What "forward GPU pricing" actually means

A forward price is a price quoted today for delivery (or settlement) at a future date. In oil, the NYMEX quotes WTI for delivery every month for the next several years. In grain, the CBOT does the same for corn and wheat. In compute — which is what this post is about — a forward price answers a simple question: at today's spot price, what does the market believe a GPU hour will cost in 30, 90, 180, or 365 days?

The mechanism is the same in every market. Someone willing to take the other side of a risk (a "market maker" or "speculator") commits to a price for a future delivery. The price they quote is shaped by what they believe supply, demand, and inventory will look like at that future date. If many market makers believe compute will be cheaper six months from now, the 6m forward will sit below the 1m forward. If they believe it will be more expensive, the 6m forward will sit above.

Definition (one sentence)

A GPU forward price is the market's quoted price today for one GPU-hour of compute to be rented at a specific future date — typically quoted as a ladder of tenors (1m / 3m / 6m / 12m) so buyers can see the full forward curve at a glance. The ladder's shape (rising, falling, or flat) tells you whether the market expects supply to be tighter or looser over the next 12 months.

Until recently, AI infrastructure buyers had no real forward signal — they procured on spot alone. Kalshi Prediction Market now hosts event contracts that let speculators express views on GPU pricing, which produces the first credible market-implied forward signal for compute. GridStackHub's role is to surface this curve alongside the live spot dataset so infrastructure teams have both numbers in one place. We are not a broker, exchange, FCM, or trading venue — we are a price intelligence site.

The Kalshi-implied ladder: 1m / 3m / 6m / 12m

The forward curve is almost always shown as a ladder — a small set of dated tenors the market agrees to quote. The horizons we use are:

Tenor Days out What it answers Why it matters
1m 30 What will a GPU-hour cost me next month? Short-cycle procurement — burn-rate planning, near-term budget sign-off.
3m 90 What does a one-quarter reservation approximate? Quarterly commit cycles — the most commonly used tenor in vendor pricing.
6m 180 Where does the model put the mid-H2 floor? Mid-horizon — sets expectations for H2 2026 fiscal planning.
12m 365 What does the market think a 12-month reservation should cost me? Annual commit math — the bottom-line number for a fixed-price contract.

The shape of the ladder matters as much as the absolute numbers. Three shapes procurement teams care about:

  • Contango — the curve rises with tenor. The market expects tighter supply or higher demand in the future, and a 12m contract will cost you more than 12 months of rolling 1m spot. The current shape of the B200 ladder is a partial contango — see the live curve.
  • Backwardation — the curve falls with tenor. The market expects looser supply or softer demand, and forward contracts get cheaper as tenor extends. This is what the A100 ladder has looked like for most of 2026.
  • Flat — every tenor cluster around a similar price. The market is balanced and has no strong view on the next 12 months. The H100 ladder is currently in this regime.

The numbers shown on /gpu-forward-curve are the Kalshi mid quote for each tenor — the average of bid and ask across the most recent event-contract markets that settle on that date. When the Kalshi feed is unreachable from the sandbox, we substitute a 60-day GridStackHub spot trend as a transparent fallback (you'll see a forecast_proxy tag on each cell) so the ladder stays readable rather than disappearing.

How to read the 1m vs 12m spread: if the 1m cell is $4.29/hr and the 12m cell is $5.15/hr, the spread is +20% — the market expects B200 scarcity twelve months out. If the 12m cell is $3.95/hr, the spread is -8% — the market expects B200 oversupply. The spread is a procurement signal, not a price target. A flat curve tells you locking in the same horizon costs roughly the same as rolling 1m spot; a steep contango tells you reservations are expensive relative to spot but worth it if you need supply certainty.

Provider-native forward rows layer on top of the Kalshi mid

The Kalshi ladder is a market signal — it tells you what speculators believe, not what providers actually offer. Procurement teams need both. That's why we run a second layer above the Kalshi mid called provider-native forward rows: independently verified reservation rates from each provider, structured at the same 1m / 3m / 6m / 12m tenors.

