In the long arc of technological disruption, the assumption has always been that yesterday's tools become tomorrow's relics — but Nvidia is quietly rewriting that assumption. Despite being nearly four years old and outpaced by newer architectures, the H100 GPU has seen rental prices rise 22 percent in a single month, suggesting that the AI infrastructure buildout is broad and layered enough to sustain demand across generations of hardware. The bears who wagered on rapid chip depreciation as the crack in Nvidia's foundation may have misread not just a company, but the shape of a technological e
Nvidia's Aging H100 Chips Defy Depreciation Fears, Strengthening Bull Case
Older chips aren't becoming worthless—they're finding a stable place in the market.
Mark
So the story here is that Nvidia's old chips aren't depreciating as fast as people feared. But how much of this is just normal supply and demand? If there's a shortage of H100s, of course the price goes up.
Mimi
That's a fair question. The 22 percent monthly increase does suggest tight supply. But the broader point is that even as Nvidia launches newer, better chips, the H100 isn't becoming worthless. It's finding a stable place in the market—customers who don't need the latest generation can still use it profitably.
Luke
I want to push back on the framing here. The article says H100 prices are "up 22 percent over the last month." That's a snapshot. We don't know if that's a seasonal blip, a temporary shortage, or a genuine trend. And we're comparing to 2023 peak prices—they're still down from there, right?
Mimi
Yes, substantially down from 2023. But the point is they're not going to zero. They're not becoming obsolete.
Luke
Agreed. But "holding value better than expected" and "prices rising" are two different claims. One is about durability; the other is about a recent uptick. The article conflates them.
Mark
Does the multi-tiered pricing model actually prove Nvidia's moat is safe? Or does it just mean Nvidia is good at price discrimination?
Mimi
Both, maybe. If Nvidia can sell the same basic technology at different price points and keep both tiers profitable, that's a sign of real competitive strength. It means even if newer chips come out, the old ones don't become dead weight.
Luke
But we should note: the article doesn't give us actual numbers on how many H100s are being rented versus R100s, or what the margin is on each tier. We're inferring sustainability from price stability, which is reasonable, but it's not the same as seeing the actual unit economics.
Mark
Fair. So what's the real risk that Burry and others are worried about? Is it still live?
Mimi
The risk is that AI companies spend billions on chips, then can't generate enough revenue from AI services to pay for them. If that happens, the whole boom collapses. But if chips hold their value and can be resold or rented out, that risk shrinks.
Luke
And the evidence here—one month of price data on one index—is suggestive but not conclusive. We'd want to see this pattern hold over quarters, across different chip types, and with actual transaction volumes. One month could be noise.
The Pulse
- Skeptics, including famed short-seller Michael Burry, have argued that fast-depreciating chips would hollow out the economic logic of the entire AI boom — a thesis that has gained serious traction on Wall Street.
- The H100, now two chip generations old and competing against Nvidia's own Rubin platform, defied expectations by posting a 22% rental price increase in just one month.
- Rather than collapsing into obsolescence, older Nvidia chips appear to be finding stable footing in a multi-tiered market where budget-conscious buyers absorb what premium customers leave behind.
- Nvidia's stock trades at a price-to-earnings ratio barely above the S&P 500 average — a striking undervaluation for a company that nearly doubled both revenue and net income in its most recent quarter.
- Each failed bearish prediction tightens the case that Nvidia's competitive moat is not a temporary spike but a durable structural advantage in the AI hardware landscape.
In the long arc of technological disruption, the assumption has always been that yesterday's tools become tomorrow's relics — but Nvidia is quietly rewriting that assumption. Despite being nearly four years old and outpaced by newer architectures, the H100 GPU has seen rental prices rise 22 percent in a single month, suggesting that the AI infrastructure buildout is broad and layered enough to sustain demand across generations of hardware. The bears who wagered on rapid chip depreciation as the crack in Nvidia's foundation may have misread not just a company, but the shape of a technological era.
