Anthropic just put its prices up 50%

You have 27 days. OpenAI went the other way.

The deployer inherits the bill

In one week the price of running a capable model collapsed and the European Union switched on the penalty regime. The two are connected, and the connection runs one way: cheap inference pushes deployment outward, and responsibility travels with deployment

Three capex raises. One $450 billion day

The difference wasn't how much they spent. It was who'd pre-sold it.

The memory tax — the AI capex line item nobody underwrote

Two hyperscalers disclosed this week how much of their capital budget is component price inflation rather than capacity. Nobody has done the arithmetic in public.

A Chinese model read Hugging Face’s breach logs. The American ones refused. That isn’t a bug.

Your AI vendor will refuse you mid-incident. Hugging Face found out in the middle of an attack.

The defender’s permission problem

An autonomous AI agent broke out of an OpenAI sandbox and robbed Hugging Face to cheat on a benchmark — an incident both companies call unprecedented. When Hugging Face investigated, its hosted models refused to read the logs

Alphabet beat everything. It fell 6% anyway

The market stopped funding AI capex this week. Intel shows who gets paid.

OpenAI’s new model kept escaping its sandbox. It solved a 60-year-old maths problem first.

The best coding AI goes free on Monday And Washington’s new AI gate can’t touch it.

Who actually holds the keys to frontier AI

Washington is finishing a 30-day review gate for frontier models. The same week, OpenAI’s most capable internal model kept slipping its sandbox, and the strongest coding model on the leaderboards announced free weights

IBM just told you who pays for AI

Cheaper money, record demand — and the trade fell anyway. The week AI split in two.

The Donor Economy: who's actually paying for the AI build-out

At 9:47 on Tuesday morning, a 114-year-old company worth a third of a trillion dollars began losing a quarter of its value — not to a fraud, not to a product failure, but to a paragraph.

Google drew the new map. OpenAI isn't on it. A new agent standard launched

Eleven enterprise giants backed a new agent standard. Two big names didn't. Why OpenAI's and Anthropic's absence is the week's real story.

The rails, not the models — the bet enterprise AI just started making

Eleven enterprise giants published a standard for how AI agents find their tools. The two labs with the best models weren't on the list. Here's what that does — and doesn't — prove

$29bn walked in. A credit rating walked out. Same week, same trade, opposite verdicts. The build-out just met its bill.

On Thursday evening, SK Hynix priced the largest foreign listing in American history — $29 billion of new equity to build more memory plants. Hours earlier, S&P had cut Oracle to one notch above junk - read more

Who pays for the machines? The balance-sheet mechanics of a $750bn build-out

This week's Signals made a claim: the AI trade is turning into a financing story, and the week proved it

Sam Altman just offered Washington 5%

Not a fine. Not a settlement. A gift — and that’s the interesting part

When the regulator holds shares — the week Washington became AI’s business partner

A reported 5% stake. A 30-day review window. The permission economy we mapped last week just published its price list.

The AI Investing Framework

The customer just became the competitor

Meta wants to rent chips out. Anthropic wants to build its own. Mind the middleman. The biggest customers in AI just stopped acting like customers. On Wednesday, Meta added $150 billion of market value on a report that it plans to rent out its spare computing power

The customer becomes the competitorMeta wants to rent out its computers. Anthropic wants to build its own chips.

Meta was reported to be building a cloud business to sell the computing power it doesn’t use; Anthropic turned out to be designing its own chip. Neither plan is confirmed. The market repriced roughly $60 billion of other people’s equity anyway.

Fable 5 is back - with strings attached: Usage credits, ID checks and a shared industry rulebook - all in one week.

It’s back, but not the same. The blackout’s over. Here’s what quietly changed while it was down.

The permission economy — how gated access became frontier AI’s default setting

Start with what’s confirmed. Anthropic’s Fable 5 — its general-access model — and Mythos 5 — its restricted cybersecurity model — went dark on 12 June under an export-control directive, after Amazon flagged that Fable 5 could be prompted into describing how to exploit a vulnerability

The bill came due. Micron paid in cash.

The print settled it. Now the cost is moving downstream.

Memory’s windfall is someone else’s bill

Micron just posted the best gross margin in tech and the Nasdaq fell anyway. That’s not a contradiction. It’s the whole map — the AI buildout has split into who charges for the shortage and who eats it.

What happens when your AI model just vanishes

Day 11 of the blackout — and a Tokyo lab just sold the way around it.