Tom Traugott, SVP of Emerging Technologies at EdgeCore Digital Infrastructure.getty​The artificial intelligence buildout is on track to cross $1 trillion in annual investment by 2027, with no credible sign of contraction. The record-setting growth of frontier labs like Anthropic and OpenAI demonstrates that durable demand and pricing across the compute stack continues to increase. Spot rates for the newest accelerators are also climbing—Nvidia Blackwell pricing on the Ornn Compute Price Index rose 48% between mid-February and mid-April 2026, from $2.75 to $4.08 per GPU-hour—while older GPUs remain in production at firm prices and hyperscale clusters change hands at premiums through landmark deals, including SpaceX, Anthropic and Google. The strongest counterargument to the "AI bubble" thesis is simple: Customers are paying more, not less, for constrained supply.​Scarcity And PaceScarcity is the defining condition of the moment. Goldman Sachs forecasts that token consumption will multiply 24 times by 2030, reaching 120 quadrillion tokens per month, driven by agentic adoption across consumer and enterprise markets. Against that curve sits a hard physical reality: Chip fabrication, power interconnection and data center construction may struggle to scale at the same pace. The industry's response has been to treat every bottleneck as an investment opportunity—following the money toward the constraints and funding their relief.This is where the financing structure itself becomes the bottleneck worth examining. Today, funding for compute infrastructure, including data centers, power and the silicon inside them, looks remarkably like the natural gas market of the early 1980s. Capital flows through long-term, bilateral agreements, whose bankability rests on the corporate credit of a single off-taker and the appetite of a limited pool of lenders. The dollars available for construction are effectively capped by the contracts that can be signed, carrying a corrosive implication: Demand is assumed to vanish the moment a contract expires. Infrastructure underwritten this way must be pre-sold, project by project, relationship by relationship.Compute Architecture​History suggests this is transitional. When demand is sufficient and durable, American capital markets have repeatedly found ways to meet it—from risk-on venture capital at the frontier to the patient construction of commodity markets that bring liquidity to physically scarce resources. Natural gas made precisely this journey. Consider its physical stack: Gas molecules are the consumed commodity, but they only became tradable once the exploration rigs, transmission pipelines and storage fields existed to produce, move and hold them—and once a reference point like Henry Hub allowed the market to price the molecule independently of any single well or buyer. Compute has the same architecture. Tokens and GPU-hours are the molecules; data centers, power interconnection and the accelerators are the rigs, pipelines and storage—the physical plant that makes the commodity deliverable. Oil, gas and metals share a feature compute currently lacks: a visible, tradable market price that signals when new physical supply is justified.Compute is now approaching that threshold. As demand grows more certain and key metrics—token throughput, GPU-hours, teraflops—become standardized, the molecule can be priced independently of the pipeline. Compute can be reconceived not as merely contracted demand inside fixed-term agreements, but as a durable, ever-present consumption stream akin to the baseload draw of gas. Just as a Henry Hub price allows a producer to justify a well without a signed buyer, a liquid compute price allows a developer to justify a data center against market demand rather than a single tenant. That reframing is the precondition for everything that follows.​The Shift AheadThe most consequential opportunity lies in the paper-to-physical ratio of mature commodity markets. In oil and gas, the notional value of contracts traded routinely exceeds the physical commodity consumed by 10 times or more. That multiple is not a distortion; it is liquidity. It brings far more capital into contact with the resource than physical consumption alone ever could—dampening price spikes, distributing risk and signaling that demand is durable enough to justify continued investment. The financial layer becomes the mechanism that finances the physical plant before the offtake exists.For the data center market, this could mark a profound shift. Developers could build ahead of a signed agreement, with capital backstopped not by a single tenant's credit but by a pool of liquidity, drawing confidence from broader market demand. A contract could be written against a synthetic reference price—tradable, assignable and later allocable to one or several off-takers as the market clears. Speculative development, today financed only through pre-leasing, becomes under writable against a market price.This is no longer theoretical. ICE and Ornn plan to launch GPU compute futures based on Ornn's Compute Price Index, which tracks live-traded spot prices across major hardware types, while CME Group and Silicon Data, backed by trading firm DRW, have partnered to launch the first compute futures. The Compute Exchange operates a transparent spot venue, Architect Financial Technologies has launched perpetual futures on GPU and DRAM prices, and SemiAnalysis, Compute Desk and OneChronos round out a fast-maturing ecosystem. DRW founder Don Wilson put it plainly: "It has been clear to me for some time that compute will become the largest commodity in the world."Complications remain. Compute is ephemeral, and true fungibility across operating platforms remains unresolved, with real qualitative and performance differences demanding better benchmarking. Thin early liquidity, limited participation and the concentrated influence of a few chipmakers remain open questions, but the direction is unmistakable. The same machinery that turned natural gas from a handshake into a globally liquid market is now being built for compute—and with it, the prospect that trillions in AI infrastructure can be funded not by contracts alone, but by dynamic markets.​Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?