Legal Teams Are Getting an AI Subsidy. That's About to End.
Anthropic announced a pricing shift this week that signals the end of a temporary advantage for law firms. The AI company has begun billing enterprise clients based on actual compute consumption rather than flat-rate subscriptions. For legal departments, this matters more than most.
Legal work is the heaviest compute workload in enterprise AI use. A lawyer reviewing a commercial contract processes more than 100,000 tokens-the basic units an AI model computes-to analyze a single agreement. A complex M&A deal with exhibits, side letters, and disclosure schedules pushes higher. Three contracts reviewed simultaneously, a routine task in commercial work, can consume half a million tokens before producing a single output word.
Compare that to other professional functions. A marketing professional drafts a headline. A developer runs a code review-structured, short, repetitive. A finance analyst pulls a summary. Each task consumes a fraction of the compute that contract analysis demands.
Why Legal Tasks Cost More
Contract review requires cross-referential reasoning. Definitions sections inform operative clauses. Operative clauses inform exhibits and risk analysis. Long context windows cost more to process, introduce higher latency, and increase the risk of hallucination-meaning legal tasks demand premium models, run harder, and require more verification passes than any other professional function.
Energy consumption for long reasoning over large documents dwarfs a coding sprint or marketing brainstorm by an order of magnitude.
The Subsidy Window
Most legal teams using enterprise AI tools today pay a flat seat license or a fixed annual fee regardless of usage intensity. Heavy workloads have been effectively subsidized. A legal professional running 100,000-token document analysis, multi-pass review, and cross-contract queries is almost certainly paying 90 cents on the dollar less than the actual computational cost.
The meter wasn't running because there wasn't one. That's changing.
AI providers set flat-rate pricing before they understood how lawyers actually use these tools. They're absorbing the compute costs temporarily while scaling. But that subsidy window is narrowing measurably.
The Strategic Play
Enterprise agreements signed in 2024 or early 2025 lock in flat-fee pricing for the contract term. Once that agreement expires, the economics reset-and legal's disproportionate token consumption will get priced accordingly.
The legal department that maximizes AI for Legal adoption now builds workflows, trains staff, ingests contract libraries, and runs high-volume review cycles on someone else's dime. That builds institutional capability at subsidized rates. The legal department that waits for pricing certainty pays full freight for the same capability later.
The lawyers treating the current flat-fee window as a subsidized research budget will have a permanent structural advantage over those treating it like any other software line item.
The conventional argument is straightforward: Claude and similar tools save lawyer time, so the ROI is obvious. That's true. But it undersells the moment.
Right now, legal departments are receiving a structural subsidy from AI providers who priced before they understood the workload. Lock in that subsidy through a long-term contract. Build the workflows and institutional knowledge while the compute is cheap. Once the agreement renews, the pricing will reflect reality.
The meter isn't running yet. It will be.
Your membership also unlocks:
