Your $19 Copilot seat is now a $3,000 bill

A developer's GitHub Copilot bill went from $19 a month to $3,127 in three days. The same day, MiniMax switched its open-weights model from flat-rate to token billing and the stock fell 15%. Today's issue is the read on what both events mean for every AI tool sitting on your company's books, and what to do before the next finance meeting.

In today's issue:

• Main story: Your $19 Copilot seat is now a $3,000 bill

• Since Friday: Stargate Michigan breaks ground, Microsoft Build opens Tuesday, and Nvidia's consumer AI chip lands at $599

On Sunday afternoon, a developer named Jakub Polec posted a screenshot of his GitHub Copilot dashboard. The agentic coding workflow he ran the week before, paid for at $19 per user per month, had generated $3,127 in token charges in three days. The same evening, the Chinese AI lab MiniMax launched M3, a one-million-token open-weights model with sparse attention and multimodal IO, and switched M3 from flat-rate to token billing the same day. The stock fell 15% before the U.S. market opened on Monday.

Two billing transitions, on the same calendar day, on opposite sides of the world, in tools that share almost no users, is not a coincidence. It is the end of a phase, and the phase that replaces it is one your finance team will be reading about by Wednesday.

Setup

For three years, the AI tools your company runs have been priced below their actual unit economics. Developer seats at $19, ChatGPT Enterprise at $30, and Copilot at $39. Those numbers were never the cost of serving the workload. They were the cost of buying habits. The land grab paid for itself through future contract value and lock-in, and venture money sat on top of inference costs that ran 3x to 50x what the seat ever charged.

That phase had to end. The Copilot transition was announced two weeks ago and went live on June 1. MiniMax's was scheduled the same day. Both companies shipped real capability improvements alongside the price change, and neither hid the move. The market reacted anyway because the buyer side was not ready. Buyers have been running production workflows for two years against a price tag that was not priced for production.

The turn

The headline on June 1 is "subscription pricing changed." That is the surface event. The operating event is that every AI tool sitting on your company's books today was priced for the land grab and is one product-update cycle away from being priced for the workload.

The Copilot dashboard bill is the visible version. The Starbucks dairy-counting failure I wrote about last week is the invisible version. Both are cases of an AI workflow whose true cost was absorbed somewhere outside the line item the CFO was tracking. The difference now is that one of those costs is moving onto the line item where it belongs, and the other categories sit one product update from the same move.

The bill is real, and the math used to defend it is unverified.

Polec's $3,127 is one developer over three days. The read across enterprise Copilot deployments running agentic workflows projects $1,500 to $100,000 per month per seat, depending on usage intensity. Microsoft has not yet released enterprise cost caps. The first month's invoices for early-billing customers land before end of June. Satya Nadella's Tuesday Build keynote is the first chance to put caps on the table.

The defense Microsoft will offer is the productivity number. Jensen Huang said at Computex on Sunday that GitHub-tracked code commits are running at three times the 2025 rate, and he attributed the gain to agentic coding tools. Three times is the largest single-year developer productivity number ever measured on a comparable surface. If that number is real and durable, the new pricing has a defense. If it is half real or front-loaded, the new pricing has a problem the size of GitHub's enterprise renewal book.

The point operators have to sit with is that nobody at your company is currently tracking commit volume or completion velocity in a way that would let you answer the question. The pricing change just made that measurement the difference between a defensible AI budget and an indefensible one.

MiniMax is the leading indicator, not the outlier.

MiniMax is not a critical vendor for most U.S. companies. The reason the 15% drop matters is that the stock fell despite a same-day capability launch that any serious technical reader would call a real upgrade, and the market priced the billing change as a bigger event than the model release. That is the first time the market has done that on a public AI ticker.

Two other tickers are in the same seat. OpenAI is preparing an IPO, and Anthropic confidentially filed its S-1 on Sunday at a reported $965 billion valuation. Both companies have priced enterprise tiers below the land-grab line for the same reason MiniMax and GitHub did, and both will move that line inside the next 18 months. The MiniMax tape is the first read on how the market scores the move when it lands.

The platform team's bill from last week now has a number.

Last Tuesday's Memo argued that AI deployments create operational debt the platform team absorbs without it ever appearing on the AI line item. The CFO had to find that cost manually, inside overtime, incident response, and quiet feature retirements. June 1 moves a chunk of that cost onto a visible invoice, which gives platform leadership a cleaner case to present to finance. It also means a CFO who has been under-funding the platform side of every AI pilot now has a vendor invoice telling them the bill is coming due. The conversation moves from "are we over-spending on AI" to "are we under-spending on the team that absorbs it."

AI READY PRO · FREE

A 30-day AI program. In your inbox. Then it ends.

