AI Compute Futures: The Financialization of a Non-Fungible Resource

The CFTC’s public docket for AI compute futures is not just a regulatory formality. It’s a quiet admission that compute has become a strategic resource, and the market needs a price discovery mechanism that goes beyond the opaque contracts of cloud providers. I’ve spent the last decade watching liquidity flows, from ICOs to DeFi to Terra, and every time the market tries to financialize a non-standardized asset, the same pattern emerges: a rush to launch, a scramble for liquidity, and a quiet retreat when the index fails to capture reality. CME’s target launch for AI compute futures is October 2025, according to the CFTC’s request for public input. The product is still in the regulatory sandbox, but the implications reach far beyond the exchange floor. If approved, this would be the first standardized financial instrument pegged to the cost of AI computing power—a derivative of GPUs, cloud services, and the energy that powers them. The question is not whether the market needs this product. The question is whether the market can agree on what “compute” even means. Let’s start with the basics. The CFTC is seeking public input on whether to classify AI compute as a commodity under the Commodity Exchange Act. This is a structural pivot. If the CFTC designates compute as a commodity, it opens the door for a whole new asset class: compute ETFs, compute swaps, compute options. The “commoditization of compute” is a phrase thrown around by VCs, but the legal reality is that without a regulatory definition, no financial product can survive. The CFTC’s move is the first step in turning compute from a negotiated service into a traded asset. But here’s the problem I identified in my 2017 ICO analysis: liquidity mining APY is essentially the project subsidizing TVL numbers. Stop the incentives and real users vanish. The same logic applies to compute futures. The underlying asset—AI compute—is not a homogenous barrel of oil. It’s a multidimensional bundle of GPU type, memory bandwidth, latency, location, and energy cost. An H100 chip in Texas with cheap renewable energy is not the same asset as an H100 in Tokyo with high power costs. Yet the futures contract will treat them as interchangeable. That’s a recipe for basis risk. I’ve seen this before. In 2020, I analyzed the composability trap in DeFi. Aave and Compound looked like separate markets, but when ETH dropped below $200, the liquidation cascades hit both simultaneously. The correlation was hidden until the stress test. For AI compute futures, the composability risk lies in the index construction. If the index is based on a basket of data sources—say, AWS, Azure, GCP, and a few independent GPU aggregators—the correlation between those sources during a sell-off could break the index. If Amazon slashes prices to win a hyperscaler deal, the index might drop, forcing margin calls on hedgers who are long compute. But the real spot price for smaller players might not have moved. The index becomes a fiction. Algorithms don’t fail; models do. The model for AI compute pricing is still primitive. Today, the price of compute is determined by bilateral negotiations between cloud providers and enterprises. There is no transparent spot market. The futures contract will create a synthetic spot price, but that price will only be as trustworthy as the data feeding it. If the CFTC requires the index to be based on reported transaction data from a diverse set of market participants, the index will be robust. But if the index is based on quotes from a few large players, it will be vulnerable to manipulation. The 2022 Terra collapse taught us that a synthetic price mechanism can hold for a while, but when the underlying collapses, the peg breaks fast. This brings me to the contrarian angle. The market narrative is that AI compute futures will unlock hedging for hyperscalers and AI labs, reduce volatility, and create a new asset class. But I see a different future: the product may fail to attract real hedgers and instead become a speculative playground for macro funds. The reason is simple: the entities that need to hedge their compute exposure—the hyperscalers—are also the ones that control the index data. They have no incentive to create a transparent price that might reveal their cost advantage. They will either stay out of the market or use the futures to further entrench their pricing power. The real beneficiaries may be the arbitrage funds that trade the basis between the futures and the illiquid spot market, not the AI companies that need cost certainty. Composability is a double-edged sword. In the DeFi world, composability created the liquidity crisis of 2020. In the compute futures world, the index composability with other macro assets—like NVIDIA stock, energy futures, or even carbon credits—could create unforeseen contagion