CoreWeave earnings show Nvidia AI chips last years longer than expected

AI cloud provider CoreWeave delivered evidence this week that addresses one of the biggest concerns around artificial intelligence infrastructure investment: how long expensive chips remain economically useful.

The company reported that it recently signed a contract extending through 2029 for Nvidia A100 GPUs—chips introduced in 2020. CoreWeave CFO Nitin Agrawal disclosed the deal during the company's earnings call Tuesday, noting the contract carries "an attractive price" despite the hardware being six years old.

"Older generations of GPUs are going to have a longer useful life than anyone anticipated," CEO Mike Intrator told CNBC Wednesday morning. "They are going to contract for a longer term, and they are going to contract at a higher price."

The A100 belongs to Nvidia's Ampere generation—the same chip OpenAI used to train the original ChatGPT in 2022. Since then, Nvidia has released three newer GPU families: Hopper, Blackwell, and Rubin, which entered full production earlier this year.

Why it matters

The extended economic life of AI chips directly counters a major bear case that has shadowed the sector: technological obsolescence. Critics have argued that rapid hardware advancement would force frequent, costly replacements, making current infrastructure spending irrational. CoreWeave's contracts demonstrate the opposite—older chips retain value even as newer generations arrive, extending the return-on-investment window for data center operators and making massive capital expenditure cycles more sustainable.

Implications for hyperscale spending

Amazon CEO Andy Jassy recently revealed that the company's AI servers break even in under three years but have useful lives of five to six years, with most capacity contracted for at least five-year terms. CoreWeave's experience suggests the actual productive lifespan may extend well beyond those initial projections.

If hyperscale cloud providers—Amazon, Microsoft, Google, Oracle, and Meta—can generate returns from chips for longer than required to justify the original purchase, each additional year represents upside that wasn't factored into investment decisions. CoreWeave's CFO confirmed the company "built a business whose economics do not rely on re-contracting after initial customer term," yet is "increasingly seeing longer utilization at higher prices."

Intrator attributed the extended chip longevity to three factors: Nvidia's hardware quality, the CUDA software platform that makes chips fungible across different workloads, and cloud delivery infrastructure that maximizes utilization.

Supporting new financing models

The chip longevity data arrives as Nvidia announced a $500 billion financing initiative Monday with BlackRock, Blackstone, Brookfield Asset Management, Goldman Sachs, and KKR. The partnership aims to create compute-backed securities—financial instruments similar to mortgage-backed securities—that use data centers as cash-generating collateral.

For such securities to work, investors need confidence that the underlying assets will maintain value and generate returns over extended periods. Longer chip lifespans reduce replacement frequency and increase predictable cash flows, making data centers more attractive as collateral.

Intrator said CoreWeave has "excellent visibility to our target of at least 8 gigawatts by 2030" and expects "demand to meaningfully exceed supply for years."

The company's strong quarterly results—revenue above expectations, narrower losses, and raised full-year guidance—sent its stock up nearly 20% Wednesday. Nvidia shares rose 3%, while suppliers including Corning (up 5.2%), GE Vernova (up 2.7%), and Micron (up 4.9%) also gained.

These details were first reported by CNBC.