Nvidia Stumbles as Memory Chips Seize the AI Profit Crown
Nvidia's shares have slipped 15% as GPU prices cool and memory makers bank a tenfold windfall. The compute economics of AI are quietly shifting, and India is watching.
The News
Nvidia, the company whose graphics chips became the backbone of the AI boom, is learning that market dominance can curdle into vulnerability. Its shares have fallen roughly 15% from a peak set in May 2026, and the stock now trades below the average S&P 500 valuation measured against projected earnings. That is a striking demotion for a firm that spent much of 2025 as the market's most coveted name.
The reversal, laid out by TechCrunch AI editor Russell Brandom, has a clear culprit: the compute marketplace Nvidia itself brought into being. Spot rental prices for its H100 accelerators have slid from a peak near $3.20 an hour in May 2026, as the frantic GPU shortage of 2025 finally eases. Rival silicon from Google, Amazon, Microsoft and OpenAI, all now shipping custom chips, has loosened Nvidia's grip on the biggest buyers.
Meanwhile the profits have migrated elsewhere. Prices for DRAM, the ordinary memory that every data centre needs in bulk, have jumped tenfold since August 2025, and Micron's share price has nearly tripled over the same stretch that Nvidia slipped. "More GPU and accelerator players are entering the market," Ornn co-founder Wayne Nelms told the publication, adding that "no one is making their own DRAM."
Why It Matters
The irony is hard to miss. Nvidia's accelerators represent some of the most advanced engineering on the planet, while DRAM leans on decades-old incremental refinement. Yet it is the humbler component, suddenly scarce, that is minting fortunes. Value in a supply chain, this episode reminds investors, tends to pool wherever the bottleneck sits, not wherever the cleverest technology lives.
There is a historical rhyme here. When railways were built out in the nineteenth century, the operators who laid the track often struggled while the suppliers of steel and rolling stock prospered. The same pattern surfaced in the dot-com fibre glut, when bandwidth turned briefly worthless even as router makers thrived. Nvidia is not collapsing; it remains enormously profitable. But the market is repricing the lazy assumption that owning the flashiest layer of the AI stack guarantees the fattest margins.
For the wider industry, cheaper GPU time is not bad news at all. Falling rental costs lower the barrier to training and serving large models, which should speed the flow of AI products from anyone who is a buyer of compute rather than a seller of it.
Indian Angle
For India, the memory-versus-logic story lands close to home. Micron, the very company riding the DRAM surge, is building a roughly $2.75 billion assembly and test plant in Sanand, Gujarat, backed by the India Semiconductor Mission. A world in which memory commands premium prices strengthens the economic case for that facility and for the packaging ecosystem the government hopes will grow around it.
Cheaper GPU compute, meanwhile, is a quiet gift to India's model builders. Startups such as Sarvam and Krutrim, along with the many fintech and enterprise firms fine-tuning models on rented accelerators, spend precious dollars on cloud GPU hours. Every dip in H100 pricing eases their burn rate and narrows the cost gap with better-funded Western rivals, a meaningful edge when your funding is counted in rupees.
Indian data-centre operators and the hyperscalers expanding across Mumbai, Hyderabad and Chennai will feel the shift too. If memory, not GPUs, becomes the scarce input, capital expenditure plans and procurement strategies may need rewriting, and the RBI-regulated lenders financing these builds will want to understand where the real pricing power now sits.
FAQ
Why has Nvidia's stock fallen?
Its shares are down about 15% from a May 2026 peak as the 2025 GPU shortage eases and rivals ship custom chips. The stock now trades below the average S&P 500 valuation against projected earnings, a notable derating for a former market darling.
Who is profiting instead?
Memory makers. Prices for DRAM have risen tenfold since August 2025, and Micron's shares have nearly tripled over the same period that Nvidia declined, despite memory relying on far older technology than cutting-edge GPUs.
Does this help Indian AI startups?
Yes. Cheaper GPU rental time lowers the cost of training and running models, easing budgets for firms such as Sarvam and Krutrim that pay for cloud compute in dollars while raising capital in rupees.
Where can I read the original coverage?
TechCrunch published the full analysis by AI editor Russell Brandom. The link appears in the source attribution paragraph directly below this section.
This story was reported by TechCrunch. Read the full original coverage at TechCrunch.