channels. If a hedge fund is long AI compute futures and short NVIDIA stock, a sudden tariff on semiconductors could cause both legs to move in unexpected ways. The model that works in calm markets will fail in volatility. That’s the lesson from every financial innovation: the correlations are not stable. I’ve been tracking the institutional maturation of crypto since the 2024 ETF inflows. The pattern is clear: retail exits, institutions enter, volatility drops. The same could happen with AI compute futures. But the difference is that crypto has a relatively standardized unit—a Bitcoin is a Bitcoin. AI compute does not. The institutional players will demand a high-quality index, and if the index is opaque, they will stay away. The launch will be a slow burn, not a boom. Now, let me embed my own experience. When I analyzed the 2017 ICO bubble, I modeled the liquidity flows of 50+ Ethereum ICOs. I found that projects with high-TVL but low user retention were the first to crash. The same will happen here: the futures contract with high open interest but low physical delivery will be the first to fail. The CFTC’s public input is a signal that they are aware of this risk. They are asking the right questions: Who will provide the data? How will the index be audited? What happens if the underlying GPU market becomes concentrated? The answers will determine the product’s survival. I also recall my work on the DeFi composability trap. In 2020, I wrote a controversial piece predicting a liquidity crunch if ETH dropped below $200. The market ignored me until it happened. With AI compute futures, the hidden risk is the double-counting of capacity. If a data center is used to back both a futures contract and a spot lease, the capacity is being over-allocated. This is not a problem if the contract is cash-settled. But if the CFTC allows physical delivery, the capacity constraints could cause a squeeze. The market will need to decide: is this a financial instrument or a commodity? The answer will shape the design. Let me shift to the macro context. The Federal Reserve’s rate cycle is turning. Lower rates reduce the cost of carry for futures, making it cheaper to hedge. But the real macro story is the export controls. The US has restricted the sale of high-end GPUs to China. This creates a bifurcated market: restricted compute and non-restricted compute. The futures contract will likely be pegged to unrestricted compute, but the price will be influenced by the black market premium for restricted compute. This is a new kind of basis risk, one that involves geopolitical sanctions. The CFTC will have to decide whether to include Chinese data centers in the index. If they exclude China, the index is incomplete. If they include China, the index is subject to sanctions risk. Either way, the index design is inherently political. The bubble burst, the lessons remain. The ICO bubble taught us that narrative-driven assets without economic moats collapse. The DeFi lending crisis taught us that hidden correlations cause systemic failures. The Terra collapse taught us that algorithmic pegs are fragile. AI compute futures are a new iteration of the same pattern. The product is a solution to a real problem—price discovery for compute—but the solution is only as good as the index. If the index is built on a few data sources, it will be captured. If it is built on a diverse, transparent, and auditable set of transactions, it will succeed. Here is my takeaway: The real signal to watch is not the CFTC approval date, but the composition of the index committee. If the committee includes only AWS, Azure, and GCP, the futures will be a tool for the oligopoly, not for the market. If it includes independent data centers, GPU aggregators, and even retail miners, we might have a true benchmark. The launch date is a distraction. The index design is the story. Cross-border payments are evolving, and so is the financialization of compute. The same infrastructure that allows stablecoins to move value across borders will eventually allow compute futures to move capacity across geographies. But that is a 2030 vision, not a 2025 reality. For now, the market is in the regulatory sandbox, and the most important experiment is not the futures contract itself, but the process of defining what compute is worth. That process will determine whether this product becomes the next oil futures or the next Tulip. I’ll be watching the CFTC docket. The public comment period is the first test. If the feedback is dominated by exchange and bank interests, the index will be compromised. If independent data providers and AI startups speak up, the index might be fair. The market is waiting for direction, but the direction will be set by the data providers, not the algorithm. And as I’ve learned from 27 years in this industry, the data is always the first casualty of financialization.