GPU forward pricing as a 1m / 3m / 6m / 12m ladder across B200, H200, H100, and A100.

Two data conventions on one ladder

Most buyers who think about forward prices for compute start with the prediction-market signal — Kalshi-listed GPU contracts — and stop there. That signal is useful but not enough: prediction markets imply a market-clearing rate, not a rate any provider will actually quote you. We pair the Kalshi-derived mid with provider-native forward quotes (hourly rates a provider publishes for a 1m, 3m, 6m, or 12m commit) so the same ladder shows both signals.

Both layers render under the same explicit as-of stamp. Provider rows only appear if a row in provider_forward_pricing has verified_at set; otherwise the cell reads data pending instead of inventing a number. We treat invented forward prices as the single most expensive category of error in long-horizon procurement modeling.