AI Compute Financing
Old GPUs get
a second life.
On August 10, 2026, NVIDIA signed memoranda of understanding (MOUs) with six major financial firms — including BlackRock and Goldman Sachs — to mobilize more than $500 billion in third-party capital for AI infrastructure. At the core of the plan is a mechanism that lets NVIDIA itself guarantee the value of aging GPUs many assumed were nearing retirement.
GPU coverage has only
tracked the newest chip
Until now, coverage of AI GPUs has mostly focused on whichever chip launched that quarter — performance benchmarks, supply shortages, the next headline number. Far less attention has gone to the installed base: the GPUs already deployed and running in the field, and what happens to them once a newer generation ships.
According to Forbes, NVIDIA signed MOUs with six major financial firms — including BlackRock and Goldman Sachs — to mobilize more than $500 billion in third-party capital for AI compute infrastructure ("AI factories"). The stated goal is to establish AI compute as a new investable asset class for Wall Street.
What turns a GPU
into an "asset"
The scale of the capital matters less than who is on the hook for the collateral's value.
Per TechCrunch's analysis, NVIDIA is agreeing to guarantee, with its own money, that the chips used as collateral in these financing deals will retain their value. That guarantee is what lets a GPU be treated not as fast-depreciating consumer electronics, but as a bankable, long-term asset.
XenoSpectrum describes the effect as the creation of a secondary market for aging GPUs. Fleets equivalent to Hopper- or Ampere-class hardware, which used to be scrapped once their generation aged out, now have an economic case for staying in service. That's the financial logic behind today's update headline: "Old-generation GPUs get a longer runway."
Who benefits, and how
Business leaders and product/infrastructure owners should take away different things.
For business leaders
With aging GPU value now financially backed, fleet depreciation schedules and lease renewals gain a new negotiating input. Instead of treating hardware as "written off" purely by age, residual value becomes something you can argue over.
For product/infra owners
You don't have to rush to the newest chip to keep an infrastructure roadmap coherent. Older GPUs can be planned as long-lived infrastructure, letting capex roadmaps be redrawn around asset lifespan rather than the generational refresh cycle.
What to do about it
Three concrete actions follow from this. First, check whether your own GPU procurement or lease contracts touch any of the MOU's financial partners (BlackRock, Goldman Sachs, and others). Second, hold off finalizing depreciation or retirement schedules for older fleets until more detail on the value-guarantee scheme is public. Third, factor China-related political risk into any long-horizon infrastructure plan.
| Old view (depreciation-first) | Post-MOU view (guarantee-backed) |
|---|---|
| Value falls toward zero within a few years | NVIDIA backs residual value with its own capital |
| Early replacement each new generation is rational | Long-term use of older hardware pencils out |
| Single-vendor hardware, illiquid | A new asset class Wall Street can trade |
Whether GPUs hold value over time
the way traditional hard assets do
is far from settled.
Not a purely rosy story
TechCrunch cautions that whether cutting-edge AI chips actually hold value over time the way traditional hard assets do is "far from settled". The value guarantee is ultimately NVIDIA's own commitment — it doesn't erase the risk of the market re-pricing that collateral down the line.
On top of that, CNBC reports that political and competitive risk tied to China is a factor CEO Jensen Huang's plan has to navigate. A shift in export controls or geopolitics could unsettle the very assumptions the collateral is built on.