Stranded Assets: The Real Risk of Compute-Backed Securities

The honest deep dive into the biggest criticism of CBS — and why the market is building the tools to answer it

Author: Arlo | Date: 2026-08-14 | Tags: stranded assets, risk, MBS, CBS, GPUs, obsolescence

What Is a Stranded Asset?

A stranded asset is an investment that loses most of its value before its expected life is over. Coal plants became stranded when gas got cheap and regulation tightened. Office towers became stranded when remote work stuck. The fear for compute-backed securities is simple:

"What if a new chip architecture arrives in three years, and today's H100s — which CBS securitised for ten — are worth scrap?"

That's the MBS comparison in one sentence: securitise long-dated cash flows from an asset that can suddenly lose its earning power, and you recreate 2008 — this time with GPUs instead of subprime mortgages.

Why the MBS Fiasco Happened

The 2008 crisis wasn't really about securitisation itself. It was about:

No price discovery — nobody could see what mortgage-backed assets were really worth
No hedging market — banks held the risk themselves, with leverage, in the dark
Ratings corruption — agencies stamped AAA on garbage (the Lou Ranieri story)
Wrong incentives — originators didn't keep the risk they sold

The question for CBS isn't "is securitisation dangerous?" — it's "do we have the safeguards that MBS lacked?"

The Stranded-Asset Scenarios for Compute

1. The Architecture Jump

A genuinely new architecture (beyond GPUs — think photonic, neuromorphic or something unknown) could make today's chips obsolete for frontier training. This is the nightmare scenario: CBS securitises 10 years of H100 revenue; in year 3 nobody wants H100s.

2. The Demand Collapse

If AI progress stalls — or inference becomes dramatically more efficient — demand for raw compute could flatten. "Compute is revenue" only holds while compute is scarce.

3. The Energy Crunch

AI factories are useless without power. If grid constraints or geopolitical shocks cut power, revenue stops even though the assets are physically fine.

4. The Overbuild

If everyone builds AI factories at once (the $500B platforms all deploying simultaneously), supply could outrun demand — the classic commodity boom-and-bust, oil in 2014 style.

The Bull Case Against Stranding

The counter-arguments, fairly stated:

Compute prices are RISING, not falling — 6-year-old GPUs trade above their purchase price; that's the opposite of a stranding curve
Inference demand is exploding — older chips don't go obsolete, they get demoted to inference, which is where the volume is
Contracts are real — hyperscalers and labs sign multi-year commitments; the cash flows are contracted, not speculative
HBM memory scarcity — the binding constraint (memory) has kept prices rising for the first time in history

The most important point: even in the 2008 comparison, the asset itself (houses) kept some value. GPUs keep earning in inference even when they stop being frontier. The stranding risk is real but partial, not total.

The Hedging Answer

This is where the new markets change the calculation. With compute futures on ICE (via Orian) and the OCPI index, CBS investors can now:

Short compute futures against long CBS positions — if prices collapse, the hedge pays
Buy puts on the OCPI index — capped downside, unlimited upside
Price the risk — the futures curve tells you what the market thinks compute will be worth in 1-5 years

MBS failed because there was no market for "I think house prices will fall." CBS has that market — and it's regulated, cleared and transparent. That doesn't eliminate stranding risk. It turns it from an unhedgeable bet into a priced, manageable risk.

What Would Make CBS Repeat 2008?

Watch for these warning signs:

Ratings inflation — CBS products getting AAA without rigorous stress tests
Originators selling 100% of risk — no skin in the game
Leverage spirals — CBS used as collateral for more borrowing
Opaque structures — investors can't see which GPUs, which contracts, which counterparties

If those four appear, the MBS analogy is fair. If not — and the market is transparent, hedged and properly rated — then CBS is closer to securitised aircraft or power plants: real assets, real cash flows, manageable risk.

The Bottom Line

Stranded-asset risk is the most serious criticism of compute-backed securities — and it deserves to be taken seriously, not waved away. The honest answer is threefold: (1) the evidence so far shows compute appreciating, not stranding; (2) inference demand gives old chips a second life; and (3) the new futures/options markets let investors hedge the tail instead of just hoping.

CBS is not risk-free. But it's no longer an unhedgeable bet — and that's the difference between a bubble and an asset class.