The AI boom was supposed to follow a familiar technology cycle. More demand would spark more production, supply chains would catch up, and prices would gradually fall as hardware aged. That’s how semiconductors have behaved for decades. Yet AI infrastructure is rewriting those rules in real time.
Demand for computing power has accelerated so quickly that even older chips are rising in price instead of depreciating. And when a three-year-old graphics processor suddenly becomes more expensive, investors should pay attention. That is not normal semiconductor behavior — yet that’s exactly the surprising signal Nvidia (NASDAQ:NVDA | NVDA Price Prediction) just revealed.
Nvidia’s Triple Play Keeps Rolling
Nvidia’s earnings report last week delivered what has become a familiar pattern for shareholders — another “triple play.” The AI chipmaker beat Wall Street’s revenue estimates, topped earnings expectations, and raised forward guidance all in the same report.
Revenue climbed to another record as hyperscalers including Microsoft (NASDAQ:MSFT), Amazon (NASDAQ:AMZN), Alphabet (NASDAQ:GOOG), and Meta Platforms (NASDAQ:META) continued spending hundreds of billions of dollars on AI infrastructure.
But the most revealing detail came from CFO Colette Kress. She noted that rental prices for Nvidia’s H100 GPUs have risen 20% so far in 2026 while older A100 GPU rental prices climbed 15%. That is the kind of pricing behavior investors normally see during commodity shortages — not in aging semiconductor hardware.
Let’s put that in perspective.
- The A100 launched in 2020 based on Nvidia’s Ampere architecture
- The H100 debuted in 2022 using the Hopper architecture
- Nvidia has already moved on to newer Blackwell GPUs in 2025 and 2026
In a normal chip cycle, older hardware gets cheaper as new generations arrive. Surprisingly, AI demand has flipped that dynamic upside down.
Why Older GPUs Are Getting More Valuable
The H100 remains one of the most widely deployed AI training chips in the world. Major cloud providers still rely on it to train large language models and run inference workloads at scale. The older A100, meanwhile, continues serving enterprise AI customers, academic research labs, and inference applications that don’t require the bleeding-edge performance of newer Blackwell systems.
The problem is simple, though: there are not enough GPUs to go around. Current-generation Blackwell chips are being absorbed almost immediately by hyperscalers building AI data centers at historic speed. That leaves smaller cloud providers, startups, and enterprises scrambling for whatever capacity they can secure — including older Hopper and Ampere GPUs.
In any case, the shortages are no longer isolated to one generation of hardware. The bottleneck now stretches across the entire AI compute stack. And GPUs are only one part of the squeeze.
AI infrastructure demand has also created pressure across:
- High-bandwidth memory supplied by Micron Technology (NASDAQ:MU) and SK hynix
- CPU “scrap” from Intel (NASDAQ:INTC)
- Networking equipment from Broadcom (NASDAQ:AVGO) and Arista Networks (NASDAQ:ANET)
- Power generation and electricity transmission infrastructure
- Cooling systems and liquid thermal management equipment
When all is said and done, the AI boom increasingly looks less like a software cycle and more like a global industrial buildout.
The Opportunity — And the Risk
For Nvidia shareholders, this demand surge remains enormously profitable. Rising rental prices for older GPUs suggest Nvidia’s installed base retains value longer than critics expected. Skeptics like billionaire investor Michael Burry argued AI accelerators should depreciate more rapidly than companies are accounting for, as newer chips emerge every two years. So far, the opposite is happening.
That also raises an interesting question about Moore’s Law. Traditional computing hardware tends to become cheaper and more powerful over time. Yet GPUs tied to AI workloads appear to be behaving differently because software demand is scaling faster than manufacturing capacity can expand. In short, AI demand is outrunning semiconductor supply.
Granted, that creates risks too. If hyperscaler spending eventually slows, the market could swing from shortage to oversupply quickly. Investors saw similar cycles hit memory chips and networking hardware in prior decades. Nvidia also faces rising competition from custom AI chips developed internally by Amazon, Alphabet, and Microsoft.
That said, the current numbers suggest the AI infrastructure cycle still has plenty of momentum.
Key Takeaway
In short, the most revealing part of Nvidia’s earnings report was not revenue growth or guidance. It was the fact that three- and five-year-old GPUs are becoming more expensive instead of cheaper. That tells investors one thing clearly: the AI bottleneck is getting worse, not better.
For Nvidia, that remains a powerful tailwind for now. But for the broader economy, it shows just how difficult — and expensive — the AI arms race is becoming.
Top Gaining Stocks
Top Losing Stocks