Anthropic’s path to its anticipated IPO is revealing rough financial terrain: sharply rising compute costs are eating into margins even as revenue climbs. A recent SEC update extends the company’s financial reporting to September 30, showing revenue jumped to $17 billion from $11.5 billion in the prior quarter — but operating losses swelled alongside it. Compute costs now consume 58% of revenue, up from 48% previously. That shift marks a worsening, not improvement, in margins exactly when better ratios would help reassure investors.
Compute Margins: A Deepening Concern
Compute margins — essentially, revenue versus what it costs to power AI models — have been slipping. In previous quarters, Anthropic’s revenue per dollar spent on compute hovered near twice the expense; in the most recent quarter, it dropped to approximately $1.72 per compute dollar. That’s a stark signal that increased demand, combined with more costly contracts, is taking a toll.
A key driver: nearly $10 billion in compute expenses for the quarter, half of which came via short-term commitments. The company says these deals were significantly more expensive than long-term commitments. To keep up with surging demand, Anthropic has leaned heavily on “neoclouds” — smaller, more flexible providers like SpaceX, Nscale, Akamai, and Volta — as opposed to traditional hyperscalers. Those neoclouds typically come with a premium.
Balancing Demand, Infrastructure, and Risk
Anthropic is also investing in its own infrastructure under a project named Anthropic Direct, aiming to bring more compute in-house and reduce dependency on external vendors. But demand is growing so fast that internal efforts won’t keep pace, meaning external, high-priced options are likely to remain a core strategy for the near term.
Despite the slipping margins, revenue growth — nearly 50% quarter-over-quarter — gives Anthropic some leeway. But that only holds as long as demand remains strong and external compute stays available at tolerable prices. The company acknowledges that compute and infrastructure expenses will continue to rise as it enters more multi-year contracts with cloud service providers (CSPs).
Still, the report highlights broader risks. The infrastructure that underpins AI labs depends on massive data center buildouts. If bond market volatility or local resistance slows that pace, Anthropic and peers could face steeper compute costs and constrained supply. Such macro trends may ripple back to business models that assume stable or falling infrastructure expenses.
Why This Matters: Compute costs lie at the heart of scaling large AI models profitably. Anthropic’s recent filing shows that rising costs and demand are pressing on profitability even in a high-growth phase. Watching how its infrastructure plays out — both in-house and via external partners — will reveal how sustainable its margins can be as it moves toward an IPO. Investors and rivals alike should be tracking whether Anthropic Direct can offset some of those external costs, and whether compute prices stabilize as demand remains intense.
