OpenAI is filing to go public at $730B

THE AI BRIEF

Today's signal: OpenAI is preparing to become a public company, and the procurement decision you make this year will be made under a different vendor than the one you started buying from. Plus five other things that moved.

In today’s issue:

• Main story: OpenAI is about to file to go public

• Also worth knowing: AI agents lose to baselines on production tuning, OpenRouter raises on routing volume, Qualcomm wins ByteDance agent hardware, Meta open-sources a faster attention kernel, and Netlify's CTO flags where stacks assumed a human was always there.

THE READ

The Wall Street Journal reported on May 20 that OpenAI is preparing a confidential IPO filing in the coming weeks, with Goldman Sachs and Morgan Stanley drafting the prospectus and a possible September 2026 listing. The reported valuation is $730 billion after the latest funding round. The Musk litigation outcome removed one structural roadblock, but public investors will still test OpenAI's corporate structure, Microsoft relationship, and revenue commitments before pricing.

The cultural read of an OpenAI IPO is that the company has won. Maybe. The operating read is that the vendor your team has been buying from since 2023 is about to become a different kind of company, on a quarterly schedule, with different incentives. Public-market AI companies have to defend gross margins. They have to manage capex disclosure. They have to deprecate older products to protect the next earnings call. None of that is bad. It is just not the same risk profile a procurement team built its OpenAI integration around two years ago.

Three changes worth tracking. The first is pricing. Private OpenAI absorbed loss-leader pricing on enterprise APIs to win lock-in. Public OpenAI will not. Expect price discipline on tokens and a faster sunset on cheaper model tiers. The second is roadmap volatility. Public companies trim products that do not contribute to the story they tell on earnings calls. Anything you are running on a deprecated or low-margin endpoint is a candidate to be cut. The third is concentration risk. A bigger share of enterprise AI demand now sits with a company whose primary obligation is shareholder return. That is the same arrangement enterprises have with Salesforce, ServiceNow, and Microsoft. Familiar, manageable, and worth pricing in.

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.

What I keep hearing from operators who built their AI stack on OpenAI in 2023 and 2024 is that they treated it like a research relationship. A startup that picked up the phone, took requests, custom-priced enterprise deals, and shipped what was asked for. That posture does not survive contact with public markets. The reasonable thing to do this quarter is to start assuming the OpenAI you renew with in 2027 looks more like Salesforce than the OpenAI you signed with.

Two things to do this quarter. Pull your OpenAI invoices for the last 12 months and identify which workloads sit on which model tier. The ones running on older endpoints are the ones most exposed to a forced upgrade. And add a second-model routing layer to the renewal conversation. OpenRouter's funding round, also today, is the market telling you the same thing the IPO is. Routing is no longer optional infrastructure.

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

InferenceBench shows frontier AI agents losing to simple baselines on server optimization. A new benchmark from Maksym Andriushchenko and collaborators measures whether AI agents can optimize OpenAI-compatible LLM servers on H100 GPUs. Frontier models fail to beat plain vLLM hyperparameter tuning, with smaller models like Claude Sonnet 4.6 ranking higher because they preserve simpler valid configurations. The signal worth knowing: agents demo well and lose to baselines on real production tuning, which is the work that actually drives inference cost.

OpenRouter announced a major funding round on the back of production routing volume. Usage data continues to point toward routing across many models as core infrastructure rather than a developer convenience. The signal is that the AI stack is moving from one-model applications to routing by cost, latency, and reliability across labs, which means the procurement conversation expands beyond a single frontier vendor.

Qualcomm reached a deal with ByteDance to supply millions of ASICs for the Doubao agent data centers. Bloomberg surfaced the reporting via Techmeme. The relevance is that agent infrastructure is now a volume hardware market on its own, not a Nvidia-only story, and the cost curve for running agents at the scale ByteDance is targeting begins to diverge from the cost curve for training frontier models.

Meta open-sourced a Blackwell GPU attention kernel claiming 2.3x faster attention. PyTorch announced TLX Block Attention, a warp-specialized Triton kernel for block-diagonal attention patterns on Nvidia Blackwell. The signal: model serving and training cost is being attacked at the kernel layer in the open, which compresses the speed-and-cost advantage closed labs have been earning from proprietary inference stacks.

Netlify CTO Dana Lawson said AI agents are exposing where developer stacks assumed a human would always intervene. Posted via AI Native Dev. Lawson's point is architectural. Software delivery systems have a thousand quiet assumptions about a human at the keyboard interpreting errors, approving deploys, or bridging gaps between tools. Those assumptions become production bugs when an agent is running the workflow.

WATCHING TOMORROW

Any OpenAI's public response to the WSJ report, and whether Anthropic or Google adds procurement-side signals in the same week.

REPLY

Hit reply and tell me: if your largest AI vendor went public next quarter, what is the first thing you would want your procurement team to do? I read every reply.

FORWARD

If a colleague is sitting in a meeting this week about renewing an OpenAI contract or standardizing on a frontier model, forward this issue to them.

Back tomorrow,
Haroon