The data convention (anti-hallucination protocol)

Every cell in the provider-native layer has one of three states:

  • Kalshi-derived — the ladder mid for that tenor; carries an explicit as-of stamp and source tag.
  • Provider-native — a row in our provider_forward_pricing table that has its verified_at field set by an independent paste-up edit; carries the verified date and the source URL the row was sourced from.
  • Data pending — explicit placeholder rendered when a row has not been independently verified. We do not interpolate or fabricate a number here; invented forward prices are the most expensive category of error in long-horizon procurement modeling.

Paste-up writes go through POST /api/forward-pricing/admin/upsert with the provider-forward-pricing key; unverified rows simply do not render. The full matrix, including the live basis between the Kalshi mid and each verified provider row, is available at /gpu-forward-pricing as the live ladder for procurement.

The reason this matters for buyers: the market-implied forward and the provider-quoted reservation can disagree — sometimes by a lot. A high-quality basis opportunity is exactly that disagreement: a provider giving you a 6m rate materially below the Kalshi 6m mid. /gpu-forward-pricing shows both layers side by side so you can compute the basis yourself.

Why procurement teams should care this quarter

Pulling the three signals together — Kalshi mid, provider-native reservation, and live spot — is the procurement math for Q3 2026. The picture is meaningfully different for each GPU family:

  • B200 floor softening — into a $4.05–$4.25 plateau The lowest Blackwell spot floor dropped from $4.65 in late June to $4.49 mid-July and now $4.29 on CoreWeave, inside the $4.10–$4.30 forecast band we published in the Week 30 GPU Pulse. Softening pace is flattening toward a stable phase; the 30-day outlook stabilizes at $4.05–$4.25 by mid-August. If your team is timing a B200 commitment, the window toward that stable phase is closing now.
  • H100 $1.49 holds 7 weeks flat — equilibrium, not a floor The H100 spot floor at $1.49/hr on Vast.ai has held for 7 consecutive weeks — the longest uninterrupted flat run we've tracked all year. Forward shape is correspondingly flat. This is equilibrium pricing, not a softening baseline; locking a 12-month H100 contract at $1.49 today is not a discount, it's the market clearing rate.
  • H200 cooling — Blackwell transition is real H200 dropped to $2.65/hr this week (-7.0% WoW). B200 supply growth is visibly eroding H200's premium positioning. The 30-day forward curve for H200 is the cleanest backwardation on the ladder; if you're timing an H200 commitment, the market is signaling cheaper forward tenors.
Procurement takeaway for Q3: the B200 / H200 forward curve is in mild contango with softening trend — if your team is buying Blackwell, the next 30 days are likely as cheap as the curve offers; reservations done now will look reasonable six months from now only if you actually need supply certainty. The H100 curve is flat — locking at $1.49 is a clarity buy, not a savings buy. The H200 backwardation is the cleanest case for waiting 90 days on commit timing. See the Week 30 GPU Pulse for the full underlying baseline numbers and the Pro pricing page for tooling that automates this math against your own workload.

Run the math on your own workload

GridStackHub's calculator, optimizer, and reserved advisor all read from the same spot + forward dataset this post describes.

🧮 GPU Cost Calculator — Model any GPU × workload × provider combination against the live dataset.
📦 Reserved Instance Advisor — Break-even analysis for H100, H200, and B200 commitments.
📊 GPU Forward Curves — Live Kalshi mid quotes across the 1m / 3m / 6m / 12m ladder.
🪜 GPU Forward Pricing — Provider-native reservation rows layered on top of the Kalshi ladder.
📰 Weekly GPU Cost Pulse — The Week 30 baseline this explainer cites (B200 $4.29, H100 $1.49, H200 $2.65).
💼 Pro Pricing — Alerts, forecasts, and forward-curve tooling for procurement teams.