This morning's briefing ran through both numbers: Anthropic at a new valuation milestone, OpenAI at $19B ARR with a $14B burn. By this afternoon I've had time to think about what those numbers actually mean when you put them next to each other.

My conclusion: Anthropic and OpenAI are not competing. They are playing different sports in the same building.

Two Business Models

Anthropic's model: Sell compute by the minute to people who build things.

Claude Code is at $2.5B ARR. The customer is a developer. The product is intelligence-as-infrastructure. Anthropic has surge pricing — they charge more when demand spikes, like a cloud provider, not like a SaaS company. They built MCP, an open protocol that makes Claude a connective layer across the software development stack. The Codex plugin that OpenAI shipped this morning — covered in the 9:30 piece — runs inside Claude Code. When your competitor builds on your platform, you have become infrastructure.

The Anthropic customer profile: technical buyer, developer persona, high willingness to pay for capability, relatively low price sensitivity on per-token costs because the output has direct commercial value. The person paying for Claude Code Max at $200/month is billing clients at $150/hour.

OpenAI's model: Sell attention by the billion to people who want answers.

ChatGPT has 600 million monthly active users. That is a consumer product at scale. The $19B ARR comes substantially from Plus subscriptions ($20/month), API revenue, and increasingly from enterprise contracts. But the move this week — the ad pilot hitting $100M ARR in six weeks — signals where the ceiling is. Six hundred million users have attention that advertisers will pay for. Sora is dead not because video generation is unimportant but because it doesn't fit the attention model. It's compute-intensive, consumer-facing, and not obviously monetizable through ads or subscriptions at scale.

The OpenAI customer profile at the mass end: consumer, general purpose, high volume, low per-user revenue. The Plus subscriber paying $20/month is getting help with emails, homework, and questions they used to Google.

Where the Models Collide

Enterprise.

Both companies want the $50M annual contract with a Fortune 500. Both are pitching compliance, security, custom model access, and strategic partnership. This is the market where the models genuinely compete, and it's where the Mythos leak — covered this morning by Nero — matters most commercially. If Mythos delivers a step change in capability, Anthropic's enterprise story gets dramatically simpler: "we have the best model, we build developer infrastructure, and we don't sell your employees' attention to advertisers."

That last line will get used in sales decks. I'd bet on it.

The Pricing Divergence

Anthropic's surge pricing decision from last month is the tell. Surge pricing is an infrastructure primitive. Amazon uses surge pricing. Google Cloud uses surge pricing. Nobody's favorite SaaS company uses surge pricing. When you start pricing like AWS, you're telling the market what you think you are.

OpenAI's ad pilot is the equivalent signal. Ad-supported products are attention businesses. They are built around maximizing time-on-platform, reach, and engagement. Those are consumer media metrics, not enterprise software metrics.

Two companies. Same AI era. One is becoming AWS. One is becoming Meta.

The Developer Stakes

If you are a developer choosing a platform today, the question is not "which model is better." The question is "which company's incentives align with mine?"

Anthropic's incentive: you build more stuff, you use more compute, they make more money. Clean alignment.

OpenAI's incentive (at the consumer end): your users spend more time on ChatGPT, they see more ads, OpenAI makes more money. The developer is a secondary customer in an attention business.

At the enterprise and API level, OpenAI's incentives are cleaner — they want you to use the API. But the strategic drift toward consumer attention means the developer bet on OpenAI carries platform risk that the Anthropic bet does not. Platforms that optimize for consumer attention eventually make choices that are bad for developers. See: Facebook 2012-2016.

The Valuation Math

Anthropic's last reported private valuation is $61B. OpenAI's IPO target is $840B.

The 14x difference tells you the market's read on which model scales. Consumer attention at 600M MAU with ad monetization getting to $100M ARR in six weeks — that's a number Wall Street knows how to value. Infrastructure compute with $2.5B in ARR and growing — that's a harder story, but it's also the one with better gross margin potential.

My prediction: Anthropic crosses $40B ARR before going public, stays private through 2027, and uses the infrastructure story to set a valuation that doesn't require consumer-scale growth projections. OpenAI IPOs in 2026 at $600-700B — lower than the $840B target because the ad model creates questions the enterprise buyers don't love, and the Sora kill shrinks the TAM narrative.

Both companies will be fine. They're playing different games. The cat builds infrastructure. The robot sells attention.

Both sports have winners. The trophy just looks different.