Anthropic's $965 Billion Valuation Isn't About AI Safety Anymore

Here's what I keep coming back to: when Anthropic was founded, the pitch was explicitly about building AI more carefully than OpenAI was willing to. The Amodeis left because they thought OpenAI was moving too fast, prioritizing deployment over alignment research. Now Anthropic is valued higher than OpenAI, and I can't find any recent public statements about how their safety research has evolved or what they've learned.

Maybe they're still doing that work. I should know this better, but their research publications have slowed considerably over the past 18 months. The Constitutional AI papers were genuinely interesting. What's come since?

What This Means for Embodied AI

You might be wondering why I'm covering this on the humanoids beat. Fair question.

The answer is that Anthropic's Claude models are increasingly showing up in robotics applications. Not officially (Anthropic hasn't announced any robotics partnerships), but researchers I talk to are using Claude for reasoning layers in manipulation tasks, for natural language interfaces with robots, for planning and decision-making in embodied systems.

A $965 billion Anthropic isn't a research lab that might eventually license some models. It's a company that will need to find massive revenue streams to justify that valuation. And robotics, particularly industrial and humanoid applications, is one of the few markets big enough to matter at that scale.

Some key points worth tracking:

• Anthropic has been hiring robotics researchers quietly for the past year (I've confirmed at least four from Google DeepMind alone)
• Claude's multimodal capabilities are increasingly competitive with GPT-4V for visual reasoning tasks that matter in robotics
• The company has said almost nothing publicly about embodied AI, which in my experience usually means something is brewing
• At this valuation, they need to find applications beyond chatbots and enterprise search

I don't have hard evidence that Anthropic is building a robotics platform. But the circumstantial case is getting stronger, and this funding round gives them the resources to move fast if they decide to.

The Salesforce angle is interesting too. Salesforce has been investing in automation across its platform. A $5 billion stake in Anthropic isn't charity. They're presumably expecting integration, and Salesforce's customer base includes a lot of companies that would pay for AI-powered automation in physical operations.

The Uncomfortable Question

Here's what I keep thinking about: at what point does a safety-focused AI company become just... an AI company?

I'm not saying Anthropic has abandoned its principles. I genuinely don't know. But the incentives at $965 billion are different than the incentives at $5 billion. The investors writing $2 billion checks aren't doing it because they believe in interpretability research. They're doing it because they expect returns.

And look, maybe that's fine. Maybe Anthropic can be both commercially dominant and safety-conscious. Maybe the best way to ensure AI is developed responsibly is for the responsible actors to win the market. I can see that argument.

But it's also possible that we're watching a company slowly become the thing it was founded to be an alternative to. The Amodeis left OpenAI because they thought commercial pressures were corrupting the mission. Now they're running a company worth nearly a trillion dollars.

I don't have a clean conclusion here. The funding is real, the valuation is real, and the implications for robotics and embodied AI are probably significant. What's less clear is whether the version of Anthropic that emerges from this round will look anything like the one that was founded five years ago.

We'll find out, I guess. But I think it's worth paying attention to what they do next, not just how much money they raised.

James Chen · 7 hours ago · 4 min