closed an agreement with Google and Broadcom to secure 3.5 GW of TPU power starting in 2027. Mizuho analysts estimate that Broadcom will receive $21 billion in AI revenue from Anthropic alone in 2026. Anthropic uses AWS, Google TPU, and NVIDIA, avoiding dependence on a single provider.

And then there’s the IPO topic. Anthropic is likely to go public in October 2026, according to The Information. Banks expect to raise over $60 billion, which would be the second-largest tech IPO in history after SpaceX. The implied valuation has already risen from $380 billion two months ago to $600 billion now, and in the secondary market, almost no one wants to sell Anthropic shares, but there are also no buyers for OpenAI’s shares valued at $600 million.

The money-burning numbers are wild: OpenAI expects to spend $121 billion on computing alone by 2028, but will lose $85 billion that year. Anthropic expects to reach positive cash flow in 2027.

Some analysts say Anthropic’s growth is slowing down, from 10x to 7x annually. Still, it remains surprising compared to OpenAI. But nothing is guaranteed: Chinese open-source models are gaining enterprise ground, and OpenAI continues to push in the international consumer market.

At the core, both are betting on time. OpenAI bets that inference costs will fall enough to turn those 900 million users into a profitable business. Anthropic bets that the enterprise market will keep paying before it saturates. Who will reach the goal first or crash first? We don’t know. But one thing is certain: it’s no longer about who has the best model, but who manages to survive with a sustainable business model. For now, Anthropic has found theirs. OpenAI is still searching.