Key Takeaways

  • Anthropic reported a third-quarter revenue of $16.8 billion, which annualizes to a $67.2 billion run rate, but its 45% quarter-over-quarter expansion missed investor targets pegged to 10x annual compounding.
  • Model distillation makes up between 5% and 7% of Anthropic's total revenue, revealing that billions of dollars flow directly from rival labs and clone developers extracting weights.
  • Jordi Hays pointed out that unauthorized distillation has become the industry's real benchmark: “Being on the frontier just means you're getting distilled. If you're not getting distilled, that's the definition in your mind. It's like unless you're not frontier.”
  • Wrapper startups are posting phantom revenues because distillation rings open thousands of programmatic accounts on their services to hide high-volume scraping from Anthropic.
  • Institutional investors are stepping back as capital costs rise; Brad Gerstner confirmed on CNBC that his fund sold its entire core AI exposure over the past three months in anticipation of market consolidation.

The 10x Expectation Trap

Growing 45% in a single quarter would mark an extraordinary milestone for any normal software business. For Anthropic, reporting $16.8 billion for the third quarter, equal to a $67.2 billion annualized run rate, produced immediate skepticism. John Coogan outlined the math: “It's crazy that you can be growing at 45% quarter on quarter and still not meet expectations. But when people have had in their minds 10x every year, 10x revenue growth every year and Anthropic goes to their investor day and says that Q3 revenue was 16.88 billion which annualizes to 67 billion.”

Public market investors had priced in continuous 10x compounding. When a company misses that curve, sentiment flips overnight. Gerstner told CNBC that his fund dumped its positions: “We have sold all of our core. We don't own any of it today. We've taken exposure down over the last 3 months. We felt AI had overshot and would lead to a period of consolidation.”

The Phantom Revenue of Model Distillation

The surprise inside the numbers is that between 5% and 7% of Anthropic's cash flow comes from people stealing its outputs. Competing teams run automated pipelines against Claude to train cheaper open-weight models. Coogan noted that political adviser David Sacks and the Anthropic leadership rejected regulatory bans: “One of the proposed solutions was ban open source, do all this crazy stuff, slow down, and David Sacks didn't want to do that. And then Anthropic came out and said, 'We deployed our own frontier technology to fight this.'”

This creates a ghost economy inside smaller software products. Coogan explained: “There's a lot of wrapper companies that have gotten to crazy revenues and basically no one uses the products because it's the distillers or the distilleries that are like, 'New wrapper company, can't wait to spin up a thousand accounts on you because you won't detect me.'”

Why Plugging the Leak Hurts Near-Term Growth

Anthropic is building systems to detect and block distillation clusters. Doing that protects their proprietary moat and widens the performance gap between closed frontier labs and open clones.

The downside is purely financial. If Anthropic successfully eliminates distillation, they immediately erase up to 7% of their top line. That turns a 45% quarterly growth rate into something closer to 38%, right when public markets are demanding impossible acceleration.

What to Do With This

Audit your paid API user accounts tomorrow morning. Export your top 5% of token consumers, group their request timings into 100-millisecond buckets, and look for uniform variance across IP subnets. If a customer runs thousands of parallel queries asking for structured answers without human latency, you are not serving an end user; you are subsidizing a distillation farm that will vanish the moment frontier rate limits hit.