The Race to List Compute Futures ETFs
Six ETF issuers have filed preliminary prospectuses for funds that will track GPU compute futures. These are among the first ETFs in history to be structured before the underlying futures contracts even begin trading. That's not a footnote — it's a signal about how fast the AI compute market is moving.
This post breaks down what compute futures ETFs are, how they work mechanically, and what the different design choices mean for investors, hedgers, and the broader GPU market.
Futures ETFs vs. stock ETFs
Most people are familiar with equity ETFs — funds like SPY or QQQ that hold baskets of stocks. These stay aligned with their benchmark through a mechanism called creation and redemption: when demand for the ETF rises, authorized participants (APs) create new shares by delivering the underlying stocks to the fund. When demand falls, they redeem shares and receive the stocks back.
Futures ETFs work differently. You can't practically deliver futures contracts the way you can deliver shares of Apple — doing so would require transferring positions between futures clearing firms (FCMs), which is operationally complex and impractical. So futures ETFs use cash creation and redemption instead: APs put in cash, and the fund's advisor uses that cash to buy futures or swaps to track the index.
This structural difference matters for how the fund operates and how closely it tracks its target.
CFTC-regulated commodity pools
Futures ETFs aren't just investment funds — they're commodity pools, operated by commodity pool operators (CPOs). This is a specific legal structure under CFTC regulation that allows a pool to accept cash from outside investors and deploy it into futures or derivatives markets.
The commodity pool structure gives the fund's advisor meaningful discretion. Unlike a passive equity ETF constrained to replicate an index, an actively managed commodity pool advisor can choose how and where to invest the pool's assets — selecting venues, instruments, and exposures — as long as the result tracks the fund's stated investment objective.
Not naming specific exchanges
If you read a compute futures ETF prospectus, you'll notice that they deliberately don't name the exchanges where they'll trade. This isn't an oversight. It's a feature.
The compute futures market is nascent. Contracts are being developed across multiple venues, liquidity is still forming, and the competitive landscape between exchanges is unsettled. By preserving flexibility in the prospectus, advisors retain the ability to route to wherever the best contracts and deepest liquidity emerge — without needing to file an amendment every time the landscape shifts.
Five ways to build a compute ETF basket
Beyond the choice of trading venue, compute ETF issuers have significant latitude in how they construct the portfolio itself. Current filings show five distinct approaches:
- Front-month, single model. Track only the nearest-expiry futures for one GPU model (for example, the H100). Simple, concentrated, and maximally sensitive to near-term compute pricing.
- Front-month, multi-model. Track the nearest-expiry contracts across multiple GPU generations (for example, the H100 and B200 side by side). This provides broader coverage of the GPU market without taking a view on the forward curve.
- Weighted basket across the expiry curve. Rather than just the front month, hold positions across multiple expiries for one or more GPU models. This is more complex, but it can better reflect the economics of compute procurement, which often happens on longer-term contracts.
- Compute within a diversified commodities basket. Add GPU compute alongside more established commodity futures such as energy, metals, and agricultural products. This positions compute as a commodity class in its own right and gives investors diversified exposure.
- AI-themed multi-asset baskets. The most expansive design: combine compute futures with power contracts (relevant because data centers are major electricity consumers) and GPU-related equities (chipmakers, data center REITs, and the like). This creates a thematic AI infrastructure basket rather than a pure compute play.
Each approach implies a different view on what "compute exposure" means — and who the fund is designed for.
Who will actually use these funds?
There are two distinct buyer groups, and understanding them depends on how these markets develop.
Natural hedgers are the companies for whom GPU compute is a real input cost: hyperscalers, AI labs, and enterprises building on rented infrastructure. For them, compute futures are a risk management tool — similar to how airlines use jet fuel futures. Being able to lock in compute costs months in advance is genuinely valuable for financial planning and budgeting.
Speculators and investors are a second, probably larger, category. The sustained demand for AI equities — from retail investors buying Nvidia to institutions taking positions in private AI companies — suggests that many market participants simply want exposure to the AI infrastructure buildout. Compute futures ETFs offer a new, more direct way to express that view.
The presence of both groups matters for market structure. Hedgers provide natural two-sided flow; speculators provide liquidity. A healthy futures market needs both.
Why this is historically unusual
It's worth pausing on the timeline here. Normally, a futures ETF comes to market well after an active futures market already exists. Bitcoin ETFs came years after Bitcoin futures were trading. Oil ETFs launched into a futures market that had existed for decades.
Compute futures ETFs are being filed before the underlying futures are actively trading. That's a bet on the market coming — and a signal that ETF issuers see enough demand from both the hedger and investor communities to build the product infrastructure in advance.
How these first funds perform — how well they track their benchmarks, how efficiently they roll contracts, and how they handle illiquid periods — will shape the microstructure of GPU compute pricing for years to come.
They're not just financial products. They're infrastructure for a new commodity market.
We hope to see the first compute ETFs launch in Q3, and we expect the American Innovation Exchange to play a key role.