BlackRock Claims AI Computing Power May Be Tokenized in the Future
The world's largest asset management company, BlackRock, released a research report titled "Machine Native Economy," suggesting that computing power is becoming an investable resource. In the future, digital contracts could represent established rights to use computing power, which can be transferred, financed, or used as collateral on programmable infrastructure.
The report identifies three overlaps between artificial intelligence and digital assets: payments between agents, the invocation of tokenized financial assets, and the computing power market. Revenue from cloud giants is used as a scale proxy, estimating that Amazon Web Services, Microsoft Azure, and Google Cloud will collectively reach about $1.1 trillion by 2030, with a compound annual growth rate of about 29% relative to 2025. The adjusted trading volume of stablecoins is cited to exceed $11 trillion by 2025.
BlackRock noted that standardized products may include tradable computing power futures, facilitating hedging of prices and availability for both supply and demand sides; agents can automatically select and settle based on price, performance, latency, location, and hardware inventory. The report views GPU mortgage loans as an early form of financialization of infrastructure, specifically mentioning Stripe's acquisition of model routing company OpenRouter as a signal of integrating computing power allocation into financial pipelines.
This aligns with the company's existing tokenized products: on-chain money market BUIDL, proposed tokenized money market shares, Aladdin's access to synthetic dollar liquidity, and the data center platform promoted with Microsoft and MGX, with an initial target of about $30 billion, potentially leveraged or expanded to about $100 billion. NVIDIA has signed memorandums of understanding with BlackRock, Blackstone, Apollo, KKR, Goldman Sachs, and Brookfield to treat AI factory-level computing power as a financeable long-term asset, with third-party capital potential exceeding $500 billion.
The report also emphasizes that tokenization does not exempt qualified investors, customer due diligence, and anti-money laundering, with compliance being independent of the on-chain ledger. This is a research outlook, not an announcement of issuing computing power tokens.
The market mechanism is to rewrite the scarcity of training and inference durations from cloud billing into rights certificates that can be pledged and hedged. Buyers are model companies and agent networks looking to lock in future machine hours, while sellers are large-scale clouds and infrastructure funds holding cabinets, power, and chips. The event-driven aspect comes from the release of the white paper. Beneficiaries are asset managers and chip manufacturers that can already securitize machine hours, while those under pressure are decentralized computing nodes that can only rent by the hour and cannot enter the collateral pool. Funding will shift from subscription cloud fees to pre-sold capacity and futures margins.
Source: Public Information
ABAB AI Insight
BlackRock's approach has consistently been to first write research and then turn that research into manageable products. Bitcoin spot ETFs, BUIDL, and tokenized treasury shares are all about moving existing assets onto transferable ledgers. This time, integrating computing power into the same framework acknowledges that training and inference have become as scarce as electricity and positions; merely buying NVIDIA stock is not enough, one must also buy "rights to chips for future hours." Larry Fink previously discussed computing power futures at the Milken Conference, and this report translates that verbal concept into three market categories: payments, assets, and machine hours.
The capital mobilization is very specific. On one side are tokenized cash and treasury bonds, providing agents with 24-hour settlement currency; on the other side are data centers and GPU financing platforms, turning cabinets into long-term assets similar to real estate. OpenRouter is mentioned in a footnote, indicating that once the routing layer is acquired by a payment company, model invocation will become a quantifiable financial action rather than just engineering scheduling. Compliance is emphasized separately to allow pension and insurance funds to participate, rather than handing computing power over to unlicensed computing networks.
Benchmarking targets include power forwards, bandwidth contracts, aviation fuel hedging, and decentralized GPU leasing markets. The former has standardized delivery and ratings, while the latter has spot availability but struggles to enter the collateral list. The industry is at the starting point of upgrading "cloud billing" to "tradable capacity": futures and tokens are still conceptual, while financing memorandums have already arrived.
The structural change is a transfer of pricing power. Whoever can standardize machine hours into contracts will determine the cost curve for model companies without needing to own all the chips. The mechanism is that agents must order their own inference; traditional invoicing systems cannot handle small, all-time, programmable transactions; digital assets provide the settlement layer, asset management provides qualified packaging, and chip manufacturers provide collateral. Once computing power can be pledged, cloud discounts become interest spreads, no longer just procurement terms.
ABAB News · Cognitive Laws
Scarce resources will not see price increases next, but will become contracts.
When agents can pay themselves, invoices will turn into tokens.
Whoever turns machine hours into collateral will reprice model companies.