How tokenized securities and compute are expanding the financing market

Every lending market runs on one question: what counts as collateral? Government securities underpin enormous financing markets because lenders know how they’re priced, how they move and what happens to them in a default. Real estate is far slower, but a century of property law, appraisal practice and foreclosure procedure has made it financeable anyway. The same question is now being put to tokenized securities, private shares and compute infrastructure. All three already hold real value. What’s unsettled is how much of that value a lender will lend against. Plenty is happening on the supply side. Asset managers are putting conventional securities onchain. Digital-asset firms are building lending books around assets that trade continuously. AI companies are looking for ways to finance GPUs and data-center capacity that don’t involve selling equity. Each of those expands the pool of things that could be pledged. None of them makes a lender accept anything. Acceptance gets settled in the eligibility schedule, the haircut table and the provisions governing what happens when the borrower can’t perform. That’s where the next phase of institutional adoption actually happens.

Market value and recoverable value are different numbers
An investor starts with what an asset is worth. A secured lender has to start with what it can recover, and how fast. That second number depends on how long it takes to take control, whether a quoted price is real at size, how many buyers exist on a bad day, and how short the legal path runs from default to sale. An asset can be perfectly good in a portfolio and poor collateral, for the simple reason that the market for it thins out exactly when the lender needs it. Which is why “unconventional collateral” is a useless risk bucket. The assets in it don’t behave alike. Bitcoin prices are transparent and it trades around the clock, but it moves fast and the lender’s position depends on a custody arrangement. A tokenized Treasury may reference a security every credit officer recognizes while carrying a different ownership record, different transfer mechanics and exposure to an intermediary nobody has underwritten. Private shares can sit on enormous enterprise value with no reliable way to sell them. A GPU produces cash, but what it’s worth depends on the model, its age, where it physically sits, how hard it’s been run and what contracts are attached to it. A borrower presents these as one pool. The lender is underwriting four different recovery paths. That gap is the commercial opportunity, and it’s narrower than it sounds. Once a lender can measure and control an asset with enough confidence, some fraction of its value converts into borrowing capacity. Better custody, more reliable valuation and standard documentation raise the conversion rate. They don’t change the asset. They change how much of it a lender is willing to count.

The eligible list is the risk policy
Category labels don’t survive contact with an eligibility schedule. “Digital asset,” “tokenized security” and “compute asset” each cover instruments that behave nothing alike. Institutions end up approving collateral at the instrument level, with rules attached to pricing source, liquidity, concentration, custody and transfer. One senior lender we spoke with monitors an accepted list of roughly 10 to 20 assets, testing them against measures like average daily volume and recent price behavior to decide whether each one still belongs on it. That list isn’t paperwork. It’s the risk policy, written in a form the operations team can act on. It matters most when collateral changes inside a facility. A borrower can swap one asset for another, leave reported market value untouched, and hand the lender a different liquidity profile, a different custodian and a different recovery path. Haircuts and concentration limits have to see that. A visible price doesn’t mean executable size, and a price source that works for a mark isn’t necessarily good enough for a margin call. Mixed collateral adds a quieter problem. When assets trade on different hours and are valued at different frequencies, a consolidated coverage ratio blends fresh and stale inputs. The number comes out arithmetically correct and economically misleading, which is worse than being obviously wrong. For investment firms this is a familiar portfolio problem in an unfamiliar operating environment. Correlations rise under stress, liquidity thins, and assumptions that looked conservative fail together. Heterogeneous collateral makes those dependencies harder to see, because the assets sit in different systems under different agreements.

Tokenization changes the path, not the rights
Tokenization makes ownership records and transfers programmable. It doesn’t tell a lender what has actually been pledged. Two tokens can reference the same underlying security and give their holders different rights and different exposure to third parties. A ticker and a price don’t establish collateral quality when the legal interest, the custodian and the redemption path all differ underneath. ISDA’s work on tokenized collateral shows how fast this reaches the operating agreement. Its model provisions for variation margin deal with which tokenized assets count as eligible credit support, how they transfer, and what a ledger that never closes does to a concept like the local business day. Enforceability, tax, accounting, legal title and settlement finality are still left to the parties, and those are the terms that decide whether a secured party can act when it needs the collateral. The case for tokenized collateral is still strong. DTCC has argued that moving collateral faster could shrink liquidity buffers, improve intraday funding and sharpen capital allocation. That case holds wherever the token, the legal record and the operational record say the same thing. Where they don’t, speed is a liability.

Compute is turning into a credit question
AI infrastructure puts the same problem into an entirely different asset class. GPUs and data-center capacity absorb enormous capital and their economic value is easy to see. What a lender can underwrite, and recover, is a separate question with a much longer answer. Equipment value depends on specification, age, location and resale demand. Contracted cash flow depends on customer quality, utilization and the terms of the service agreement. Possession, insurance and remarketing can matter more than any digital record of who owns what. As compute owners look for capital that isn’t equity, lenders have to decide what they are lending against: the equipment, the contracted revenue, a digital interest in either, or some combination. Each choice produces a different collateral package and a different way to lose money. Compute is also the clearest sign that this market won’t stay limited to assets that start out onchain. A physical or contractual asset can sit inside a digital financing workflow as long as its ownership, valuation and control information carry across accurately. Minting a token is the easy part. The work is connecting an existing asset to a lending process without losing the facts that determine its value under stress.

The market sorts by acceptability, not by size
The largest collateral pools won’t be built from the assets with the largest quoted values. They’ll be built from the assets lenders can price, control and realize confidently enough to offer terms worth taking. That pushes on everyone in the chain. Issuers can make an instrument easier to finance by improving transferability and legal certainty. Custodians can make control and balance information more reliable. Lenders can build an edge in assets their competitors still find inconvenient. The paths won’t converge. Tokenized Treasuries have familiar economics and a short route to broad collateral use. Bitcoin pairs deep liquidity and real institutional custody with volatility that demands active margining. Private shares and compute infrastructure may end up being the better business precisely because they’re harder, since lenders who can structure around infrequent valuation and slow recovery get paid for doing it. Advance rates, haircuts and terms will settle into a hierarchy on their own. That hierarchy is the real answer to which forms of value have become dependable collateral.

What we’ve seen operating it
Membrane builds the software institutions run their loan and collateral operations on. The Collateral Management Engine handles multi-asset collateral, configurable eligibility and margin rules, live valuation and loan-to-value monitoring, threshold alerts and collateral workflows across custody arrangements. It connects collateral positions to loan terms and operational approvals without asking anyone to replace their custodian or core systems. We have managed more than $13B in loans and currently support more than $4B in collateral each day. One thing keeps showing up in that work: expanding the collateral universe is easy to describe and hard to operate. Every new asset carries its own assumptions about price, liquidity, control and recovery. When those assumptions live in a PDF, three spreadsheets and one person’s memory, they don’t survive a volatile Tuesday. A broader collateral market only helps an institution whose operating record is as sophisticated as the assets inside it.

Request a collateral management demo
Membrane works with institutional lenders, borrowers, custodians, prime brokers and asset managers who need to monitor collateral and coordinate margin workflows across counterparties and custody arrangements. If you’re working through a complex collateral portfolio, or want to see how our Collateral Management Engine sits alongside an existing operating model, request a demo.