- Monday Momentum
- Posts
- Your Pension Is Buying GPUs Now
Your Pension Is Buying GPUs Now
How Nvidia and the six biggest firms on Wall Street turned AI compute into a $500 billion asset class, and why the structure built to prove this is not a bubble is made of the exact parts a bubble is made of
Justin Wright
August 17, 2026 • Est. Reading Time: 6 minutes
Happy Monday!
Nvidia CEO Jensen Huang told CNBC he approached only six firms, and none of them said no. (Source: Nvidia)
On August 10, Nvidia announced it had partnered with the six largest capital allocators on Wall Street to raise money for artificial intelligence. Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR agreed to establish financing platforms designed to mobilize more than $500 billion of third-party capital for AI infrastructure. Jensen Huang told CNBC he approached only those six firms, and not one turned him down.
The purpose of the structure is to fund the AI buildout without loading the debt onto the balance sheets of Nvidia or its customers. Outside investors put up the capital while labs, clouds, and enterprises use it to build data centers and buy Nvidia chips. Huang described the result plainly: his GPUs are now an "investable asset." Compute has become an asset class, financed like real estate or infrastructure, and drawing on the deepest pools of long-duration money in the world.
When BlackRock, Goldman, Apollo, and the others raise $500 billion, they raise it from insurance floats, sovereign wealth funds, and pension systems. This means the ultimate lender behind the AI boom is increasingly becoming your retirement account. Nvidia built this to reassure a market nervous about a bubble, and that reassurance is the worrying part.
Nvidia partnered with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to mobilize over $500 billion in third-party capital for AI infrastructure, keeping the debt off its own balance sheet. GPUs serve as collateral, and Nvidia guarantees up to 25% of their value if they depreciate. The problem: GPUs lose roughly 73% of their value in three years, the guarantee grows exactly when demand weakens, and the capital comes largely from pension and insurance money. Nvidia framed the deal as proof the AI boom is not a bubble, but its structure is a catalog of bubble mechanics.
What Nvidia Actually Built
The platforms are structured as memorandums of understanding that let outside investors finance data centers built on Nvidia hardware with debt as the central instrument. Rather than Nvidia lending to its customers directly, which is the circular-financing pattern that has worried analysts for a year, the six firms raise institutional capital and lend it to the buyers of Nvidia chips. The debt sits with the financing platforms, not with Nvidia.
The innovation, and the risk, is what backs the loans. The GPUs themselves serve as collateral. To make chips acceptable as collateral to conservative lenders, Nvidia agreed to guarantee their value: if the GPUs backing a loan fail to hold their worth, Nvidia will cover up to 25% of the shortfall. This is what turns a depreciating piece of hardware into something a pension fund can lend against. It is also what keeps Nvidia on the hook.
Forbes called it Nvidia's bet to make AI compute Wall Street's next asset class. That is exactly right, and the company is now underwriting the financial instrument that lets the world buy them.
The web of money the $500 billion structure is layered on top of. Nvidia invests in customers, who commit to buying Nvidia chips, who fund the clouds that buy more chips. (Source: Bloomberg)
The Collateral Melts
Every secured loan depends on the collateral holding its value. This is where the structure gets uncomfortable, because GPUs depreciate fast. H100-generation chips lose roughly 73% of their value over three years. That is simply the nature of a product improved every cycle. But this also means the asset backing these loans is melting even in good times. Lenders are extending long-duration credit against hardware that will be worth a fraction of its purchase price before the debt matures.
Now add the failure mode: Nvidia's guarantee kicks in precisely when GPU values fall. GPU values fall hardest when AI demand weakens and data centers sit idle. So Nvidia's financial obligations are designed to balloon at precisely the moment its business is contracting. Financiers call this "wrong way" risk, where your liabilities grow exactly when you can least afford them. If demand ever softens, distressed operators would dump GPUs onto the secondary market simultaneously, crashing prices at the very moment lenders need to recover value and Nvidia's 25% backstop comes due across the entire portfolio at once.
There is also a geopolitical accelerant, and this newsletter has been tracking it. China is ramping domestic compute aggressively and has every incentive to flood the global market with low-cost silicon. If Chinese hardware pushes prices into free-fall, the collateral behind hundreds of billions in private loans erodes far faster than the loan terms assume. The same Chinese cost pressure that pushed enterprises toward DeepSeek could, in a downturn, detonate the collateral value underpinning the American buildout.
