An AI model company buys wholesale electricity by the megawatt and resells it as cognitive work. Dylan Patel of SemiAnalysis notes that the base cost of compute runs $10–15 million per megawatt, while Anthropic has generated up to $50 million per megawatt in revenue. This implies that every $10 spent on inference capacity yields $50 of revenue, creating a reinvestment loop. Gross margin as a function of electricity demonstrates that model companies can be profitable on a contribution basis.

Anthropic‘s gross margin was −94% in 2024, meaning $1.94 of compute cost for every $1 of revenue. By 2025 the corner turned, swinging from −94% to a 40–50% gross margin. In 2026, revenue passed cost: $50 million per megawatt against a $10–15 million cost. Anthropic booked its first profitable quarter with $10.9 billion of revenue and $559 million of operating profit. That 5% operating margin sits well below the 70–80% gross margin the megawatt math implies; training runs and headcount consume the difference.

Gross profit per megawatt depends not just on intelligence but also on efficiency. A model delivering the same intelligence at a fraction of the compute generates more profit per megawatt, even at a lower price. For example, GLM-5.3-Flash scores 57 on the Artificial Analysis Intelligence Index — identical to Claude Opus 4.8 — but uses 18 billion active parameters at a 90–97% reduction in cost. Fewer active parameters and less attention compute mean more tokens per megawatt. However, the frontier keeps climbing: three jumps of 3 points or more carry half of the 25-point gain from 37 to 63. Efficient models chase a target that resets every quarter.

Patel’s key point is that the profit per megawatt is reinvested entirely into training. This margin funds the factory. Nvidia paid $6 billion to acquire Poolside and invested another $1 billion on that premise. Model building becomes an industrial process: thousands of experiments across a search space, not artisanal hand-tuning. As Jason Warner of Poolside puts it, ‘The model is the output. The ability to keep building better models, faster & more efficiently each time is the actual innovation. The Model Factory is the compounding asset.’ Laguna S 2.1 went from kickoff to release in 52 days. Inference margin funds the factory that makes the next model cheaper to build and more efficient to run.