Ugly Baby Newsletter
June 26, 2026
Morning. The market is still happy to fund AI, but it has started to ask who is paying for the habit.
That is usually the moment the product stops being a demo and starts becoming a tariff.
Useful, if slightly less glamorous. Breakfast rarely respects glamour anyway.
Menu of the day
Markets: picky again
The article: pricing gets serious
Shot: Micron meets Anthropic
Startup Lesson: price the usage
Doggy Bag: receipts beat vibes
What to watch: predict the bill
Market pulse
Quick note: latest session before drafting.
Markets: tech is still the adult in the room, but it is no longer getting free drinks. Micron’s blowout quarter and the lift in memory names say the AI trade is still alive; it just wants infrastructure and cash flow, not slogans.
Names: JPMorgan’s read on retail caution says the easy options-and-margin lever is cooling. That is not a panic signal. It is a reminder that expensive growth stories are being graded on how they behave when the support tape gets thinner.
The article: pricing gets serious
Quick reminder: the most important AI product change this week is not a new model. It is a new sense of price discipline.
OpenAI’s June 24 help-center update says new ChatGPT Business workspaces do not automatically get Codex seats, and the company is leaning harder on credits, shared pools, and usage controls for higher-end features. OpenAI’s API page is already split into per-token pricing, Batch discounts, Flex processing, Priority processing, reserved capacity, and separate enterprise options. Anthropic, meanwhile, just said its compute deal with SpaceX adds more than 300 megawatts and over 220,000 NVIDIA GPUs, while also increasing Claude Code and API limits. Micron then tied its own future more tightly to Anthropic with a strategic agreement that spans memory, storage, supply, and enterprise adoption.
That is the pattern. AI is moving from “can we make it work?” to “how do we meter it without surprising procurement?” The new product surface is not only the model. It is the price sheet, the usage cap, the fallback when capacity is tight, and the audit trail when the buyer gets nervous. If the old software sales question was “does it save time?”, the new one is “can I predict the bill and explain the control model to finance?”
In practice, this is where a lot of AI startups get sloppy. They price for adoption, then discover the adoption comes with bursty usage, support load, and customer anxiety about who can use what. The result is a product that feels cheap in the demo and weirdly expensive in the month-end close. That is not a GTM problem. It is a packaging problem wearing a GTM costume.
The better answer is not always to raise prices. It is to separate the cases. Seats for the steady workflow. Credits or usage for the burst. Reserved capacity for customers who actually need throughput. Controls for the regulated buyers. That sounds bureaucratic because it is bureaucratic. It is also what happens when AI stops being a toy and starts becoming infrastructure.
There is a founder lesson hiding in there that is worth keeping close: the best AI businesses will not just sell model quality. They will sell a calm finance conversation. Whoever can make the bill legible, the limit predictable, and the failure mode boring is going to look a lot more enterprise-ready than the company with the prettier demo. Not exactly a small detail.
Shot: Micron meets Anthropic
Micron’s June 22 agreement with Anthropic is not just a supplier handshake. It links memory and storage design, supply, and Anthropic adoption into one working relationship, which is another way of saying the AI stack is now negotiating with itself all the way down to components.
Shot: Anthropic eyes the window
Anthropic’s confidential S-1 does not mean a public listing is imminent. It does mean Anthropic now has a public-price reference point it can use when talking to customers, talent, and suppliers. Optionality is useful when the market still believes in the category.
Startup Lesson: price the usage
Price the usage before you price the hope.
If a customer can go from casual to expensive in a week, your packaging needs to show that shift clearly. Shared pools, caps, per-seat limits, and reserved capacity are not admin clutter; they are how the buyer trusts the product enough to use it. If the bill is a surprise, the churn will eventually look like a product problem even when it started as a packaging problem.
Doggy Bag
Fun fact: Micron’s Q4 guide is $50 billion in revenue. Memory used to be the boring part. Now it is the part investors keep checking first.
The number: 300 megawatts and 220,000 NVIDIA GPUs. That is how much capacity Anthropic says it just picked up through SpaceX.
The quote: “The price is the product.” Also true: if the controls are unclear, the product is unfinished.
Reco: If your AI buyer asks about credits before features, pay attention. They are not nitpicking. They are telling you where the real budget lives.
What to watch
- Whether more AI products move from simple per-seat pricing into layered usage, credits, and reserved-capacity plans.
- Whether enterprise buyers start demanding auditability and cap logic before they care about benchmark scores.
- Whether the AI infrastructure winners are the ones that can make power, memory, and price feel predictable.