Nvidia's Toughest Competition in 2028 May Be the Chips It Already Sold

Nvidia (NASDAQ:NVDA) can't build artificial intelligence (AI) hardware fast enough for its customers. But the chips it has already delivered aren't going anywhere.

Nvidia's data center business generated $47.5 billion of revenue in fiscal 2024. In fiscal 2025, that figure jumped 142% to $115.2 billion. And in fiscal 2026, it climbed another 68% to $193.7 billion (Nvidia's fiscal years end in late January). That adds up to about $309 billion of shipments across the last two of those years alone -- and nearly all of that hardware is likely still racked up and running.

How long it keeps running is something Nvidia's biggest customers estimate in their filings, and those estimates carry real money. When Meta Platforms (NASDAQ:META) raised its estimated useful life for most servers to 5.5 years in 2025, the change added $1.00 to its earnings per share for the year.

And the schedules raise an awkward question for Nvidia: What does demand look like in 2028, when the boom-era chips aren't yet due for retirement?

Image source: Getty Images.

Nvidia's buyers assume the chips last five or six years

According to its latest annual filing, Microsoft depreciates servers and network equipment over two to six years. Alphabet generally uses six years for servers and network equipment. And Meta's 5.5 years took effect at the start of 2025 and covers most of its servers and network assets. The change cut that year's depreciation expense by about $2.9 billion.

Amazon (NASDAQ:AMZN) went the other way. Its reasoning, I'd argue, is the most interesting part.

The company raised its server estimate from five years to six at the start of 2024. A year later, it reversed, cutting a subset of servers and networking equipment back to five. The shorter lives, Amazon said, are due to "the increased pace of technology development, particularly in the area of artificial intelligence and machine learning." The reversal added $1.4 billion to its 2025 depreciation and amortization expense.

What retires in 2028?

Not much of this hardware is due to come out of service in 2028. A machine bought in 2024 on a five-year clock retires in 2029 at the earliest. On a six-year clock, 2030.

In other words, nearly everything from the 2024 and 2025 spending waves should still be working in 2028. The demand Nvidia is counting on that year is almost entirely new capacity, not replacement.

Chief financial officer Colette Kress said on the company's late-August earnings call that Nvidia expects revenue to grow about 70% in fiscal 2028, which runs through late January 2028. She called that a supply constrained outlook.

Nvidia itself makes the case that the schedules are honest. On its earnings call last November, Kress said the A100 chips Nvidia shipped six years earlier were "still running at full utilization today," crediting its CUDA software.

That defense also describes the problem. A chip that stays productive is capacity Nvidia has already been paid for once -- and it competes with whatever the company wants to sell next.

Sure, the dollars can grow even if the units don't. On the August call, CEO Jensen Huang said each new generation carries more revenue per gigawatt of data center capacity (about $18 billion for Hopper, about $40 billion for the new Vera Rubin platform). And "customers want to race to the next generation as fast as they can," he said.

A paid-off chip can work for cheap

The competition gets sharper once a server finishes its schedule. With no cost left on the books, its owner can rent it out at any price that covers electricity and space. Priced that way, a 2024-vintage chip is cheap competition for inference (the everyday work of running AI models), which arguably doesn't require the newest hardware.

Of course, that market is only starting to form (a marketplace for used Nvidia chips opened this summer). And Nvidia's results suggest why. Kress said in August that Nvidia's computing capacity is fully utilized across every cloud it serves. Supply should remain a bottleneck at least through the end of fiscal 2028. Nobody sells a machine that's earning rent.

Ultimately, the disclosures themselves are worth watching. Amazon's cut says AI hardware ages out faster than planned, which would pull replacement demand forward. Meta's extension says the fleet lasts, leaving 2028 resting that much more on new construction.

As for the stock, shares trade around $230 as of this writing, at about 29 times earnings. Given the growth Nvidia has already guided for, that price strikes me as fair, and I'd still buy shares here.

But the next time the cloud companies change those useful-life estimates, the direction will matter. If they extend again, the chips Nvidia already sold are lasting longer -- and some of the demand investors expect in 2028 may take longer to show up.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Alphabet, Amazon, Meta Platforms, Microsoft, and Nvidia. The Motley Fool has a disclosure policy.