Wall Street Is Treating Nvidia GPUs Like Forever Cash-Flow Machines. Here’s What Shatters That Illusion

The AI boom is changing an old rule of technology investing: hardware is supposed to get cheaper, less useful, and eventually worthless. That assumption helped investors model data center equipment as a wasting asset with a five- or six-year life. 

However, the market for Nvidia (NASDAQ:NVDA | NVDA Price Prediction) GPUs is starting to look different. Rental prices remain elevated, older chips are finding long-term customers, and Wall Street is building financing markets around their residual value. That creates a powerful re-rating opportunity for companies built around Nvidia hardware. It also creates a new risk: if alternative AI chips break Nvidia’s scarcity advantage, the entire asset-class thesis could unwind faster than investors expect.

GPUs Are Starting To Behave Like Assets

Silicon Data tracks GPU rental pricing across the AI-compute market and publishes daily benchmarks for A100, H100, B200, and Advanced Micro Devices‘ (NASDAQ:AMD) MI300X. Its data show that the traditional depreciation curve for Nvidia hardware has become less predictable, while newer generations have maintained pricing strength.

That matters because the neocloud business was built around depreciation. Buy a GPU, rent it out for several years, depreciate it toward zero, and replace it with something faster.

But an A100 launched in May 2020 is still generating meaningful rental economics in 2026. CoreWeave (NASDAQ:CRWV) reported $104.2 billion in backlog in its second-quarter results, followed by more than $25 billion of additional customer commitments — putting contracted demand above $129 billion.

If old Nvidia GPUs can keep producing revenue deep into their supposed retirement years, the accounting assumption and the economic reality start pulling apart.

Nvidia Is Helping Wall Street Finance The Bet

Nvidia is not merely selling chips into this market. It is helping create the financial plumbing around them.

The company announced partnerships with major financial firms to mobilize more than $500 billion of third-party capital for AI infrastructure. The structure can include Nvidia guarantees covering up to 25% of certain projects’ residual value.

Then comes another important development. CME Group and Silicon Data plan to launch compute futures, pending regulatory approval. The contracts are designed to let AI builders and cloud providers hedge compute-price risk.

That is more important than it sounds. Once a cash flow can be hedged, lenders can underwrite it with greater confidence. Once lenders become comfortable, capital gets cheaper, and cheaper capital can push the value of the underlying assets higher.

The Nvidia Monopoly Is The Weak Link

Granted, today’s rental economics are reflecting a supply squeeze. More GPUs eventually mean more competition and potentially lower rental prices. But the bigger threat is not necessarily more Nvidia GPUs — it is fewer Nvidia GPUs being required.

AMD’s MI300X already has measurable rental activity in Silicon Data’s benchmarks, while Amazon (NASDAQ:AMZN) is moving its Trainium strategy toward a broader market. CEO Andy Jassy said in June Amazon’s chips business had surpassed a $20 billion annual revenue run rate and estimated it could approach $50 billion if operated as a standalone business selling to AWS and outside customers. He also said Amazon could eventually sell Trainium racks to third parties.

That gives AI customers another way to satisfy training and inference demand without renting Nvidia GPUs.

And that is the illusion Wall Street should worry about. Nvidia GPUs may be becoming durable, financeable cash-flow assets — but the evidence is still overwhelmingly Nvidia-specific.

Key Takeaway

The asset-class thesis is real enough to matter, and CoreWeave may be one of the companies that gets re-rated as investors recognize that GPU depreciation no longer tells the whole economic story.

But investors should not confuse Nvidia’s current dominance with permanent scarcity. AMD’s expanding footprint, Amazon’s potential Trainium sales, and other custom accelerators such as Google’s TPUs create a release valve. If alternative silicon absorbs enough AI workloads, Nvidia’s residual values and rental rates could fall together.

For now, the evidence favors Nvidia and Nvidia-heavy infrastructure providers. But the biggest risk to the thesis is becoming clear: GPUs can behave like forever cash-flow machines only if customers keep wanting Nvidia’s GPUs.

Contact [email protected] for any questions or corrections.