Microsoft Copilot, OpenAI, Anthropic, Google Gemini, Databricks AI and Amazon Bedrock all use commercial constructs that did not exist three years ago. Usage-based pricing, model-training rights, indemnity carve-outs, output liability and exit terms — we negotiate every line, before it becomes a precedent.
AI vendor contracts are not enterprise software contracts. The unit of consumption is a token, a request, a model invocation or an "active employee" depending on the vendor. The IP and data clauses are unsettled. The indemnity carve-outs are aggressive. The pricing escalators are punishing. And the buyer is usually signing the vendor's first-draft paper.
We have negotiated AI vendor contracts across Copilot, OpenAI, Anthropic, Gemini, Bedrock, Databricks and the long tail of vertical AI vendors. The objective is to bring buyer-side discipline to a contract type the market is still inventing — before the precedent locks in for the next decade.
We do not assess your AI roadmap or recommend which model you should use. We do not implement governance tooling. Our remit is the commercial and legal layer: pricing, data, IP, indemnity, term and exit.
The set of vendors we routinely negotiate is shifting quarterly as the market moves. We track pricing, terms and precedent across every meaningful AI vendor as it changes.
An AI vendor negotiation runs 6 to 10 weeks. A clauses-only review and remediation engagement on existing AI paper can run 3 to 4 weeks.
Most AI vendor negotiations run on a fixed-fee basis. Where a clean baseline exists, success-based fees against documented savings are available. See engagement models →
Model the realistic usage envelope: how many users, how many invocations, what model mix, what context length. Vendors price aggressively against speculative use cases the buyer never actually hits.
Decompose the pricing model: per-user vs per-token, included credits, throughput tiers, model-tier surcharges, fine-tuning overages. The headline rate is rarely where the cost lives.
Review and rewrite the data, IP and indemnity clauses. Customer data, prompts, outputs, model-training rights, output ownership, output indemnity, and the vendor's carve-outs for "open-source-derived" or "third-party-model" liability.
Counter-proposal with the pricing structure, model mix, term length, commitment shape and clause set we recommend. Make every concession deliberate and traded.
Lead or co-lead the negotiation. The AI vendors negotiate harder than expected; their corporate development teams are usually behind the deal. Buyer authority on the final deal always sits with you.
Hand over a re-opener calendar. AI pricing is moving fast enough that any deal locked at signature is overpriced inside 18 months unless a re-opener trigger is built in. We build it in.
"Our first Copilot proposal had pricing locked for three years with no re-opener and full model-training carve-outs left in. Their team renegotiated it into a 12-month deal with a price-protection re-opener and explicit data-training prohibition."
Send us the vendor, the proposed annual spend, and the scope of the AI deployment. We will model the negotiable headroom within one business day.
