The billionaire who once sold 98% of his bitcoin now thinks GPU futures will repeat crypto's speculative fever — just with better fundamentals underneath.
The Summary
- Mark Cuban predicts chips will become "the next crypto" as CME Group launches GPU futures trading October 5
- Cuban's crypto track record is a graveyard of bad calls — he dumped 98% of his bitcoin and once said he'd rather have bananas
- This time he might be right: compute is the oil of the AI economy, and now it has a derivatives market
The Signal
Mark Cuban says chips as an asset class will be the new crypto, and for once, the man who's been wrong about digital assets more times than he's been right might be onto something. CME Group is listing GPU futures on October 5, creating a financial instrument that lets traders bet on the price of compute power the way they bet on corn, crude oil, or bitcoin.
The timing matters. AI compute is the most constrained resource in the tech economy. Every major company from Meta to xAI is burning billions on GPUs. Nvidia can't manufacture them fast enough. And now there's a way to trade exposure to that scarcity without actually buying a server rack.
"Chips as an asset class will be the new crypto."
But Cuban's prediction raises an obvious question: why should anyone trust his judgment on speculative assets? This is the investor who:
- Sold 98% of his bitcoin holdings
- Said he'd prefer bananas to BTC as a store of value
- Watched his preferred "bananas" play age out while bitcoin rallied past $100K
The difference this time is that GPU futures have something bitcoin never did in its early days: immediate, measurable utility. A bitcoin in 2013 was worth whatever the next buyer would pay. A GPU in 2026 is worth the compute hours it produces for training frontier AI models. That's not speculation. That's a commodity with industrial demand.
The mechanics:
- CME's GPU futures let investors trade compute capacity without physical delivery
- Pricing will likely track Nvidia H100/H200 benchmark rates
- Market will create price discovery for AI infrastructure costs
The agent economy runs on compute. Every autonomous AI agent, every real-time personalization engine, every voice-to-code tool — they all consume GPU cycles. If Cuban's right and chips become a tradeable asset class like oil, it means compute pricing gets more transparent, more liquid, and more accessible to companies that can't afford to build their own data centers.
It also means speculation. Crypto's speculative mania wasn't a bug, it was the feature that brought capital and attention. GPU futures could do the same for compute infrastructure, funding the build-out of the hardware layer Web4 runs on.
The Implication
Cuban's been wrong about crypto. But he might be early on compute financialization. If GPU futures take off, watch for secondary markets: compute-backed lending, fractional GPU ownership, and derivatives that let startups hedge their training costs. The companies that figure out how to trade compute as a financial asset will have an edge in the agent economy.
For builders: this is your hedge. If you're burning $50K a month on inference costs, GPU futures let you lock in rates before the next AI boom sends prices vertical. For investors: Cuban's track record says fade him. But the underlying thesis — that compute becomes the most important commodity of the next decade — that one's hard to argue with.