Two Tides Research

Spot Premium: What On-Demand GPU Capacity Really Charges

Google's TPU rate card prices identical silicon at $12, $8.40 and $5.40 per chip-hour. The 2.2x gap is what urgency costs, and why it cannot last.

By Marco Wade · 31 October 2026 · 4 min read

Spot premium: the difference between the on-demand rate for capacity and the committed rate for the same hardware over one or three years, paid for flexibility rather than for compute.

Google publishes the only clean price ladder in this market, because TPUs are sold publicly at three commitment lengths. Ironwood, its current flagship, lists at $12 per chip-hour on demand, $8.40 on a one-year commitment and $5.40 on three years. Identical silicon, one price list.

Commitment	Per chip-hour	Multiple of the 3-year rate
On demand	$12.00	2.2×
One year	$8.40	1.6×
Three years	$5.40	1.0×

A 55% discount for signing three years is the market's own estimate of what commitment is worth. Read the other way, the premium is the price of not committing: the option to stop, to switch supplier, or to discover you were wrong.

What the premium buys at the top of the market

The SpaceX–Anthropic lease runs to $1.25 billion a month for more than 300 megawatts, which annualizes to roughly $50 billion per gigawatt-year — four to six times what long-contract capacity rents for. Either side can walk on ninety days' notice.

Nobody pays four to six times the committed rate for tokens they could buy more cheaply on a three-year contract. They pay it because they need capacity this quarter and their own buildings will not exist for a year. The premium is a schedule, not a yield: it prices the gap between when demand arrives and when supply does.

Why that matters to a buyer

If you are pricing a project off spot rates, you are pricing it off a number that is meant to close. The durable comparison is the long-contract rate, and the durable risk is that by the time your volumes arrive, the premium has gone and your unit economics were built on it.

The full treatment, with the contracts and the interviews behind these numbers, runs in the letter.

Read it in Two Tides

Free to read. Paid tiers open 1 December.

Sources and method. Definitions follow how these terms are used in capacity contracts and in company disclosures, not textbook usage. Figures are from the sources named on the flagship data page.

Research, not advice. This page is research and commentary on an industry. It is not investment, legal or engineering advice, and not a recommendation to buy or sell any security. Figures are the best estimate available at the time of writing; where one rests on a single source, the text says so.

Cite this page as: Quorsio, "Spot Premium: What On-Demand GPU Capacity Really Charges", November 2026, with a link to this URL. Short excerpts with attribution are welcome. Republishing the page in full, including its tables and charts, needs written permission: hello@quorsio.com.