👋 Hey, welcome to Predicted; the newsletter covering the business of prediction markets.

Forwarded this email and want it in your inbox? Hit the subscribe button below! 👇

Discussed in this piece:

1) Kalshi launched forward curves for GPU rental prices, the first public price for future compute

2) Why the AI buildout needs hedging, from an $8-to-$3 rental crash to $700bn of debt-funded capex

3) The benchmark business, and the race against CME and ICE to own the number

4) Whether a sports-heavy order book can produce a price anyone will trust

🖋️ This piece was written and researched by Pet Berisha and Omar El Safy

Kalshi went live with forward curves for GPU rental prices on July 14.

The curves track what it costs to rent Nvidia's B200, H200 and A100 chips by the hour, over the coming weeks and months, derived entirely from trading in Kalshi's own chip-price markets.

Kalshi CEO Tarek Mansour put it plainly. "Compute is the new oil. Like every commodity before it, it needs a real derivatives market."

A forward curve is a chart of what the market expects something to cost in the future.

Fix your energy tariff or your mortgage rate, and you have used one.

The curve itself is not a product you trade. It is a reference number that two firms can write into a private deal. If they actually lock in that price, they still have to trade the underlying markets or agree to a block trade (privately negotiated large trade) on the exchange.

Kalshi only earns fees when firms trade in the underlying market contracts to lock in a price.

CME are building a similar product. But Kalshi’s is different.

Because CME’s compute futures will cash‑settle against Silicon Data’s daily GPU rental indices, which record what chips actually rented for across providers.

Kalshi’s Forward Curve is explicitly the market’s implied view of what GPUs will rent for next month.

And a mess is where derivatives earn their keep.

H100 rental prices went from above $8 an hour in 2023 to around $3 by late 2025, as Nvidia ramped supply and 300+ new providers entered the market. Then the tide turned. By early 2026, Silicon Data showed zero on-demand availability across 90% of providers, with renters subletting clusters to each other. A glut, then a shortage, inside a year.

Meanwhile, spending in this market keeps climbing.

Meta, Microsoft, Alphabet and Amazon plan $700bn in AI capex, and more of it is debt-funded every quarter. Meta borrowed $27bn from Blue Owl for a single data centre project.

The neoclouds live and die on this number. CoreWeave, Lambda, Nebius and a few hundred smaller operators borrowed to buy chips, and those loans are collateralised by the GPUs themselves. If rental prices fall faster than the debt amortises, the equity is gone.

A neocloud borrows, say, $1bn against its chips and pays it down over four years out of rental income. The GPUs are the collateral, so the rental price sets both what the business earns and what the collateral is worth. If rents fall 40% in a year, the loan balance has barely moved while the income servicing it and the collateral backing it have both shrunk. The lender is underwater on the security, refinancing is off the table, and the shareholders are wiped out before the bank takes its first loss.

And nobody can underwrite a loan against a GPU without a reference rate for what a GPU earns.

Power plants borrow cheaply because electricity has forward curves. Neoclouds want to lock in what their chips will earn next quarter. AI labs want to cap what training will cost. Lenders want a number to write into covenants.

Larry Fink told the Milken conference that a new asset class will be buying futures of compute.

Five days later, CME and Silicon Data announced compute futures.

ICE are building GPU capacity futures with Ornn.

DRW's Don Wilson, whose firm backs Silicon Data alongside Jump Trading, called compute "the largest commodity in the world."

Both of those products are waiting on regulatory review.

Kalshi didn't have to wait. Event contracts on a CFTC-licensed exchange can be self-certified, listed first without asking permission. It is the same mechanism that let Kalshi launch perpetual futures after CFTC Chairman Michael Selig loosened the delivery-date definition.

CME's compute futures are due later this year, while Kalshi's new curve offering is live now.

Prediction markets have spent four years selling the wisdom of crowds. Until now, the output was a probability, and the customer was a bettor, a journalist, or someone arguing on X.

A forward curve is a number other people write contracts against.

Oil is priced against Brent. For decades, loans were priced against Libor. Whoever owns the reference number earns fees on the trading around it, which is why CME run 66% operating margins and ranks among the most profitable companies in the S&P 500.

