Akash Unveils GPU Forward Contracts for AI Compute, With Q1 2027 Pilot Planned
Key Takeaways
- •Akash announced Akash Forwards at Solana Breakpoint on November 17, 2026, presenting forward contracts built for physical delivery of verified GPU capacity rather than cash settlement against a synthetic index.
- •The published design standardizes contracts as node-months, covering hardware grades including eight-GPU NVIDIA B300 nodes, H200s, and RTX Pro 6000 Blackwells, with each contract specifying a GPU grade, a region, and a delivery month.
- •Prices form through daily open call auctions, and an automated clearing program manages daily settlement in U.S. dollars while sitting between every buyer and seller.
- •A pilot for certified participants is scheduled for Q1 2027 and a wider public launch is targeted for Q2 2027, but the market is not open and no trades have cleared yet.
- •Under the delivery mechanism, Akash validators inspect leases hourly, and if a provider fails to serve, buyers are not charged for missed hours and can finish the month on one node or receive cash at a reference price.

Akash unveiled GPU forward contracts, branded Akash Forwards, on November 17, 2026, at Solana Breakpoint. Founder Greg Osuri outlined a cleared forward market for future GPU capacity in which contracts are built for physical delivery of verified compute rather than serving as a cash bet on a synthetic index. Forwards are a fixture of commodity markets: agreements struck today that fix the price of an asset for delivery on a specific future date, letting producers and buyers plan around a known number instead of chasing a spot price. According to the official BusinessWire release, a pilot for certified participants is set for Q1 2027, with a wider public launch targeted for Q2 2027.
The product's status remains firmly prospective. Akash Compute Forwards is a published market design: the market is not open, and nothing on the design site trades yet.
What Akash Forwards Are Designed to Buy
Most AI labs still secure GPUs the hard way, signing private, bilateral deals. Hardware grades, regions, contract lengths, and prices rarely match across agreements. When supply tightens, weaker buyers get outbid; when plans change, they are left sitting on capacity they no longer need. Investors and lenders face the same fog: without a public forward price tied to deliverable machines, valuing the AI buildout stays guesswork.
Akash Forwards treats one unit as a node-month. Under the published design, a buyer can lock eight-GPU NVIDIA B300 nodes for calendar months ahead, with other grades including H200s and RTX Pro 6000 Blackwells. The price is fixed when the trade clears, and each contract names a GPU grade, a region, and a delivery month, so both sides trade against a common standard.
Eligible providers whose hardware passes Akash verification can sell future capacity before it comes online. Buyers can sell contracts back into the market before delivery if their plans change. Prices form in a daily open call auction — a format in which orders collect and are matched together at a single price rather than trading continuously — and an automated clearing program sits between every buyer and seller, managing daily settlement in U.S. dollars.
How Delivery Is Supposed to Work
Following one filled position through the published timeline, the contract trades at daily calls until shortly before its delivery month, with price changes settling in dollars every day. Right after the last call for that month, the clearing program pairs each remaining buyer with attested nodes and records a one-month lease.
Every hour of the month is checked. Akash validators inspect each lease on an hourly cadence. Providers are paid for the hours they serve; if a provider stops serving, the buyer does not pay for the miss, and the system then looks for replacement compute through the spot path. On a longer outage, the design offers two exits: a buyer may finish the month on one node, or take the rest of the month in cash at a reference price.
That is the physical-delivery claim in one line: buyers receive attested GPU hours for the month they bought, or a cash payment covering the gap when delivery fails. The chain's job, as described, is lease accounting and hourly checks — not proof that a model trained on those GPUs was correct.
Why It Matters for AI Compute Pricing
Cheng Wang, chief financial officer and chief operating officer of Overclock Labs, the creator of Akash Network — a decentralized marketplace for cloud compute — said in the release that three U.S. exchanges have already announced compute futures in 2026. He argued that settlement trust remains the open question for those contracts, and that bilateral GPU deals hide the prices buyers and sellers need. In the project's framing, Akash Forwards answers that problem the way power and gas markets did: every contract delivers real GPU-hours to a defined standard, and every recorded price is one somebody paid.
Osuri put the buyer's pain more simply. Unless a lab is among the best funded, he said, it cannot see what compute really costs. It gets outbid when capacity is tight and overbuys to stay safe, and that uncertainty climbs to investors trying to value the AI buildout. In commodity markets, the forward curve — the run of quoted prices across successive delivery months — is the reference traders use to see how the same asset is priced for different dates. A public forward curve for standardized node-months is the core bet: AI and blockchain rails would turn opaque capacity into a commodity-style price signal. The concept sits alongside other experiments that turn capital into a claim on compute, including 0G's path from staked assets into AI compute credits.
What Is Demonstrated, and What Remains a Claim
What the primary record shows today is an unveil, not a live order book. Osuri announced Akash Forwards at Breakpoint, and the company published a white paper and design site at forwards.akash.network. The release names the hardware grades and the month-long physical unit, as well as the daily auction, dollar clearing, the Q1 2027 certified pilot, and the Q2 2027 public target. Simulated terminal prices on the design site are marked as illustrative.
The harder part is what remains project claim. The pilot has not cleared a real trade, and default protection, certification agents, guarantors, and assignment rules are design layers rather than a proven market with depth. Hourly validator checks can attest that a lease ran, but they do not prove that replacement capacity will always be available at a fair price. Wang's comparison to power markets is an analogy for how forward prices can finance supply, not evidence that GPU forwards will attract the same liquidity.
The mental model is simple: Akash GPU forward contracts attempt to put a public, physical-delivery price on future AI capacity, available before anyone has to guess from private deals. Whether the Q1 2027 pilot clears real node-months at verified grades, whether those prices stay visible after assignment, and whether buyers treat the curve as a hedge rather than a brochure will determine whether the design becomes a working market.