CoreWeave's Old GPUs Still Print Money: What It Means

CoreWeave, a cloud computing company that rents Nvidia GPUs to AI developers, has been under scrutiny for its heavy debt and the risk that its hardware becomes obsolete quickly. However, recent disclosures on its Q2 2026 earnings call revealed a multi-year renewal for a cluster of Nvidia A100 GPUs, extending into 2029. The A100 was introduced in 2020, meaning this architecture will generate revenue for nearly a decade.

Management explained that as older fleets finish their initial contracts, they continue to deliver returns because much of the capital cost has already been paid down. This contradicts Wall Street's assumption that older GPUs would be retired or deeply discounted. Instead, they remain online at attractive prices.

The demand for older chips is driven by scarcity of power and the 'workload cascading' effect. Newest chips handle speed-critical tasks, while older ones are used for fine-tuning, inference, and smaller models where cost matters more than performance. This creates a market where older GPUs don't need to beat new ones on speed; they just need to be cheap and reliable.

The economics of a GPU cluster change after the first lease. During the initial contract (typically 3-5 years), payments cover most of the hardware cost and debt service. By the time it expires, the remaining capital burden is low. Subsequent leases generate higher incremental returns because the heavy lifting is done.

Debt markets are starting to price in this residual value. CoreWeave's $2.6 billion debt facility has a five-year maturity, but the backing contracts average only three years. This gap suggests lenders are comfortable with the residual value beyond the initial contract. Additionally, Nvidia's partnership with major asset managers to mobilize $500 billion for AI compute infrastructure signals that GPU compute is becoming a recognized asset class.

This news challenges the core bear case against CoreWeave: that GPUs depreciate rapidly, forcing constant reinvestment. If older GPUs can generate cash flow for years beyond their initial contracts, CoreWeave's business model becomes more sustainable and profitable than previously thought.

For investors, this means CoreWeave's stock could be undervalued if the market hasn't fully priced in the residual value of its GPU fleet. The company's heavy debt load remains a risk, but the ability to extract value from second and third leases could improve its financial health over time.

This also has implications for Nvidia. By supporting the residual value of its chips through partnerships and potential guarantees, Nvidia can maintain demand for its GPUs even as newer models are released. This could stabilize Nvidia's revenue streams and justify its premium valuation.

The broader AI infrastructure sector could benefit as well. If GPU compute is treated like long-duration infrastructure, it may attract more institutional investment, lowering capital costs for companies like CoreWeave. This could accelerate the build-out of AI capacity, benefiting the entire ecosystem.

However, the thesis depends on continued demand for older chips. If AI workloads shift dramatically or power constraints ease, the residual value could diminish. Investors should watch how H100 contracts mature and whether renewal economics hold up.