Key Takeaways

  • Cloud operators like CoreWeave, Lambda, and Crusoe cannot build a GPU exchange because their direct hardware ownership invites market manipulation on forward contracts.
  • A functional compute forward curve requires a neutral operator that owns zero physical GPUs and takes no spread on cloud hosting.
  • Model inference platforms like Fireworks, Modal, and Base10 cannot run the trading venue because their internal compute consumption biases order books.
  • External lenders require independent, liquid market-clearing prices before they can lower the cost of capital on data center debt financing.
  • Physical settlement requires deep infrastructure: programmatic data center virtualization, multi-constraint order matching, and subdivided InfiniBand fabrics.

The Forward Market Demands an Operator with Zero Hardware

Evan Conrad founded SF Compute to establish an open trading venue for GPU capacity. A common question from founders is why NVIDIA, CoreWeave, or hyperscalers will not simply launch the exchange themselves.

The barrier is structural. If an exchange operator owns hardware, counterparties cannot trust market clearing prices. As Conrad explains: “We don't own the hardware. The reason why we don't own the hardware is because if we did, we would be able to manipulate the forwards market. So, we have to be Switzerland, we have to be independent.”

Hardware owners face an immediate temptation. They can hold back capacity, front-run spot spikes, or alter forward delivery dates to favor internal cloud margins. A functioning market demands that participants believe the order book reflects unbiased supply and demand. The moment an exchange benefits from price swings, external traders exit.

Lower Financing Costs Require Killing Cloud Margins

NVIDIA gains direct financial upside from a working compute forward curve, even though it cannot build one. Data center builders borrow billions to buy hardware clusters. Debt lenders charge steep interest markups because they lack a reliable price discovery mechanism to predict what a GPU hour will sell for next year.

Liquid forwards solve that problem by generating real market prices that lenders can underwrite. As Conrad notes: “So our impression is that Nvidia can't make it, but benefits tremendously from it existing. The reason why they can't make it is because they would then not be independent enough for people to trade on it.”

Yet building that trust requires giving up high-margin revenue. Incumbent cloud providers generate cash by marking up compute for enterprise inference workloads. Running an exchange forces the operator to walk away from those profits. Conrad summarizes the dilemma: “This is the thing that benefits the entire industry because it makes the financing costs go down. But nobody can make it unless they give up the ability to do all the big inference money and all the big cloud deals. You have to kill your business in order to make this.”

Market Making Fails When You Consume Your Own Supply

The same operational conflict stops inference platforms from launching the marketplace. Conrad points out why software layers cannot be the clearinghouse: “If you don't trade on it, then you don't have enough liquidity for lenders to finance against the forwards market. This is also why Crusoe and Coreweave and Lambda why the hyperscalers can't make it, why Base10 and Fireworks and Modal can't make it cuz they have too much demand on that side.”

Inference platforms run high-volume consumer and developer workloads. If Modal or Fireworks operated an order book, they would face constant incentives to reallocate compute internally during traffic spikes. swyx summarizes the position: “That's why nobody does it. It's like structurally they're not set up to do that. And you're basically reducing friction and that's all your revenue.”

Physical settlement goes beyond software matching. SF Compute must programmatically virtualize remote data centers and subdivide InfiniBand fabrics to route sovereign compute dynamically. The exchange operator must capture thin transaction fees and resist every temptation to enter the cloud hosting market.

What to Do With This

Audit your compute commitments this week to see if you are paying an illiquidity penalty on reserved clusters. If your cloud contracts lack secondary market transfer rights or dynamic subleasing clauses, negotiate shorter billing windows before signing your next GPU agreement.