Full narration of the briefing. Click a section heading to jump the player to that moment.
Every negotiation playbook you own assumes there is someone on the other side of the table. With OpenAI and Anthropic, that assumption fails first. These are companies with more demand than they can serve, published pricing, and almost no sales force. The tactics that move Oracle or Salesforce do nothing here, because there is no quarter-end rep chasing your signature.
There is a rate card, a queue, and a very small team focused on the largest deals on earth. Here is how to operate in that market.
Section one. Accept the structure of the market. In 2026, OpenAI and Anthropic each run fewer than fifty sales representatives globally... and those teams are concentrated on transactions of one hundred million dollars and more.
Below roughly ten million in committed spend, it is unlikely a negotiation even starts: you buy at published rates like everyone else... and no amount of procurement process changes that. This is not vendor arrogance. It is arithmetic.
Demand exceeds supply, and scarce sales capacity goes where the money is. Plan your strategy for the market that exists.
Section two. Calibrate your expectations on price. When you are large enough to get a conversation, the discounts are modest and mechanical... five to twenty-five percent, driven almost entirely by the size and length of your commitment.
There is no deal desk theater, no end-of-quarter collapse, no forty percent off for waiting. The rate card is public, margins are tight against compute costs, and both vendors know the next customer in the queue pays list. Anchor your business case on engineering the bill down, not on negotiating it down... and treat any commercial discount as the second win, not the first.
Section three. Build the one lever that works. The only leverage in this market is credible competition between OpenAI, Anthropic, and Google's Gemini, backed by a valid, documented case. That means workloads actually benchmarked across models, an architecture built for portability so switching is an engineering task rather than a rewrite...
and evaluation results you can put on the table. A buyer who can genuinely run on two of the three, and prove it, gets attention... that headcount and logo prestige do not buy. A buyer who cannot is a price taker with a procurement process.
Section four. When you do get the conversation, spend it on structure. Commit to spend, not to specific models, so your agreement flexes as the model lineup changes. Insist on price-decline protection or short terms, because list prices in this market fall fast...
And a three-year lock at today's rates is a gift to the vendor. Nail the enterprise terms that actually carry risk. Data usage, retention, training exclusions, and capacity priority. And size the commitment conservatively...
exactly as you would any consumption deal, because unconsumed commit is margin you donated to a company that did not need the help.
Section five. Take the bigger discount nobody has to approve. Model routing, sending small tasks to small models, prompt caching, and batch processing routinely cut AI bills far more than any negotiated percentage... Engineer first, consolidate your spend second, and they require no meeting with a sales team that does not exist.
and when your committed volume finally crosses the threshold where the vendor will talk, arrive with clean consumption data and a portable architecture. In this market, the prepared buyer does not negotiate harder. They need the negotiation less, which is the same thing.
One last point. At Redress Compliance we advise large enterprises on OpenAI, Anthropic, and Gemini agreements on a pure contingency basis. Our fee is 25 percent of what we save you. If we save you nothing, you pay nothing.
Before you commit at list, let us look at the structure. com.
Redress Compliance works on contingency: our fee is 25 percent of what we save you. Nothing saved, nothing paid. Independent, buyer side only, never vendor funded.
Talk to a OpenAI / Anthropic negotiator