Nvidia has ascended to become the world's most profitable company, with a market cap exceeding $5 trillion, yet Wall Street remains strangely hesitant. The stock trades at a price-to-earnings ratio roughly in line with the broader S&P 500 — an almost paradoxical valuation for a company that nearly doubled both revenue and net income in its latest quarter. Three persistent doubts explain the caution: the semiconductor industry's historical boom-bust cycles, the threat of in-house chip development by Nvidia's own customers, and the fear that rapid hardware depreciation will undermine the economics of AI infrastructure.
That last concern has been most forcefully articulated by Michael Burry, who argued that if Nvidia's chips lose value too quickly, the entire AI buildout loses its financial justification. The logic is clean and alarming: obsolete hardware means stranded investment, and stranded investment means the AI boom was built on sand.
The market, however, is offering a rebuttal. Rental prices for the H100 GPU — a chip launched nearly four years ago and now two generations behind Nvidia's current roadmap — rose 22 percent over the past month, even as the company unveiled its next-generation Rubin platform. The H100's predecessor, the A100, continues to see strong demand as well. What's emerging is not a story of cannibalization but of stratification: premium customers pay for the newest silicon while cost-conscious operators absorb older generations at lower price points, keeping the entire product stack economically alive.
This pricing resilience strikes at the heart of the bear thesis. With strong business fundamentals projected through at least 2027, extraordinary growth already on the books, and chips holding value far better than predicted, Nvidia's valuation looks less like fair pricing and more like a systematic underestimation of its staying power. The H100's quiet comeback may be the most persuasive argument yet that Nvidia's dominance in AI hardware is not a moment — it is a structure.
Nvidia has become the world's most profitable company, with a market cap exceeding $5 trillion, yet Wall Street remains cautious. The stock trades at a price-to-earnings ratio of 29—roughly level with the S&P 500—despite the company nearly doubling revenue in its most recent quarter and doubling net income. This valuation disconnect reflects three persistent doubts among investors: the semiconductor industry's historical cyclicality will eventually slow the AI boom; competitors and Nvidia's own customers are building chips to replace its products; and depreciation of Nvidia's hardware poses an existential risk to the entire AI infrastructure buildout.
That third concern has gained particular traction. Michael Burry, the investor famous for shorting the housing market in "The Big Short," has argued that if Nvidia's chips lose value too quickly, customers won't generate enough revenue from AI services to justify the purchase price. This theory has been weaponized against hyperscalers and cloud companies alike. The logic is straightforward: if yesterday's cutting-edge GPU becomes worthless overnight, the economic foundation of the AI boom cracks.
But the data tells a different story. Rental prices for Nvidia's H100 GPU—a chip that launched nearly four years ago and is now two generations old—rose 22 percent over the past month, according to the Ornn H100 SXM Index, which tracks live cloud GPU rental costs. This happened even as Nvidia announced the Rubin platform and its R100 GPU, which will offer several times the memory of the H100 and employ a more advanced architecture. The H100's price appreciation in the face of newer competition directly contradicts the depreciation thesis.
Certainly, H100 prices have fallen substantially from their 2023 peak, when generative AI was just beginning its explosive growth phase. But the fact that they're climbing again now—years into the cycle, with superior alternatives available—suggests something more durable is at work. The older A100 GPU, the generation before the H100, remains in high demand as well. Rather than cannibalizing its own product line, Nvidia appears to have built a multi-tiered market in which premium customers pay premium prices for the newest silicon while cost-conscious buyers rent or purchase older generations at lower rates. Both tiers remain profitable and both remain in use.
This pricing resilience matters because it undermines one of the few remaining bearish arguments against Nvidia stock. The company has already doubled in size, the business fundamentals remain strong through at least 2027, and the chips themselves are holding their value far better than skeptics predicted. The stock's valuation—in line with the broad market despite extraordinary growth—suggests Wall Street has systematically underestimated Nvidia's staying power. Previous bearish calls on the stock have repeatedly failed to materialize. With the H100 proving that older chips can retain and even gain value, the case for continued Nvidia dominance in AI hardware looks harder to break.
Notable Quotes
Rather than disrupting itself with newer chips, Nvidia seems to be benefiting from a multi-tiered pricing model in which premium customers can pay premium prices for the newest chips and budget customers can pay lower prices for older chips.— Analysis from the reporting
Want the full story?
Read the original at The Motley Fool ↗
Economy & Finance