One email every morning at 7 am ET. Each one is a short read and one real thing to try before lunch. By Day 30, you have nine concrete capabilities, including prompting that hits your bar on the first draft, AI workflows that take five hours a week off your plate, and the language to lead the AI conversation at your company.

Built for ops leaders, COOs, chiefs of staff, founders, and team leads at mid-market companies. The four weeks move you from foundations (effective prompting, catching hallucinations, learning AI with AI) to connect (a personal context layer, daily automation, five hours a week back), to build (reusable skills, agent design and debugging), to lead (tool and model evaluation, leading the conversation at work). Twenty-one mornings, no upsells, no community Slack.

Counterargument

The strongest objection is that this is a one-time correction, not a structural reset. Vendors will offer caps, contracts will get re-negotiated, and Microsoft will respond at Build before the market fully absorbs the move. The land-grab pricing was always temporary, and the customers who were going to stay are the same customers who will pay the new rate.

I want to be honest about the bear case. For the correction to be one-time, three things have to hold. Microsoft has to ship enterprise cost caps in the next 30 days. The 3x productivity number has to hold up to a second-party audit. And nothing else priced underneath unit economics can flip in the next two quarters. None of those is impossible, and none of them is the default outcome either. The default outcome, based on what every other software category has done after a buy-side reset, is a year of repricing and a smaller, denser group of vendors on the other side.

What to do this week

Run a five-line audit before the next finance meeting. List every AI tool you pay for, the current per-seat or per-token price, the billing model (flat, metered, or both), who owns the renewal, and the last time anyone measured the workflow's output against a number the CFO would accept. If any of those five come back blank, the tool is sitting where Copilot was on Saturday.

Then have one conversation with the person actually running the workload. Ask the question Mthat iniMax's users were not asked before Sunday. "If the price of this tool went up 10 times next month, would you still use it?" If yes, the workflow has product-market fit, and the budget conversation is winnable. If the answer is "we would have to think about it," the workflow has habit fit, not product fit, and the budget conversation is going to be a hard one.

The Build keynote on Tuesday morning is the first vendor response to the reset. Watch the cost-cap announcement specifically. If caps land cleanly, one tool gets back on stable ground. If caps land vaguely, every Copilot-anchored AI strategy in the market needs a 30-day stress test.

From the field

The pattern I keep seeing in client engagements over the last six months is that the CFO knows the AI line item, the head of engineering knows the workflow, and neither of them knows the unit cost. The seat price was a number both sides could live with, and the real cost sat on the vendor's balance sheet where neither side had to ask. Sunday moved that question onto the buyer's side of the table.

In about half the engagements we have run this year, the team running the AI tool could not tell me how many queries or tokens their team consumed last month. That answer was not their job. Starting this week, it is. The tools that survive the next two quarters will be the ones whose users can answer the question in a meeting.

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.

SINCE FRIDAY

• OpenAI broke ground on Stargate Michigan, a $7 billion, 1-gigawatt data center on a 250-acre Saline Township campus with closed-loop cooling, 2,500-plus construction jobs, and $40 million in Codex credits earmarked for 400,000-plus Michigan students.

• Luma AI opened a Physical AI Science Lab with open-science publication commitments, focused on the generalization problem in robotics. Runway joined as a founding member of the same coalition alongside NVIDIA's Cosmos 3 release.

• NVIDIA's RTX Spark Windows AI PC chip was confirmed at a $599 starting price with native anti-cheat support for Fortnite, Valorant, PUBG, and Denuvo DRM. Eight PC brands committed to fall device shipments.

• Microsoft Build 2026 opens Tuesday June 2 in San Francisco with Satya Nadella's 9:30am PT keynote, the first Microsoft platform statement since the Copilot billing controversy and the RTX Spark announcement landed together.

P.S. If you have already seen a Copilot bill, an Anthropic bill, or any other AI-tool invoice in the last 72 hours that surprised you by an order of magnitude, hit reply and tell me which tool and how big the gap was. I read every reply, and I am keeping a running list to share back anonymously next week.

If a colleague is going into a Tuesday or Wednesday budget meeting where someone is going to ask "is our AI spend safe," forward this issue to them.

IF YOU GOT VALUE FROM THIS ISSUE

Keep going. The next 30 mornings, in your inbox.

The analysis you just read came out of the same operator practice that the 30 Days program teaches. Foundations, then connect, then build, then lead. By Day 30, you own nine concrete capabilities: prompting that hits your bar on the first draft, AI workflows that take five hours a week off your plate, agent skills your team can run, and the language to lead the AI conversation at your company.

Free. Thirty mornings, then it ends. No upsells, no community Slack. Inspired by Hilary Gridley's Couch to 5K for AI.