A $100M AI buyer swapped in a model 23x cheaper

THE AI BRIEF

Today's signal: Frontier-model bills stopped falling, and the largest AI buyers started swapping in cheaper tiers from the same vendors.

In today’s issue:

• Main story: A $100 million AI buyer just told the press the cheaper model is fine

• Also worth knowing: SpaceX renting Colossus, OpenAI's Q1 burn, Qualcomm circling Modular, Meta paying $900M to hire a WhatsApp CEO, and quantum on the President's desk.

THE READ

Ensemble Health Partners switched to an OpenAI model, 23 times less expensive than the frontier tier, and said it worked.

Ensemble Health Partners, a software vendor that processes hospital revenue cycles, told The Information yesterday it is on track to spend up to $100 million on AI this year, and that it has had "success" moving workloads from OpenAI's most advanced model to a model 23 times cheaper from the same vendor. The reporting pairs the Ensemble disclosure with an OpenRouter data point: 65% of tokens processed by the developer-routing platform in June came from open-source models.

I've been in three procurement conversations over the last two weeks where the same question came up in a different form. The CFO wants to know why the AI line item keeps growing when the vendors keep advertising lower prices. The answer is usually that the team started on the most expensive model in the first pilot, then never moved off it. Nobody owned the question of whether the cheap tier would do the same work. Ensemble is the first $100 million-class buyer to put a name on that question and answer it publicly.

AI READY PRO · FREE UNTIL FRIDAY

I will get you AI trained in 30 days for free

After several months of running it quietly with top AI operators and teams, I’m excited to launch AI Ready Pro today to our newsletter subscribers like you.

It’s a 30-day personalized AI training program that is personalized to you and where you are in your AI journey. It’s the culmination of 100’s of hours spent teaching AI to top Fortune 500 AI teams and operators (and working with Mark Cuban).

Each day, you receive a 10-15 minute exercise to complete to improve your AI skills.

This isn’t vague AI theory or a way to pitch you a tool. It’s 30 days of learning to use AI in your real work. So you come out the other side with the skills to use AI to actually improve you and your teams output.

Until this Friday (May 15), it’s free for newsletter readers who complete the extended assessment. This assessment will help us personalize the learning experience to you.

If you lead a team, AI Ready Team is open today too, also for readers first. It’s the same engine, but to train your whole organization. You’ll get a detailed view of:

• Your team’s AI readiness level + detailed strengths & weaknesses

• Ranked list of AI automation opportunities

• Get tactical advice on how to advance AI efforts in your org

Leading a team? Take the Team assessment instead.

That changes the procurement conversation in a small but real way. When the question used to be "are we on the best model," it could only be answered by trying every vendor. The new question is closer to "are we on the cheapest model that does the work," and that question has a measurable answer per workload. The work is unglamorous. Someone has to read the prompts, run the smaller model against the same evaluation set, and confirm the output is within tolerance. This is line-item work, not strategy work, and the companies that do it are quietly compounding savings against the ones that don't.

The honest hedge: this only matters where the workload is high-volume and the evaluation criteria are clear (classification, extraction, structured output, routing). For complex reasoning, agent loops, or anything customer-facing without a safety net, staying on the frontier tier is still the right call. Ensemble is in revenue-cycle automation, which is exactly the high-volume, structured-output category where the cheaper tier should win. Read the Ensemble disclosure as confirmation of where the swap works, not as a general blessing.

SPONSORED BY CLUTCH

Hire secure AI teammates that work 24/7.

Hire pre-built AI teammates. Give your engineers and operators a platform to ship their own AI apps. Stop losing sleep about what is running where.

Clutch is the platform behind both: pre-built agents for the workflows your ops team should automate first, plus the integration plane your team's vibe-coded apps and Claude Code projects plug into. One platform. Real production. Visible and safe by default.

Built for ops, engineering, and security teams that are tired of the shadow-AI surface area inside their own company.

ALSO WORTH KNOWING

SpaceX signed a compute deal worth up to $6.3 billion with Reflection AI for access to its Project Colossus supercluster. Reflection will pay roughly $150 million per month to train open-source models on Nvidia GB300s. SpaceX is now in the business of renting compute to other AI labs, not just running its own.

OpenAI launched GPT-5.5-Cyber and a security partner program called Daybreak in limited release alongside an updated Codex Security plugin. Access is governed through partner integrations rather than direct API. Operators evaluating AI for security tooling now have a named OpenAI track to ask their existing vendor about.

Qualcomm is in talks to buy Modular for about $4 billion, per Bloomberg. Modular makes infrastructure software for running AI models across different chips. If the deal closes, Qualcomm acquires the abstraction layer that lets buyers pick non-Nvidia silicon without rewriting their stack.

Meta is paying $900 million for a roughly 20% stake in Indian fintech Cred and hiring its founder Kunal Shah to run WhatsApp. Will Cathcart, the outgoing WhatsApp head, is stepping back after seven years. The move signals Meta is treating WhatsApp as a payments and commerce surface, not a messaging app.

WATCHING TOMORROW

OpenAI's $3.7 billion Q1 cash burn (Erin Woo's exclusive in The Information) will set the framing for the IPO timeline conversation through the week. Watch for whether Sam Altman or Sarah Friar comments on the burn figure on the record, and whether any S-1-stage disclosures move.

REPLY

Have you actually moved a workload from a frontier model to a cheaper tier this year? What was the workload, and what was the swap? I read every reply.

If a colleague keeps approving AI invoices without asking which model the bill is for, forward this issue to them.

Back tomorrow,
Haroon