Mark Cuban has made another bold call on the artificial intelligence business, saying chips may end up trading like a new class of crypto-linked assets.
In a post on X on Aug. 16, the billionaire investor wrote, “chips as an asset class will be the new crypto.”
Cuban did not spell out which chips he meant, nor did he outline a specific investment product or financial plan. The report said his comment has largely been read as referring to high-end GPUs such as Nvidia’s H100 and B200, hardware widely used for AI training and inference and now central to large-model development, cloud computing expansion, and data center buildouts.
Scarcity and compute demand sit at the center of the argument
The logic behind the comparison rests on two factors: scarcity and demand for computing power. As generative AI has expanded, companies have needed far more high-performance compute. That has turned advanced GPU supply and data center capacity into key competitive inputs across the AI sector.
The article cited Nvidia’s most recent quarterly data center revenue at $75.2 billion, up 92% year over year, as a sign of how large demand for AI compute infrastructure has become.
Wall Street is starting to finance GPUs like income-producing assets
The report said parts of the financial market are already moving in that direction. AI cloud computing company CoreWeave has used GPUs extensively to support financing in recent years, and in August it completed a $2.6 billion delayed-draw term loan for high-performance computing infrastructure.
That financing carries a term of about five years, longer than the duration of some customer contracts backing the loan. The report also noted that CoreWeave completed another $3.1 billion public syndicated financing in May and described AI infrastructure financing as an “emerging asset class.”
Those deals suggest that high-end GPUs are beginning to show some of the traits normally associated with financial assets. Lenders are willing to evaluate the future cash-flow potential of chips and extend financing on that basis.
GPU rental futures are also on the way
The next development is showing up in derivatives. CME Group is expected to launch futures products on Oct. 5 based on rental prices for Nvidia H100 and B200 GPUs, giving market participants a way to hedge or trade future compute costs.
For AI developers and cloud service providers, that would allow part of their GPU rental costs to be locked in ahead of time. The report said the compute market is gradually moving toward standardized commodity-style trading.
From buying chips to buying compute
What makes AI chips different from ordinary hardware is that their value can be turned directly into cash flow through rented computing capacity. Once a company buys GPUs and deploys them in a data center, it can sell that compute to AI model developers, cloud providers, and other businesses under long-term agreements or usage-based pricing.
That shifts GPUs from being internal equipment to becoming infrastructure assets with measurable revenue potential. As lenders accept GPUs as a financing base and exchanges start building futures markets around compute, chip prices, rental rates, utilization, and projected cash flows can all feed into valuation models.
In that sense, Cuban’s “new crypto” phrase is closer to an asset-class analogy than a literal comparison. The article said AI chips could form a market defined by scarcity, market pricing, collateral value, and derivatives trading. As demand for compute rises, what gets traded may expand from physical chips to the computing power those chips deliver.
But GPUs and Bitcoin remain fundamentally different
The article also stressed the limits of the comparison. Bitcoin (BTC) allows ownership and transfers to be verified on a public blockchain, trades around the clock in a global market, and operates under protocol-defined supply rules. GPUs are physical hardware. They must be housed in data centers and carry depreciation, wear, energy costs, and the risk of fast technology turnover.
Bitcoin advocate Pierre Rochard challenged Cuban’s analogy as well, arguing that chip production lacks mechanisms comparable to Bitcoin halving and mining difficulty adjustment. When demand rises, semiconductor companies can expand output. And when a new generation of chips arrives, the economic value of older GPUs can fall quickly.
The report said the market is still some distance away from treating AI chips as a complete and independent asset class, and Cuban did not give a timeline or a formal financial structure. Even so, GPU-backed financing, compute rental markets, and related futures products are beginning to appear, suggesting that AI infrastructure is starting to develop the conditions for financialization.
If compute becomes more standardized in pricing, collateralization, and trading, investor exposure could move beyond chipmaker stocks such as Nvidia and into GPUs themselves and the ongoing computing output they generate.