Enterprise just took the wheel
Reuters reports that OpenAI president Greg Brockman said Wednesday at the WSJ Tech Live conference in Napa that the company now generates the majority of its revenue from enterprise sales.
· The Aigentic · Morning Brief
Good morning. For years ChatGPT looked like a consumer product that happened to sell to companies. Overnight, OpenAI said the balance has flipped. President Greg Brockman told a Wall Street Journal conference that enterprises now bring in most of the company’s sales, and that a 2027 IPO is within reach. Around that pivot, OpenAI is buying developer infrastructure aimed at GitHub, its agents just found hundreds of ways to root Linux, and Wall Street is shopping slices of a forty-billion-dollar SpaceX chip lease. Here is what moved while you were away.
the aigentic
Know what matters in AI.
The stories shaping AI, the tools worth trying, and what they mean for your work.
Morning Brief + Closing Time. Two emails every weekday.
Free. Unsubscribe anytime.
Reuters reports that OpenAI president Greg Brockman said Wednesday at the WSJ Tech Live conference in Napa that the company now generates the majority of its revenue from enterprise sales. Asked whether OpenAI will be ready for an IPO in 2027, he said, “I think so.” Brockman tied the shift to a push against Anthropic for business customers, including Codex, and said Anthropic is still likely to list first before year-end. The consumer story is not over. The buyer that pays the bills has changed.
Axios reports that OpenAI will acquire Pierre Computer, a developer-tools startup whose Trees and Diffs libraries already power parts of Codex. OpenAI plans to fold Pierre’s Code Storage into Codex and keep Pierre’s open-source projects open. Terms were not disclosed. The deal lands as OpenAI builds Pando, its in-house source-control backend for engineers and agents, and as rivals race to replace a GitHub that was built for humans, not fleets of coding agents. OpenAI is no longer just writing code. It wants the rails the code rides on.
OpenAI says an autonomous research pipeline powered by an upcoming cyber model identified 2,308 Linux kernel findings, reproduced 1,724 of them, and produced 948 working local privilege-escalation exploits that could turn an unprivileged user into root. The mid-October run cost about $1.21 million in API credits, or roughly $1,279 per exploit, and generated candidate patches for human review. Linux maintainers had already patched 61 of the LPEs upstream by the time the work wrapped, with more in progress. Finding bugs at machine speed is the easy part. Keeping the world’s most important kernel patched at that speed is the hard one.
The Wire
The latest headlines from our live AI news feed.
Bloomberg Law reports that Goldman Sachs and Apollo Global Management have begun approaching Wall Street investors to sell portions of SpaceX’s roughly $40 billion AI chip financing, with some buyers offered stakes of up to $500 million ahead of formal syndication. An Apollo affiliate is buying GPUs and server hardware and leasing them to SpaceX for a little over five years from delivery, expected in the first half of 2027, in a structure meant to keep the debt off SpaceX’s balance sheet. The biggest GPU financing on record is now looking for more hands to hold it.
Anthropic open-sourced FlashProve, a research prototype that uses zkSNARK proofs to show a reply came from a specific committed model without revealing the weights. On Meta’s 400-billion-parameter Llama 4 Maverick, a single Nvidia H100 generates proofs at about 110 tokens per second, with each roughly 19 MB proof checkable on an ordinary CPU in under half a minute. The prototype is limited to 1,024-token contexts and greedy decoding, and Anthropic says it has no plans to deploy it yet. Still, it is a first real crack at ending the “just trust me” era of model serving.
International Business Times reports that AMD booked a roughly $1.1 billion non-cash charge in the third quarter to revalue warrants granted to OpenAI and Meta in exchange for multibillion-dollar AI chip commitments. Each customer can buy up to 160 million AMD shares at a penny apiece as shipment and share-price milestones are hit. CFO Jean Hu said the charge mainly reflected AMD’s rising stock price and will grow as the deals progress, though it does not hit free cash flow. Equity kickers helped lock the customers. They are now showing up on the income statement.
TechCrunch reports that Meta has acquired Conductor, an AI coding startup that lets developers assign separate pieces of a project to multiple agents running in parallel workspaces. Founder Charlie Holtz said Conductor’s cloud service will shut down on November 20, with cloud data deletion after that, while teasing a new product for early 2027 under Meta’s Superintelligence Labs. Meta confirmed the deal but gave no terms. The labs are not just shipping coding agents. They are buying the orchestration layers around them.
Deals Desk
The newest compute and power deals on our tracker:
Nov 4 — Unnamed Frontier AI Lab × Riot Platforms: A 15-year triple-net powered land lease for the entire Corsicana, Texas campus. Capacity is 910 MW of critical IT on 332 acres, with about $28 billion in total contract value, or about $54 billion with both five-year extensions. Delivery is staged as 400 MW in September 2027, 200 MW in January 2028 and 310 MW in July 2028. GlobeNewswire
Nov 4 — Google × Elmina Malaysia campus: Google’s data center campus at Elmina Business Park in Selangor is now live, with further campuses planned in Selangor and Negeri Sembilan. Campus IT megawatts were not disclosed. Google pledged more than 1 GW of new clean-energy capacity to Malaysia’s grid under clean-energy PPAs. Enterprise IT News
Full tracker → theaigentic.com/deals
Yesterday’s movers
Watch
Four picks this morning: Intelligent Machines on tech billionaires meeting the voters, Cole Medin stress-testing open models, Alex Finn letting Claude Opus 5.5 build games, and How I AI inside Figma’s product-management OS.
Intelligent Machines — Unsustainably High - Tech Billionaires vs. Voters
The Intelligent Machines panel digs into what happens when the people funding the AI boom run into the people who live next to the data centers.
Coming right after a midterm night that punished parts of the buildout, the conversation tracks the political risk that capital markets are only starting to price.
Watch this if you want the hangover briefing after election night for the AI trade.
Cole Medin — I REALLY Want Open Models to Win, But They Just Don't Cut It (Yet)
Cole Medin wants open-weight models to catch the frontier labs, then shows where they still fall short on the work he actually ships.
He walks through agent and coding tasks where the open stack looks close on a benchmark and then loses on reliability, tool use and long-horizon work.
Watch this if you are choosing between open weights and a closed API for a production agent.
Alex Finn — Claude Opus 5.5 is INCREDIBLE at building video games
Alex Finn hands Claude Opus 5.5 a game brief and lets it build, iterate and debug in public.
The demo is less about a finished title than about how far a coding model can push interactive software when you keep feeding it playtests and constraints.
Watch this if you want a vivid sense of what the newest Opus release can do beyond a CRUD app.
How I AI — Figma's PM OS is just a Git repo
How I AI sits down with a Figma product leader to show how the company runs product work out of a Git repository instead of a tangle of docs and slides.
The walkthrough covers how specs, decisions and agent-friendly context live next to the product itself, and why that shape matters when AI tools join the team.
Watch this if your product process still lives in twelve tabs and zero version history.