The subsidized-platform pattern
We have seen this movie. A new technology platform arrives priced to build dependence: ride-hailing below cost, cloud credits by the truckload, delivery for less than the driver’s pay. The service is genuinely useful and genuinely underpriced — until the platform has enough of the market, and the real cost, plus margin, arrives on the invoice.
Why does this matter for a law firm’s AI?
If a firm builds a core function — drafting, review, analysis — entirely on rented frontier compute, it hands a provider the ability to raise the rent later on something the firm now depends on. The objection is not to paying fairly for frontier models; it is to letting a central business capability sit at the mercy of someone else’s future pricing decision.
Hybrid, not absolutist
The answer is not “local only” or “never touch a frontier model.” A frontier model is a superb tool for the hardest reasoning and the best prose. But you do not use a blowtorch to light a candle. A great deal of a law practice’s recurring work — extraction, structuring, summarizing, first drafts — can run on smaller local or specialized models the firm owns, reserving the expensive frontier calls for the work that genuinely needs them.
Owned hardware changes the cost curve. After the up-front investment, the marginal cost of a local inference is electricity. There is no per-seat meter, no surprise price change, and no model that quietly changes behavior overnight because a vendor shipped an update.
Owning the stack is also owning the behavior
The economic argument has a control twin. When a firm owns the models it runs, it decides which model runs, when it updates, and how it behaves. A cloud model can change under you between one Friday and the next, altering the output of a workflow you had carefully tuned. A local model you control does not move unless you move it — which, for a system that produces court filings, is a feature, not a limitation.
What the assessment answers
None of this is a blanket recommendation to buy a server. The right split between owned and rented depends on a firm’s volume, sensitivity, and budget. That is exactly what a scoped assessment is for: what your work actually requires, what it costs to own rather than rent, and where — honestly — you should not spend the money at all.
This article is general information from a technology consultancy, not legal advice, and does not create an attorney-client relationship. Figures describing the founder’s own practice are illustrative, not a promise of results.