{"id":651,"date":"2026-07-21T07:47:00","date_gmt":"2026-07-21T14:47:00","guid":{"rendered":"https:\/\/quantcha.com\/news\/?p=651"},"modified":"2026-07-23T19:34:27","modified_gmt":"2026-07-24T02:34:27","slug":"how-prediction-markets-could-defuse-the-ai-compute-crash","status":"publish","type":"post","link":"https:\/\/quantcha.com\/news\/how-prediction-markets-could-defuse-the-ai-compute-crash\/","title":{"rendered":"How Prediction Markets Could Defuse the AI Compute Crash"},"content":{"rendered":"\n

Kalshi is building a stack of instruments on the price of compute. Completed, it could be a critical layer the AI buildout never had.<\/em><\/strong><\/p>\n\n\n\n

Every few weeks someone dusts off that old Cisco chart showing how they were the most valuable company in the world in early 2000 thanks to their role in selling the infrastructure required to build out the internet. Unfortunately, when the demand waned, overcapacity turned into glut and the stock crashed, along with the rest of the market.<\/p>\n\n\n\n

I would like to be the first person to suggest there\u2019s a similar story playing out with Nvidia and the AI capacity buildout.<\/p>\n\n\n\n

Just kidding. You can\u2019t use the financial internet these days without seeing that story rehashed a dozen times.<\/p>\n\n\n\n

But there\u2019s something different this time around. We have something the companies and investors of the 1990s didn\u2019t have: prediction markets.<\/p>\n\n\n\n

Reliably pricing demand<\/h2>\n\n\n\n

In the past, all you could really do was find a way to be on the record that you saw a crash coming. Or, if you had real conviction, you could find a way to profit off of the crash through savvy investments. But neither of those were economically productive in such a way that could help head off the crash and keep the economy stable.<\/p>\n\n\n\n

What we really needed\u2014and what we kind of have now\u2014is a way to trade the thing being overbuilt. Not because we want to empower investors to profit off of crashes, but rather to provide a true price discovery mechanism so that the industry involved can optimize their planning to head off the crash to begin with. If the market consensus is that there will be less demand for something\u2014like H200 chips\u2014in twelve months, then that information can be used to inform investments by companies so that they don\u2019t unintentionally drive off the cliff.<\/p>\n\n\n\n

I argued in The Options Story for Prediction Markets Just Snapped Into Focus<\/em><\/a> that retail options traders never had a home in prediction markets because binaries don\u2019t provide what they need despite being called \u201coptions\u201d. They lack convexity, offer no recovery, and support nothing an income seller can survive a losing trade on. However, Kalshi\u2019s compute curves, plus the perpetual futures (perps) and options coming behind them, can change that. They open the possibility of traditionally dated and\/or perpetual puts and calls.<\/p>\n\n\n\n

Risk transfer instruments built on information markets<\/h2>\n\n\n\n

Enabling meaningful risk transfer is the greatest value prediction markets can deliver.<\/p>\n\n\n\n

It starts at the information layer, which already exists. Binary threshold ladders\u2014\u201cwill a GPU-hour clear $K?\u201d across strikes and tenors\u2014are the market\u2019s implied distribution. String their centers across tenors and you get a forward curve. That\u2019s the threshold-to-distribution-to-curve move I walked through last time, and you can trade those right now. It\u2019s genuine price discovery because it gives capital a way to express views on market demand well ahead of time. It\u2019s also the promise these markets always make about the positive impact they could have on society. But the problem at this layer is that it\u2019s very complex to use and manage while also not providing the efficient instruments major participants want.<\/p>\n\n\n\n

This is where the risk transfer layer comes in. Simpler instruments like futures and options provide coverage that neatly maps to the needs most participants have.<\/p>\n\n\n\n

The whole stack in one grid<\/h2>\n\n\n\n

Futures come first. A perp or a dated future is a linear claim that lets a data center or a lender lock a price and move the risk symmetrically, which is most real-world hedging and does a lot of the work. But a future locks you both ways: you get downside protection at the cost of upside potential.<\/p>\n\n\n\n

Options are the second rung, and they add what futures can\u2019t: asymmetry. And a market where protection is sold as a product. Now you can buy a put to floor the downside and keep the upside. Or sell one and you\u2019re running an insurance book. Futures give you a hedging venue and options give you an insurance market.<\/p>\n\n\n\n

Put the axes together and the whole stack is a clean grid where every instrument is linear or convex, continuous or dated. And it all rests on the binary matrix.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

That matrix is the basis. And you could get most of the higher-level coverage through some combination of those binaries. But do you really want to? Why not get the exact exposure you\u2019re looking for and let other players in the stack take on their part of the risk in insuring it?<\/p>\n\n\n\n

