SDSignal Desk

Anthropic is still grappling with the economics of compute

Nov 3, 2026, 9:07 AM · TechCrunch

Image: TechCrunch

Anthropic's revenue is surging, but its own filing shows each new dollar costing more to serve, and that tells us how exposed even the strongest AI lab is to the infrastructure race.

Why it matters

Anthropic updated its IPO filing with results through September 30, and the headline is strong: quarterly revenue reached $17 billion, up from $11.5 billion the quarter before. But the operating loss also grew, and compute ate 58 percent of revenue, up from 48 percent.

The filing's own chart shows revenue per dollar of compute climbing since early 2025 to just over 2x, then falling back to 1.72x. This is the clearest public look yet at what it costs a frontier lab to serve fast-growing demand, and it's arriving just as public investors decide what these companies are worth.

From the desk

We'd call this a growth problem, and a real one. Demand outran Anthropic's long-term compute contracts, so the company bought capacity on short notice at higher prices. The filing says $5.0 billion of the quarter's roughly $10 billion compute bill came under short-term commitments that cost significantly more than its multi-year deals. That's what happens when customers show up faster than data centers do.

TechCrunch's read is that this looks more like a blip than a structural problem, and we lean the same way. As more committed multi-year capacity arrives, the expensive spot-style compute should become a smaller share. Revenue growing at nearly half again in a quarter buys a lot of patience.

But we wouldn't wave it off. The filing says compute will keep rising in absolute dollars and remain the largest expense for the foreseeable future. Anthropic is leaning on neoclouds, smaller providers with flexible but pricier terms, including reported deals with SpaceX at $1.25 billion a month and multibillion-dollar arrangements with Nscale, Akamai and Volta. Its own effort to build capacity, Anthropic Direct, can't keep pace with demand on its own.

The bigger lesson is dependence. The lab's margins are hostage to a trillion-dollar buildout it doesn't control. If bond markets tighten or local communities keep pushing back on data centers, compute gets scarcer and pricier, and there's little Anthropic can do about it. That's a risk for investors, and it's a pressure point that can show up as higher prices or tighter usage limits for the people building on Claude.

I'm watching whether the next quarter shows the ratio recovering. If it doesn't, the story shifts from a scramble to a business model question.

Context

Anthropic filed its S-1 the week before, with financials through June 30. The update adds the third quarter. Hyperscalers are the largest cloud providers; neoclouds are newer, smaller compute providers that offer more flexible deals at higher rates.

Who feels it

IPO investors
Strong growth paired with worsening compute margins; the debate is whether Q3 is temporary or a trend.
Developers and enterprise customers
Expensive capacity can show up as pricing, rate limits or availability pressure if demand keeps outrunning supply.
Neocloud providers
Clear evidence that top labs will pay a premium for flexible capacity, at least for now.
Communities near data centers
Their opposition is now explicitly part of the risk picture for AI lab finances.

What to watch

  1. Whether revenue per compute dollar recovers above 2x in the next reported quarter.
  2. The share of compute under short-term versus multi-year commitments in future filings.
  3. Progress and scale disclosures on Anthropic Direct.
  4. Any pricing or usage-limit changes for Claude tied to capacity constraints.

Read the original

Continue at the source.

TechCrunch

Companies: Anthropic