Last year, Nvidia (NVDA) announced that it would change its two-year GPU launch cycles to annual releases, indicating its adoption of Apple’s (AAPL) one-year cadence.

Nvidia has every incentive to release new products annually as the current leader of the AI chip market, especially as its customers, who are trying to stay ahead, will be eager to buy new hardware in droves.

However, to Michael Burry, this shift supports the idea that these assets cannot remain productive for five or more years.

Burry’s argument relies on the assumption that new chip launches make older ones obsolete. But in reality, it’s the opposite.

Investor Essentials Daily:
Tuesday News-based Update
Powered by Valens Research

Nvidia (NVDA) has borrowed from Apple’s (AAPL) playbook by adopting annual launch cycles.

Back in 2024, the chip-designing giant announced that it would shift from a two-year GPU launch cycle to a one-year cadence. The rollout has taken on the same cadence and cultural weight as Apple’s annual iPhone unveiling, with each generation treated as a headline event for the entire tech world.

The faster rhythm has fueled part of the bearish narrative around AI infrastructure as mentioned in yesterday’s article.

Michael Burry has argued that hyperscalers are inflating profitability by extending the useful life of servers and GPUs.

To him, Nvidia’s move toward annual chip innovation supports the idea that these assets cannot remain productive for five or six years, especially if they become outdated after just one year.

That resonates with cautious investors because the AI industry is moving quickly. New hardware grabs headlines, and cutting-edge systems often dominate the conversation. Yet the pace of releases does not reflect how long chips remain valuable inside data centers.

Nvidia has every incentive to refresh its products annually. It leads the AI chip market and as long as chips get just a little bit faster each year, customers trying to stay ahead will buy them in droves.

This does not mean older chips fall out of service in half the time, as hyperscalers are still using hardware from six or more years ago.

Just last week, Nvidia chief financial officer Colette Kress noted on the company’s recent earnings call that its six-year-old A100 GPUs continue running at full utilization. Part of this is because alongside releasing new chips, Nvidia updates the software that runs on the chips.

Nvidia’s CUDA ecosystem continues to evolve, with driver updates that extend the performance of each generation.

Google parent Alphabet (GOOGL) offers similar evidence. The company reported that its seven- and eight-year-old custom-made TPU chips still operate at full utilization inside its cloud data centers. These chips do not train the most advanced models. They don’t need to.

Many parts of what AI is doing behind the scenes can run efficiently on older hardware.

While the big data center companies are rushing to fill their new projects with brand new chips, that doesn’t mean they’ll have to throw them all out the minute Nvidia releases its next generation.

They will continue to run older chips for many more years. This is why the hyperscalers’ decision to increase the useful lives of their technology assets seems reasonable.

“Neocloud” companies that rent GPUs to other companies have been even more explicit. CoreWeave signs new GPU leases lasting four years or longer. In a recent Bloomberg article, CoreWeave said it is comfortable depreciating hardware over six years because it is able to re-lease chips years later.

Major hyperscalers are moving in the same direction. Meta extended the useful life of its server fleet to 5.5 years.

Even as they adopt Nvidia’s newest systems, hyperscalers keep older generations active. They are also investing in substantial infrastructure around those chips: the data centers and networking systems that are designed to function for much longer than a few years.

Burry’s argument relies on the assumption that new chip launches make older ones obsolete. But in reality, it’s the opposite. Across the board, older chips are still being utilized the same way they were five years ago.

As mentioned before, Burry has short positions on Palantir (PLTR) and Nvidia, so it’s in his best interest to sow seeds of doubt about AI.

However, the reality is that all of the major tech companies today are different than they were in 2020. And it seems they’re slowly changing their accounting policies to reflect that.

Best regards,

Joel Litman & Rob Spivey
Chief Investment Officer &
Director of Research
at Valens Research