Kalshi's Nvidia Compute Markets Hit $4.4M: Is AI Compute the 'New Oil'?

Wall Street's next commodity may not sit in a barrel or a warehouse. It may be an hour of computing time on a chip from Nvidia Corp. (NASDAQ: NVDA).

Kalshi's GPU rental markets generated $4.4 million in notional volume through July 27, according to data analytics firm Allium. That is tiny compared with oil or gold markets, but meaningful for a product that barely existed weeks earlier.

The real test is whether these early markets can establish a reliable forward price for AI computing power.

$4.4 Million Is Small, but Not Nothing

Bloomberg Opinion argued Monday that "barely anyone" is trading compute contracts, noting Kalshi's most popular August market has generated just over $100,000.

Bloomberg measured a single contract, while Allium's total covers Kalshi markets across five Nvidia chips. Kalshi's total was roughly 15 times the $285,000 traded on rival Polymarket.

That is nowhere near enough volume for a large company to hedge its computing costs. But for such a new market, it shows that traders are interested in GPU prices.

Is Kalshi Producing a Useful Price?

Allium examined six completed weekly markets tracking Nvidia B200 rental prices. Two days before settlement, Kalshi's forecasts were a median 10% away from the final price while remaining within 2.2% of the current spot price. That suggests traders were largely following spot prices rather than anticipating where they would move next.

That is not a compelling forecasting record, but neither is it disastrous for a market that had existed for only weeks. Six contracts is far too small a sample for a firm verdict, particularly while traders are still learning how to price compute. What to watch is whether Kalshi's forecasts begin moving ahead of spot prices as participation grows.

Why Wall Street Is Watching

CME Group Inc. (NASDAQ: CME) plans to launch futures tracking monthly rental costs for Nvidia's H100 and B200 chips on Oct. 5, pending regulatory review. The exchange compares compute's development with oil's evolution from spot trading into a global derivatives market.

Unlike barrels of crude, GPU hours are not interchangeable. Their price depends on the chip, location, network performance and reliability. But a futures market does not require every unit to be identical. It only needs a benchmark that moves closely enough with customers' costs to provide a useful hedge.

Kalshi's early volume shows there is interest in trading compute prices, even if the market is not ready for large corporate hedges.