Why This Is the Bubble Question
Nvidia did not build this in a vacuum, it built it to answer a specific accusation: that the AI industry is a bubble inflated by circular financing, in which Nvidia invests in customers who use the money to buy Nvidia chips, manufacturing demand that looks real but is partly self-funded.
The $500 billion alliance was meant to dispel that fear by bringing in genuine third-party capital, real money from real institutions, betting on real demand. Huang's message was if Blackstone and BlackRock will fund this, it is not a mirage, and that argument carries weight. These are the most sophisticated capital allocators alive, and none of them declined.
But the skeptics were not reassured, and their reasoning is hard to dismiss. Bernstein's Stacy Rasgon wrote that the move "will clearly fuel circular concerns," because Nvidia's 25% guarantee keeps the company entangled in the financing of its own demand.
Michael Burry: the $879 billion in hyperscaler commitments circling through Nvidia inflates reported revenue without multiplying real demand. (Source: Benzinga)
What This Means for Practitioners
For enterprise leaders, the financialization of compute has a practical upside in the near term: capital is about to get cheaper and more available for AI infrastructure. If you are planning a large deployment, the financing environment through 2027 will likely be the most favorable it will ever be. Use it, but write contracts that survive a downturn, because the same structure that lowers costs now transmits stress fast if demand disappoints.
For founders, the lesson is that compute is becoming a commodity you rent through financial markets, not necessarily a moat you own. When GPUs are collateralized assets financed by pension funds, access to them stops being a differentiator. Your edge has to live somewhere the debt markets cannot commoditize: proprietary data, distribution, workflow lock-in, and the product layer that we have emphasized in the past.
For everyone with a 401(k), this is the part worth understanding: you now have exposure, indirect and diffuse, to the AI buildout, whether or not you ever bought an AI stock. If the returns materialize, that exposure is a quiet benefit. If they do not, the losses do not stay on Nvidia's balance sheet. They flow to the institutions holding the paper, and behind them, to the investors those institutions manage.
The Bottom Line
Nvidia has done something genuinely impressive: it engineered a way to fund a half-trillion-dollar buildout without carrying the debt, by turning its own chips into collateral and Wall Street's deepest capital pools into the lender. If AI demand keeps compounding, it is a masterstroke that removes the single biggest constraint on growth.
The trouble is that the same design, off-balance-sheet leverage, rapidly depreciating collateral, a guarantee that grows as conditions worsen, and retirement money at the base of the stack, describes the anatomy of a bubble as precisely as it describes the anatomy of prudent capital formation. Which one it turns out to be depends entirely on whether the returns show up. Nvidia has made sure that if they do, everyone shares the upside. If they do not, the downside has already been distributed to the one group least equipped to see it coming.
In motion,
Justin Wright
If sophisticated investors financing elaborate structures is exactly what every historical bubble looked like at its peak, what evidence would actually distinguish a durable new asset class from a well-engineered bubble?
Nvidia's new $500B plan is risky but brilliant, especially for aging GPUs - TechCrunch
Nvidia found a new way to keep the AI boom funded: your retirement money - Fortune
Nvidia's $500B Bet To Make AI Compute Wall Street's Next Asset Class - Forbes
Nvidia Uses $500 Billion Financing Initiative to Dispel AI Bubble Fears - PYMNTS
Why Jensen Huang's $500 billion AI financing plan faces a big risk from China - CNBC
Anthropic strikes $9 billion deal with cloud computing firm Riot - DataCenterDynamics
Michael Burry Calls Nvidia's $500 Billion AI Financing Push a 'Wall Street Stunt' - Benzinga
AI Circular Deals: How Microsoft, OpenAI and Nvidia Keep Paying Each Other - Bloomberg
Quick Hits
Anthropic launched Theseus Infrastructure with Macquarie and GIC, an off-balance-sheet venture that will build and lease US data centers with Anthropic as anchor tenant, then signed a $9.1 billion, 20-year deal with Riot Platforms for 191MW in Texas. (CNBC)
Gemini crossed 1 billion monthly active users on August 11, driven by deep integration across Search, Gmail, and Android. (Tech Startups)
SpaceXAI launched Grok 4.6 on August 12, a flagship optimized for long-running agents, coding, and multi-step tasks. (Tech Startups)
OpenAI opened GPT-5.6-Cyber to a limited group of vetted defensive security experts, expanding its Daybreak initiative weeks after its own model breached Hugging Face. (AI Weekly)
If you haven’t listened to my podcast Mostly Humans: An AI and business podcast for everyone yet, new episodes drop every week!
Episodes can be found below - please like, subscribe, and comment!