Jensen Huang Warns of AI Chip Oversupply — and Nvidia’s Margins Are About to Feel It
Jensen Huang warned investors of 30% margin compression as AI chip oversupply builds and Intel and AMD close the competitive gap through 2026.
For the past two years, buying an Nvidia H100 felt like trying to get Taylor Swift tickets — limited supply, absurd prices, and a waiting list that made grown engineers cry. That era is ending. On Nvidia’s Q4 earnings call, CEO Jensen Huang told investors what many had suspected: the AI chip market is heading into a normalization phase, foundry capacity is catching up, and bulk accelerators are about to start trading a lot closer to commodity rates.
Huang’s warning wasn’t vague. He projected margin compression of approximately 30% through 2026, driven by a combination of surging foundry output and intensifying competition from Intel and AMD. For a company that built its empire on scarcity-driven GPU pricing, that’s a significant admission.
Too Much Chip, Not Enough Scarcity
The core problem is simple: production capacity for H200 and B200 chips — Nvidia’s current flagship AI accelerators — is scaling faster than demand can absorb it. TSMC and Samsung have both ramped capacity aggressively over the past 18 months, responding to the hysteria of 2023–2024. Now those fabs are humming, and the market is shifting from a seller’s game to a buyer’s one.
Huang described it plainly on the call:
“We are seeing a normalization phase where foundry capacity is coming online and bulk AI accelerators will trade closer to commodity rates.”
— Jensen Huang, CEO, Nvidia Q4 earnings call
That’s a polite way of saying the pricing power Nvidia has wielded since GPT-4 broke the internet is fading. When your $30,000-a-unit chips start getting compared to commodity hardware, the income statement starts looking different fast.
Intel and AMD Are Actually Showing Up
The competitive picture matters here too. Intel’s Gaudi accelerators and AMD’s MI300 series have gone from “technically interesting but commercially irrelevant” to genuinely eating into Nvidia’s customer conversations. Hyperscalers — Google, Amazon, Microsoft — have also been quietly building their own silicon (Google’s TPUs, Amazon’s Trainium) specifically to reduce dependence on Nvidia. Huang acknowledged that this competitive pressure from Intel and AMD accelerators intensifies through 2026, which is precisely why the 30% margin compression isn’t a one-quarter blip.
To be clear, Nvidia isn’t about to become a budget brand. But the window where they could charge whatever they liked because nobody else had anything competitive? That window is closing.
Nvidia’s Answer: Go Custom, Go Enterprise
Nvidia’s strategic response is to pivot away from the commodity GPU race and lean into custom silicon for specific enterprise clients, alongside its software ecosystem — CUDA, Omniverse, and enterprise AI platforms. The idea is to move up the value stack before the bulk accelerator market turns into a margin grinder.
It’s a reasonable play. CUDA alone has moat-like properties; a decade of developer lock-in doesn’t evaporate just because AMD drops a competitive chip. And custom silicon deals with hyperscalers tend to come with longer contracts and higher switching costs than off-the-shelf GPU purchases. Huang is essentially arguing that the commodity layer gets commoditized, but Nvidia won’t live there.
What This Means for the Broader Market
The ripple effects go beyond Nvidia’s income statement. If AI accelerator prices compress meaningfully, that’s actually good news for companies buying the hardware — cloud providers, enterprises, and AI labs all benefit from falling GPU costs. The irony is that cheaper chips could accelerate AI deployment even as it deflates the chip suppliers’ profits.
For investors, however, a 30% margin compression warning from the dominant player in a sector is a real signal worth taking seriously. Nvidia’s valuation has baked in years of premium pricing power. If that assumption softens, the math gets uncomfortable fast. The company is betting its software pivot happens fast enough to offset the hardware margin slide — which is a reasonable bet, but still a bet.
The AI chip gold rush of 2023–2024 produced extraordinary profits for one company that happened to have the right GPU at the right moment. What comes next looks a lot more like a normal, competitive semiconductor market — which is fine, just not nearly as exciting for anyone holding Nvidia stock at a 30x revenue multiple.
