Anthropic closed a $65 billion Series H yesterday at a $965 billion valuation. Read that again. They’re one funding round away from being worth a trillion dollars. And according to their own announcement, their run-rate revenue crossed $47 billion earlier this month.

That sounds incredible. It is incredible. It’s also misleading in a way that should make everyone nervous about where AI valuations are heading.

Run-rate revenue is not revenue

Simon Willison caught the key detail here. Run-rate revenue is what you get when you take your best recent month and multiply by twelve. It’s an annualized projection, not actual money in the bank. Anthropic has made a habit of reporting this number in funding announcements, which is a clever way to make growth look explosive.

Here’s what it actually means: if Anthropic had one really good month and made $4 billion, they can claim $48 billion in run-rate revenue. That doesn’t mean they’ll make $48 billion this year. It means that if this month’s performance continued for twelve months straight, they would. Big if.

This isn’t fraud. It’s standard practice in high-growth tech. But it’s also the kind of number that can obscure what’s actually happening. Did Anthropic’s revenue genuinely increase because enterprise adoption is accelerating? Or did they land a few massive contracts that front-loaded this month’s numbers? Run-rate revenue doesn’t tell you.

The valuation math is wild even if the revenue is real

Let’s say Anthropic really is on track to make $47 billion this year. That would put them at a 20x price-to-sales ratio. For context, that’s higher than Nvidia’s peak valuation during the 2021 bull run. It’s higher than most software companies. It’s the kind of multiple you see when investors believe growth will continue at breakneck speed for years.

Maybe it will. Anthropic’s announcement emphasizes growing enterprise adoption, and there’s real evidence that Claude is becoming the preferred model for developers who care about reliability and safety. The technical quality is there. The product is good.

But here’s the tension: Gary Marcus just published a piece arguing that customers are waking up to the reality that tokens are “getting burned for millions of dollars without any real significant ROI to show for it.” That’s not just Marcus being skeptical. That’s the conversation happening inside companies that are actually using this stuff at scale.

So which is it? Are enterprises doubling down on AI spend because they’re seeing real returns? Or are they starting to pull back because the ROI isn’t materializing?

Both things can be true at the same time, which is what makes this moment so strange. Early adopters who jumped in during the ChatGPT hype are hitting the point where they need to justify costs. Meanwhile, late adopters who waited to see proof are now coming in, impressed by what the early adopters built. You can have enterprise skepticism and growing revenue in the same quarter.

This raise sets the stakes for everyone

Anthropic’s $65 billion round isn’t just about Anthropic. It’s a signal to the entire AI industry about what scale of capital is required to compete. If you’re OpenAI or Google or anyone else trying to stay in this race, you now know what the ante looks like.

That’s probably the point. Anthropic has been positioning itself as the responsible AI company, the one that takes safety seriously and builds thoughtfully. But they’re also clearly positioning themselves to win, and winning requires compute at a scale that only a handful of companies can afford.

The announcement mentions this is likely their final private fundraise before an IPO. That makes sense. At a $965 billion valuation, there aren’t many places left to go except public markets. An IPO would also force more transparency around actual revenue, not just run-rate projections.

The honesty about Opus 4.8 is refreshing

Here’s what I actually like about Anthropic: they’re willing to be honest about incremental progress. When they released Claude Opus 4.8 yesterday, they described it as “a modest but tangible improvement” over the previous model. That’s such a contrast to the usual AI lab marketing, where every release is a breakthrough.

It’s possible to hold two ideas at once. Anthropic is building good models and being more transparent than most of their competitors. They’re also raising absurd amounts of money at valuations that assume exponential growth will continue indefinitely.

The question isn’t whether Anthropic is a good company or whether Claude is a good product. The question is whether the economics of foundation model development can actually support these valuations, or whether we’re watching a very sophisticated game of musical chairs.

If enterprise AI spend keeps growing and Anthropic captures a meaningful share, then a trillion-dollar valuation might look reasonable in hindsight. If Marcus is right and companies start cutting AI budgets because the ROI isn’t there, then this round will look like the top of the market.

We’ll know which one it is soon enough. Anthropic just made sure the stakes are as high as possible.