Open App →

Sold by Label

When one constraint lands with opposite signs on different P&Ls, the market sells the whole basket anyway. Here’s how to re-sort it.

By Steve · October 30, 2026 · 👁 8 · 💬 0

The Customer Becomes the Grid

Hyperscalers are learning to build power plants because they have to. The ones who get good at it will find the plant outgrows the data center, and the contracts they sign now will decide who profits when it does.

By Steve · October 27, 2026 · 👁 13 · 💬 0

Paid in Price, or Paid in Delay

Almost every chipmaker is raising prices this fall. Only some of them have pricing power. The difference tells you where the constraint really sits, and who ends up paying for it.

By Steve · October 22, 2026 · 👁 20 · 💬 0

The Return Doesn’t Ship With the Chip

The buildout is a bet on compute. The payoff is a bet on organizations doing slow, unglamorous work, and that part can’t be bought, only earned.

By Steve · October 19, 2026 · 👁 34 · 💬 0

You Can Buy the Tools. You Can’t Buy the Yield.

Terafab, GlobalFoundries, and the one input in a fab that runs on a clock instead of a budget

By Steve · October 7, 2026 · 👁 26 · 💬 0

The Fuel Isn’t Going Anywhere

We’re planning to quadruple nuclear power on the strength of a storage rule. The back end has to become a gating item, not a cleanup task.

By Steve · October 6, 2026 · 👁 30 · 💬 0

Nobody Drills for Helium

Rare earths get summits and transformers get hearings. Helium has no lobby, no strategic reserve and no supply curve that answers to price. That's why the shortage hitting the AI buildout's memory rung is barely on the board.

By Steve · October 5, 2026 · 👁 40 · 💬 0

The Rung You Can't Streamline

The nuclear deals for AI data centers are arriving on a deal-flow clock. The review capacity that has to clear them runs on a training clock. Those clocks aren't synchronized, and the gap is where safety risk goes to wait.

By Steve · October 3, 2026 · 👁 31 · 💬 0

The Geniuses Were Four Percent

Everyone's debating what a country of geniuses in a datacenter can think. The Manhattan Project already showed us what it costs to build what they think of.

By Steve · September 27, 2026 · 👁 34 · 💬 0

The Speedup That Didn't Ship

Oracle made its engineers dramatically faster, and its customers didn't notice. That isn't an AI failure. It's the oldest lesson in operations, arriving at AI speed.

By Steve · September 24, 2026 · 👁 38 · 💬 0

Three Shortages, One Door

Why the AI supply chain's two feeder shortages, logic and memory, both come due at the same backend step, and why that step is the only number worth watching.

By Steve · September 19, 2026 · 👁 43 · 💬 0

Consent, Not Capex

Why the market traded a consent move as a spending cut

By Steve · September 15, 2026 · 👁 50 · 💬 0