The Financialization of AI Compute
Building the biggest financial market in the age of GenAI
Every company will be an AI inferencing company…
… just like every company today is an oil & gas company.
I hosted a salon for Wall Street traders and startup founders to debate the financialization of AI compute. Serendipitously, we met in the building that used to host the School of Mines at Columbia University.
Every company will be an AI inferencing company.
Let me explain.
Today, every company is an oil & gas company. No matter what a company sells, it’s affected by the global undercurrent of the supply & demand of commodities. Farmers feed grains to their livestock, and grain prices fluctuate with that of oil and gas. Toy manufacturers make plastic cars, and plastics are derived from coal and oil byproducts. Airlines run on fuel. Amazon relies on a big fleet of delivery vehicles to fulfill its Prime deliveries, and these trucks guzzle up fuel. The list goes on.
The global economy today runs on oil and gas.
Every company tomorrow will be an AI inferencing company. In a mere130 years, humanity has progressed from mining coals to mining bitcoin. The raw materials for production are also evolving from physical commodities to virtual AI work.
Inferencing simply refers to the process where a machine (AI) makes conclusions based on brand-new data. If ChatGPT has answered a question you asked it, you’ve experienced inferencing. At an industrial scale, AI inferencing will infiltrate every corner of the economy. In agriculture, a robotic laser weeder uses AI inferencing to weed crops 2x faster. Here AI makes inferences about cutting techniques based on pictures of crops it takes in real time. In public services, an AI-enabled police bodycam auto-creates training snippets based on bodycam footage. Here AI makes inferences about a snippet’s relevance to training. In B2B sales, AI SDRs have flooded the market. Here AI makes inference about a prospect’s likelihood to buy certain products.
AI inferencing is already here and will continue to take over all corners of the economy.
Every company will need to hedge AI compute
Today, to combat the price volatility of oil & gas-major inputs of production-companies engage in hedging, i.e. locking in the future price of oil today. In everyday terms, it’s like going to your local grocer and tell them “Egg prices are bonkers and I need to do something about it. So I want to lock in a fair price for 2026. Twelve months from now, I only want to pay $9.99 for a dozen of eggs; to lock in that certainty, I’m willing to put down some money today.” The grocer agrees and takes your small deposit.
You and the grocer just completed a futures contract transaction.
In fact, globally the most popular instrument of hedging is via futures. The commodity futures is a $14.1 trillion market.
Today, the market size for AI compute futures (i.e. future GPU capacity) stands at nil.
Are we on the cusp of a Cambrian explosion of the Financialization of AI Compute? Or will this market follow the path of the DRAM futures exchange?
I believe we are staring at a Cambrian Explosion. Here’s why.
Volatility
Now, is GPU price as bad/volatile as that of eggs in 2025 or that of oil in a geopolitically uncertain world? It turns out, yes. GPU prices can fluctuate even more than oil prices. With advancement in inferencing techniques, more use cases of inferencing, uncertain demand from crypto mining, and step changes in computing efficiency, the demand for AI compute will be hotly debated and watched for decades to come.
Disagreements and uncertainty brew volatility.
Liquidity
Liquidity, said differently, is the questions of whether there’ll be enough bids and offers on AI compute in any given day. Liquidity is usually measured in daily trading volume, and in the tightness of the bid-ask spread. For example, meme stock AMC’s average daily trading volume is a whopping 10.8 million; its lesser known cousin Regal Cinema’s daily volume is “only” half a million.
Typically, several key factors drive liquidity:
Number of market participants
Level of disagreement on price
Exogenous events
If you believe every company will be an AI referencing company, then every company will participate in the AI futures market, directly or indirectly.
Of course, the level of concentration among suppliers/buyers will affect the path to full liquidity. But that’s a temporal problem, not a fundamental hurdle.
Some may also argue that in the long term, the cost of inference is doomed to go a certain direction. But as many people will argue the opposite side, thus creating much needed volatility and liquidity for the market.
Standardization
Standardization is one of the most hotly debated topic at my salon. Meaning, can AI chips go the way of crude oil, and be as standardized as fitting neatly into four main types?
To respect participants’ privacy, Chatham house rules apply here :)
Other considerations
At the salon, we also debated the following topics:
Which form of financialization is the initial wedge
What market participants will be first movers
Anecdotes and forecast of the supply & demand of AI chips
Conclusion
In a few years, AI referencing will be as foundational to the economy as oil and gas are today.
The conditions for financializing the AI compute market are ripe.
Are you creating the biggest financial market of the future? Let’s get in touch