Anthropic nears first quarterly profit while locking in £930m monthly SpaceX compute deal

TL;DR:

  • Anthropic told investors its June quarter sales could reach at least £8.6bn ($10.9bn), more than double the £3.8bn ($4.8bn) it booked in the March quarter, generating an expected Q2 operating profit of about £444m ($559m).
  • Disclosed in SpaceX’s IPO filing on the same day: Anthropic has agreed to pay SpaceX £930m ($1.25bn) per month for AI compute capacity through May 2029, covering both Colossus and Colossus II training clusters.
  • SpaceX’s own AI segment lost roughly £1.85bn ($2.5bn) on £610m ($818m) of revenue in the March quarter, indicating the seller of compute is still in the red even as a frontier-lab customer turns profitable.

The two figures, published within hours of each other, reset the standing narrative on frontier-lab unit economics. Until now, the working assumption among UK enterprise buyers and investors has been that the leading AI labs would burn cash indefinitely. Anthropic’s path to operating profit on Claude — driven by demand for coding tools and the Mythos security model — suggests the question is shifting from “can the labs make money?” to “how does the compute supplier get paid back?”

A £15bn-a-year compute bill changes how enterprises evaluate AI suppliers

The £930m monthly commitment annualises to roughly £15bn a year — an extraordinary single-vendor concentration. Either side can terminate with 90 days’ notice, but the public disclosure changes how UK CIOs and procurement teams should think about supplier viability. A frontier lab is now contractually tied to a private space company’s AI infrastructure timeline. Customers buying Claude or Mythos for the next three years are buying into that triangular dependency.

Anthropic’s revenue ramp also sharpens the competitive picture. OpenAI, which is itself moving towards an IPO with a confidential US filing expected in coming weeks, has historically led on consumer ChatGPT traffic. Anthropic’s enterprise lead — particularly with developers using Claude for code and security teams using Mythos for vulnerability discovery — is now visible in the financials, not just the procurement chatter.

Looking forward

For UK enterprise buyers, the practical questions follow the disclosures. Do existing Anthropic contracts depend on SpaceX clusters in ways procurement should flag for resilience reviews? Does Anthropic’s expected profit reduce or increase pressure on Claude pricing — and how does that change the multi-vendor approach UK CIOs have been building to avoid platform lock-in? The £15bn annual compute commitment also reframes how the UK should think about its own sovereign-compute ambitions: even a profitable frontier lab needs ten figures of capacity per year, and those clusters aren’t currently being built in Britain.

Currency conversions use £1 = $1.27 at time of writing.

Share this article