There is a precedent for how these benchmarks get born, and it looks a lot like compute in 2026.

WTI crude futures launched on NYMEX in 1983, a direct product of the 1970s oil shocks. Buyers who couldn't find oil at any price demanded transparency, and the futures price became the price.

Compute had its own version of that moment this year, when on-demand availability hit zero, and the market started trading GPU capacity like a back-alley commodity.

If a neocloud signs a swap against Kalshi's B200 curve, a prediction market price has become financial plumbing.

Vitalik Buterin made the argument in February that prediction markets only become durable when hedgers replace naive traders, and offerings like these take us a step closer to that reality.

Kalshi did $33bn in volume in June 2026. Roughly 87% of it was sports wagering.

A curve is only as good as the depth behind it. Thin markets produce a number that moves when one trader sneezes, and no CFO hedges nine figures of compute against a number like that.

And there is a harder question underneath. Who quotes these markets?

Kalshi's parlay business runs on request-for-quote market makers pricing custom combinations, and the hold gets split between the makers and the exchange.

The GPU book will lean on the same kind of firms.

But an NFL parlay resolves on Sunday night.

A GPU forward is a genuine price risk that someone has to warehouse for weeks, in an asset where configuration, networking and power supply all move the price.

The firms best equipped to hold that risk, the DRWs and Jump Tradings of the world, are the ones backing the competing index.

Kalshi project compute futures will one day out-trade oil's 800 million contracts a year. WTI took a decade and an oil crisis to get there. It only happens if data centres and labs hedge routinely, in size, on an exchange most of Wall Street feels is predominantly an arbitrage on sports gambling.

On July 7, Kalshi lost a preliminary injunction in the Southern District of New York.

Judge Torres ruled New York can treat their sports contracts as gambling while the case plays out.

More than a dozen states are in open conflict with the CFTC over the same question.

Minnesota makes trading sports contracts a felony from August 1. Nevada held a contempt hearing this week.

Every commodities-shaped product Kalshi ships is also a legal argument.

A GPU forward curve looks like the CME. The more of Kalshi's surface area that behaves like a commodities exchange, the stronger their claim to be one.

But the volume runs the other way. If GPU curves stay a rounding error next to sports, the gambling label sticks, and the SDNY ruling suggests judges are reading the volume numbers too.

Prediction Markets are both controversial and fascinating, and we’re seeing this in real time. Venues where genuinely interesting markets and derivatives are being generated, whilst also fighting a dozen states over whether their biggest offering by volume, sports contracts, is gambling. 

The path where prediction markets land is messy, but the more of these types of offerings, the more likely that path is one with staying power, generational companies and one that is genuinely additive to the economy and markets.

This newsletter is for informational purposes only and is not financial, business or legal advice. These are the author's thoughts & opinions and do not represent the opinions of any other person, business, entity or sponsor. Any companies, platforms, markets or projects mentioned are for illustrative purposes unless specified.

The contents of this newsletter should not be used in any public or private domain without the express permission of the author.

The contents of this newsletter should not be used for any commercial activity, for example - research report, consultancy activity, or paywalled article without the express permission of the author.

Please note, the services and products advertised by our sponsors (by use of terminology such as but not limited to; supported by, sponsored by or brought to you by) in this newsletter carry inherent risks and should not be regarded as completely safe or risk-free. Third-party entities provide these services and products, and we do not control, endorse, or guarantee the accuracy, efficacy, or safety of their offerings.

It's crucial to provide our readers with clear information regarding the inherent nature of services and products that might be covered in this newsletter, including those advertised by our sponsors from time to time. When you trade on prediction markets (including event contracts, opinion markets and other speculative instruments) your capital is at risk. Risks associated with prediction markets include price volatility, loss of capital (the value of your position could drop to zero), illiquidity, complexity, evolving regulation and lack of protection. Many prediction market operators do not currently operate in a fully regulated industry, and availability varies by jurisdiction. Therefore, please be aware that when you place funds on prediction markets, you may not be protected under financial compensation schemes and protections typically afforded to investors when dealing with regulated and authorised entities to operate as financial services firm. Nothing in this newsletter constitutes a recommendation to place, hold, or close any position on any market.