The genuinely new pieces are the two continuous ones, the perp and the perpetual option, because those aren\u2019t combinations of any single tenor\u2019s rungs. Dated instruments carry the horizon precision while the perpetual ones stay deliberately coarse and deep.<\/p>\n\n\n\n

And because every instrument is a function of the same distribution and the same underlying, there are natural synergies.<\/p>\n\n\n\n

For the taker, it\u2019s a more precise surface than equity options. Not only do you get a continuous underlying with optionality, but you can also pepper in exposure to discrete events via binaries. I had hoped the equity world would have seen these coming for earnings and rate announcements by now, but there\u2019s no sign of them.<\/p>\n\n\n\n

For the maker, that same coherence keeps the diversity from shattering liquidity. Every instrument nets back to the same distribution and the same perp, so a maker hedges one with a basket of the others and quotes both sides more safely than on an equity surface built from sparse strikes. You could participate across the board or specialize based on expertise. I\u2019m eager to start building tools for this space once we know more about exactly what\u2019s coming because I think it would be neat to find optimal ways to balance a book based on this varied arsenal of instruments.<\/p>\n\n\n\n

Seeing a crash versus surviving it<\/h2>\n\n\n\n

We\u2019ll never know how the industry would have handled the dotcom boom if they had this combination of information and instruments available at the time. Would Cisco have locked in a price for the next year of hardware sales? Would hosters have bought capacity in bulk or just hedged near-term needs? Would non-tech companies have invested so heavily in off-brand proprietary tech plays expected to take years to profit? Would startups still have bought Super Bowl ads without explaining what they were actually selling?<\/p>\n\n\n\n

The information layer lets you see a glut (or shortage) coming and the risk transfer layer offers the opportunity to survive your exposure. This time, the major players may have what they need before it\u2019s too late.<\/p>\n\n\n\n

On the other hand, some might object that this is how 1987 happened. While I\u2019ll concede that there\u2019s some merit there, 1987\u2019s portfolio insurance was toxic because everyone synthesized the same protection by selling into the fall\u2014a real options market instead lets that risk be warehoused by someone who actually wants it. A lot of the impact for this scenario will come down to what sort of leverage is made available. If properly designed, those ultimately taking on the risk will be able to absorb it without causing a failure cascade that takes everyone out with them. I don\u2019t know where that line should be drawn, but I\u2019m sure there\u2019s a reasonable place for it.<\/p>\n\n\n\n

Back to modern reality<\/h2>\n\n\n\n

Pretty much everything I\u2019ve covered here is speculation of what might someday be. None of the convexity has even been discussed by Kalshi as far as I can tell. At this point they just have the binary matrix and forward curves available. Perps are coming, but no ETA just yet. Participants can read some information off the markets, but the books are really thin and I wouldn\u2019t put much faith in the quotes just yet.<\/p>\n\n\n\n

The incumbents are moving in too: CME is standing up an AI-compute futures market with Silicon Data and ICE is listing GPU-compute futures settled to Ornn (the same index Kalshi uses). Their arrival is the clearest sign compute is becoming a real derivatives asset class rather than a prediction market curiosity.<\/p>\n\n\n\n

The race is on. Who do you think builds the deepest stack first: Kalshi, CME, or ICE? And will it actually change how the AI buildout is financed?<\/p>\n","protected":false},"excerpt":{"rendered":"

Kalshi is building a stack of instruments on the price of compute. Completed, it could be a critical layer the AI buildout never had. Every few weeks someone dusts off that old Cisco chart showing how they were the most valuable company in the world in early 2000 thanks to their role in selling the … Continue reading “How Prediction Markets Could Defuse the AI Compute Crash”<\/span><\/a><\/p>\n","protected":false},"author":1,"featured_media":653,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":[],"categories":[8],"tags":[],"_links":{"self":[{"href":"https:\/\/quantcha.com\/news\/wp-json\/wp\/v2\/posts\/651"}],"collection":[{"href":"https:\/\/quantcha.com\/news\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/quantcha.com\/news\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/quantcha.com\/news\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/quantcha.com\/news\/wp-json\/wp\/v2\/comments?post=651"}],"version-history":[{"count":1,"href":"https:\/\/quantcha.com\/news\/wp-json\/wp\/v2\/posts\/651\/revisions"}],"predecessor-version":[{"id":654,"href":"https:\/\/quantcha.com\/news\/wp-json\/wp\/v2\/posts\/651\/revisions\/654"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/quantcha.com\/news\/wp-json\/wp\/v2\/media\/653"}],"wp:attachment":[{"href":"https:\/\/quantcha.com\/news\/wp-json\/wp\/v2\/media?parent=651"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/quantcha.com\/news\/wp-json\/wp\/v2\/categories?post=651"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/quantcha.com\/news\/wp-json\/wp\/v2\/tags?post=651"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}