Key Takeaways

  • Wall Street analysts misread Anthropic's softer third-quarter revenue growth as weakening market demand or enterprise SaaS cohort exhaustion.
  • Anthropic disclosed in its S-1 filing that it spent $5 billion on spot-market GPU capacity, which made up 50% of its entire compute expenditure.
  • Spot market prices compressed Anthropic's gross margin by 10% because on-demand pricing runs far higher than contracted multi-year cluster commitments.
  • Compute purchase cycles require 12 to 18 months of lead time, meaning capacity caps in late 2025 were locked in during early 2024 forecasting sessions.

Wall Street Misdiagnosed the Deceleration

When Anthropic reported softer revenue growth for the third quarter, equity analysts quickly circulated a familiar narrative. They claimed enterprise interest had cooled, pilot programs were stalling, and customers had hit a ceiling on token consumption.

Patrick McKenzie looked at Anthropic's S-1 filing and saw the exact opposite reality written into the financial tables. The company had hit a hard physical ceiling, not a commercial one. Anthropic was simply out of chips.

As McKenzie pointed out, the filing is direct about this operational bottleneck: “Anthropic says in the S-1, 'We have historically operated in a compute-constrained environment in which demand for compute has exceeded the supply available to us.'” (play from 22:46)

Treating a factory running at 100% capacity as a sign of customer disinterest is a basic accounting error. If you sell every token your hardware can physically generate, your top line stops accelerating. That flat spot shows where your hardware contract ends, not where buyer demand stops.

The Five-Billion-Dollar Proof

The clearest evidence against the demand fatigue theory sits in Anthropic's spot compute bill. If customers were leaving or cutting their usage, a company would let its existing clusters handle the load and protect its margins. Anthropic did the reverse.

“Supporting evidence: they talk about hitting the spot market for GPUs for $5 billion, at substantially more than their multi-year commitments typically run, in such a way as to eat 10% of the margin just from the shift in the mix,” (play from 23:43) McKenzie noted.

Paying emergency spot prices for half of your total compute budget makes zero sense for a business watching customer demand soften. A company only swallows a 10% margin hit when paying customers are waiting in line and the long-term contract clusters are entirely full.

As McKenzie put it: “I think analysts who read this as softening demand would need to explain tapping spot for 50% of one's spend. I frankly don't think there is an explanation available there. These analyst- analysts are just straightforwardly wrong.” (play from 24:37)

The Twelve-Month Forecasting Trap

The root cause of the third-quarter plateau dates back more than a year before the numbers posted. Frontier AI companies do not order hardware on a weekly credit card swipe. They sign massive, multi-year capacity agreements years before they release the models that run on them.

McKenzie explained the forecasting math that shaped the quarter: “The S-1 says historically they make these arrangements 12 to 18 months in advance. And so in roughly Q1 or Q2 of 2025, they had to make a guess as to how compute-efficient a yet-to-be-released Opus 5.5 would be, and how many customers would be willing to use it.” (play from 23:03)

When a model turns out to be more capable or popular than the forecast predicted, revenue hits a wall. The engineering team cannot conjure an extra 50,000 liquid-cooled accelerators overnight. Growth freezes at the exact limit of the paper contracts executed five quarters earlier.

What to Do With This

Audit your primary vendor constraints tomorrow morning. If your core product relies on third-party API quotas, cloud credits, or specialized hardware allocations, calculate the exact customer volume that breaks your current contract tiers. Sign your next capacity tier before you hit 70% utilization, or you will find yourself buying off the spot market and sacrificing your margins to stay online.