No profession is getting as good a deal on artificial intelligence as the legal industry today. In one sense, AI is paying lawyers to use it. But that will likely change soon, so the time to act is now.
Anthropic just announced its pricing model update for high-volume users. It has started billing some enterprise clients based on how much compute their AI workloads actually consume. The strategic logic is clear: As demand for Anthropic chatbot Claude’s capabilities surges, Anthropic needs a pricing framework that doesn’t leave it subsidizing the heaviest users, which includes the legal industry.
Legal work is the heaviest compute workload in the enterprise. A marketing professional uses AI to draft a headline or reframe a brief. A developer runs a code review — structured, short, repetitive context. A finance analyst pulls a summary from a model. Quick, targeted, low level of compute.
A lawyer reviewing a commercial contract? That’s a different beast entirely. Modern legal AI models process more than 100,000 tokens, the basic units that an AI model computes, to analyze a single agreement while maintaining an understanding of how clauses relate to each other.
A complex M&A agreement with exhibits, side letters, and disclosure schedules can push that further. Put three contracts in context at once—routine for any serious commercial review—and you’re at half a million tokens before you’ve written a single output word.
It’s not just volume. It’s the nature of legal work. Contracts require cross-referential reasoning: The definitions section must inform the operative clauses, which must inform the exhibits and the risk analysis. Long context windows cost more to process, induce higher latency, and increase hallucination risk — which means legal tasks require premium models, run harder, and demand more verification passes than any other professional function.
Energy consumption can be much higher for long reasoning or multimodal prompts than for simple queries. Legal AI sessions are, by definition, long reasoning over large documents. The compute bill per legal session dwarfs a coding sprint or a marketing brainstorm by an order of magnitude.
The AI providers know this. Lawyers mostly don’t.
The pricing model was never designed for what lawyers are doing. Right now, most legal teams using enterprise AI tools pay a flat seat license. Some enterprise contracts cap it at a fixed annual fee regardless of usage intensity.
Heavy workloads were effectively subsidized at flat subscription prices that didn’t reflect the true computational cost of continuous operation. Legal professionals doing serious contract work — pulling in 100,000-token documents, running multi-pass analysis, querying multiple agreements against each other — are almost certainly being subsidized 90 cents on the dollar or more. We haven’t been paying attention to the meter, because there wasn’t one.
That’s changing. Fast. The energy and compute costs are real. The providers are absorbing them temporarily while they scale. But that subsidy window is measurably shorter than it was six months ago.
There’s an argument most legal leaders are missing. The conventional framing is: AI saves time for lawyers, so the ROI is obvious. That’s true, but it undersells the moment.
The better framing is this: Right now, legal departments are receiving a structural subsidy from AI providers who set prices before they understood how lawyers use these tools. A flat fee on an enterprise agreement signed in 2024 or early 2025 locks in that subsidy for the contract term. Once the agreement is up, the economics reset—and legal’s disproportionate token consumption will get priced accordingly.
The legal function that maximizes AI adoption now—builds the workflows, trains the team, ingests the contract libraries, runs the high-volume review cycles — locks in institutional capability on someone else’s dime. The legal function that waits for pricing certainty will pay full freight for the same capability.
The lawyers who treat the current flat-fee window as a subsidized R&D budget will have a permanent structural advantage over those who treated it like any other software line item.
The meter isn’t running yet. It will be.
This article does not necessarily reflect the opinion of Bloomberg Industry Group, Inc., the publisher of Bloomberg Law, Bloomberg Tax, and Bloomberg Government, or its owners.
Author Information
Matthew Knowlton Fawcett is executive vice president and general counsel for DXC Technology.
Write for Us: Author Guidelines
To contact the editors responsible for this story:
Learn more about Bloomberg Law or Log In to keep reading:
See Breaking News in Context
Bloomberg Law provides trusted coverage of current events enhanced with legal analysis.
Already a subscriber?
Log in to keep reading or access research tools and resources.