Crypto Sector Establishes New Financial Market for AI Processing Power

The cryptocurrency infrastructure is creating a new financial layer for the pricing, financing, and trading of the processing power used to run artificial intelligence models. According to a new report published by Galaxy Research, GPU capacity, AI inference services, and access rights to these services are increasingly becoming financialized assets.

In the report prepared by Lucas Tcheyan, Vice President of Galaxy Research, "inference capital markets" are defined as networks, protocols, and financial instruments that develop around the processing capacity that enables AI models to generate outputs from new data.

The report indicates that the focal point of GPU demand in the AI sector is shifting from training models to daily execution, i.e., inference processes. This shift increases the need for new markets for companies looking to stabilize the price of processing capacity, manage future costs, and finance hardware investments.

Predictions shared by Galaxy suggest that investments in AI infrastructure could reach trillions of dollars in the coming years. Morgan Stanley forecasts that approximately $2.9 trillion will be invested in data centers by 2028, while McKinsey estimates the total capital expenditure needed by 2030 to be $6.7 trillion. Goldman Sachs' forecast for AI infrastructure from 2026 to 2031 is around $7.6 trillion.

It is noted that between 55% and 67% of these investments could be allocated to processing power and hardware. However, due to variables such as chip model, memory capacity, connectivity infrastructure, region, and rental duration, creating a standard market is not easy. Companies like Ornn and Silicon Data are developing GPU price indices compiled from real transactions to overcome this issue. It is also reported that the parent company of the New York Stock Exchange, ICE, and CME are preparing to list GPU futures contracts.

According to the report, crypto projects are transforming not only physical GPU capacity but also access rights to AI services into tradable assets on-chain. The VVV and DIEM tokens of the Venice platform are highlighted as prominent examples of this model.

Users who lock VVV tokens can produce DIEM. Each DIEM provides its holder with a daily credit of $1 for Venice AI usage. Thus, users can hold, transfer, or sell their future AI usage rights instead of purchasing a subscription. According to Galaxy, this structure is noteworthy as it transforms AI processing capacity from a rented service into an owned asset.

However, the report notes that DIEM is primarily held for speculative purposes rather than actual usage. It is stated that the weekly usage rate of the offered inference capacity remains below 50%, and the token's value depends on the expectation that Venice will continue to provide services in the long term.

Projects like Pearl and Ambient aim to evaluate the computational capacity used in blockchain mining for AI inference through a "proof of useful work" approach. Pearl integrates matrix multiplication operations used by AI models into block production, while Ambient employs an auction-based system where users set prices and durations for AI tasks.

According to Galaxy, these models can offer AI services below market prices thanks to token incentives. However, there is a risk that networks may become filled with processing power providers only seeking token rewards if real customer demand does not materialize.

The report emphasizes that a sustainable link between token demand and actual AI usage has yet to be established.

The most viable model presented in the report is financing GPU hardware on-chain. The USD.AI protocol directs stablecoin deposits as loans to small and medium-sized data center operators purchasing GPUs. Loan repayments are covered by GPU rental income, while interest income is passed on to deposit holders.

Galaxy states that this model distinguishes itself from other initiatives because it relies on real borrowers and cash flow rather than speculative token incentives. However, it warns that a drop in GPU rental prices, faster-than-expected depreciation of hardware, or weakening AI demand could lead to an increase in loan defaults.

According to Galaxy Research, inference capital markets are still quite small compared to the size of the AI sector. However, when GPU indices, futures, tokenized usage rights, and hardware loans are considered together, the foundations of a new asset class worth trillions of dollars are being laid. The report concludes that the advantage of the crypto sector is not to provide cheaper services than centralized AI companies but to create capital in areas where traditional finance is slow or inadequate, make assets divisible, and provide programmable payment infrastructure.