One of crypto’s best-known venture firms wants AI companies to put their graphics cards on the blockchain and borrow stablecoins against them — and the idea is drawing both excitement and warnings as AI borrowing hits record levels.
By Amir Hassan | October 30, 2026
Multicoin Capital partner Shayon Sengupta proposed creating blockchain-based markets where ownership claims on computing hardware, future data center capacity, and customer payments could be traded or pledged as collateral for loans, according to an Oct. 8 research note reported by CryptoSlate. For everyday crypto investors, the pitch matters because it could open a massive new source of real-world demand for stablecoins — the digital dollars that already sit at the heart of crypto markets.
The Hook: A 419 Billion USD Borrowing Spree With Nowhere Left to Turn
The rapid expansion of artificial intelligence has triggered unprecedented borrowing as technology companies race to build data centers and acquire the powerful chips needed to run AI models. According to The Kobeissi Letter, US dollar-denominated corporate bond issuance linked to AI reached a record 419 billion USD in 2026 as of Oct. 21. That is a 90% increase from the entire 2025 total and a 437% increase from 2024.
- 419 billion USD — record AI-linked corporate bond issuance in 2026 as of Oct. 21, per The Kobeissi Letter
- 157 billion USD — portion borrowed by Alphabet, Amazon, Meta, Microsoft and Oracle
- 103 billion USD — raised by investment-grade and high-yield data center and computing companies
- 90% jump versus all of 2025 — and a 437% jump versus 2024
But here is the catch: those eye-popping numbers belong almost entirely to tech giants with massive balance sheets. Smaller AI infrastructure operators have secured customer contracts and ordered computing equipment, yet they lack the financing relationships that let household-name companies raise billions through conventional debt markets. That gap is exactly where Sengupta believes blockchain technology can step in.
On-Chain Evidence: A 26.5 Billion USD Queue at One Platform Alone
The demand is not theoretical. David Choi, co-founder of GPU financing platform USD.AI, said the platform has received more than 700 financing requests totaling 26.5 billion USD since its launch, with about 8.7 billion USD moving through different stages of evaluation. Most requests come from operators seeking between 10 million USD and 300 million USD to expand their computing capacity.
Think of it like a pawn shop for the AI era — except instead of jewelry, companies are pledging the industrial-grade chips that power tools like ChatGPT, and instead of cash from a register, they are borrowing stablecoins from a global pool of lenders. USD.AI reported more than 420 million USD in lifetime deployed capital and 659 million USD in total value locked as of Oct. 2 — a striking illustration of how much bigger the queue is than the loans actually handed out so far.
The Core Conflict: Digital Claims versus Physical Reality
Sengupta’s vision goes beyond lending against physical hardware. Under his proposal, a data center operator could issue a transferable digital claim representing access to a GPU cluster during a future period. That claim could be sold to another customer or pledged to secure a stablecoin loan, letting the operator finance computing capacity it has yet to deliver. Companies could also borrow against payments expected from existing customer contracts.
He argues blockchain networks are well suited for this because they allow lenders to verify pledged assets, monitor existing debt obligations, and automate payments as computing services are delivered — functions that normally require banks, lawyers, and months of negotiation. He also proposed specialized intermediaries that would connect compute buyers and sellers while managing collateral, financing, and price risks across multiple operators.
The risks, however, are very real. Individual investors in USD.AI’s yield-bearing sUSDai token do not directly own the GPUs backing the loans or control what happens to them after a default. Dedicated legal entities hold the equipment, and appointed agents handle recovery — meaning blockchain investors depend on traditional legal processes when things go wrong. Withdrawals from sUSDai operate through 30-day cycles and can be paid out partially if requests exceed available cash. GPU prices can also decline as newer chips hit the market, so even enforceable claims over equipment may not recover the full loan balance after a default. Sengupta himself acknowledged that digital claims must be supported by enforceable legal rights, since blockchain records alone cannot guarantee ownership or recovery of physical assets.
Market Implications: What This Means for Your Portfolio
If compute-backed lending scales, it could create persistent new demand for stablecoins as the funding currency for AI infrastructure — a shift that would tie crypto’s fortunes even more closely to the AI spending boom. Yield seekers should note that USD.AI reported a 6.9% net annualized yield on sUSDai in its September update, well above typical bank savings rates, but those yields come bundled with the default and liquidity risks described above.
The stakes extend beyond crypto. CoreWeave, one of the largest AI cloud providers, secured a 2.6 billion USD credit facility with a roughly five-year maturity backed by customer contracts averaging about three years — a maturity mismatch that leaves lenders exposed if contracts are not renewed. A transparent on-chain market for computing claims could make such risks visible to everyone, rather than buried in bond documents.
The Verdict
Turning AI chips into bankable collateral is one of the most consequential ideas at the intersection of crypto and AI right now. The demand side — 26.5 billion USD in requests at a single platform — is proven. The technology to tokenize and track the claims exists today. What remains unproven is whether blockchain rails can enforce recovery of physical assets when borrowers fail, and whether retail investors truly understand the liquidity constraints they are accepting. Watch this space closely, but do your homework before assuming a 6.9% yield comes without strings attached.
The cryptocurrency market remains highly volatile. This article is for informational purposes only and does not constitute financial advice.
borrowing against GPUs is wild when H100s depreciate like used cars. the lender ends up holding the bag on last gen silicon
419B already borrowed for AI and the plan is to fund more of it with stablecoins against depreciating hardware. whats the haircut gonna be, 70 pct?
70 pct haircut is generous tbh, an H100 two gens from now is a space heater. oracles gonna love repricing that every month
collateral that sheds value every hardware generation is an amortizing loan with extra steps. lenders will want coupons way above anything t bill arbitrage pays
Senguptas note is from Oct 8 and everyone acts like its new. still, 26.5B queue waiting for collateral rails is the real number here
419B in AI bonds and they still want more from stablecoins. jensen pulled a similar pitch last year and it went nowhere, collateralized GPUs are a repo market not a blockchain problem
the jensen comparison undersells this imo, last year was a slide deck, now there is a 26.5B queue with names attached. borrowers waiting in line changes the pitch entirely
26.5B queue is demand for dollars not demand for chains. whoever nails gpu backed lending wins the real world asset crown imo
demand for dollars still ends up settling on some chain tho. whoever books this flow prints the fees, chain agnostic right up until the volume shows up
fees only print until the next chain undercuts you by 10 bps. the 26.5B queue shops for the cheapest rails same as everyone else, loyalty in this market is zero