Fri, Nov 20S&P 5006,910.39▼ -3.70%Dow48,147.33▼ -1.93%Nasdaq22,617.73▼ -5.28%VIX35.89▲ +26.20%10-yr5.07▼ -0.59%
Bellwize

Analysis

Nvidia Fell to Fourth by Value. Is Its Grip on AI Chips Loosening?

As Alphabet overtakes Nvidia in market value and its in-house TPUs reach rival AI labs, investors are weighing whether Nvidia's hold on AI computing is actually eroding.

By Bellwize Staff · November 20, 2026, 11:28 AM ET

Bellwize Analysis weighs the public evidence around a market question. It is general information — not a forecast, and not investment advice.

An abstract pattern of glowing blue circuit-board traces on a black background.
Image by dawnydawny via Pixabay

On Friday morning Nvidia’s market value slipped below Alphabet’s, dropping the chipmaker to fourth among U.S. companies by that measure, behind Apple, Microsoft and now Alphabet too. The stock has fallen 40% from its 52-week high. The move came inside a wider retreat from anything tied to artificial intelligence: Anthropic’s shares fell about 10% on their first full day of public trading, days after the largest IPO on record, and the Nasdaq had dropped 2.4% the session before.

A year ago the notion that anything could dent Nvidia’s lead sounded fringe. That consensus has cracked. The market is now pricing a plain question: does the company that supplied the AI boom still own it?

Google is selling chips to Nvidia’s own customers

The sharpest evidence for erosion is who is buying Google’s silicon. Anthropic, among the most compute-hungry of the AI labs, has committed to as many as a million of Google’s tensor processing units, the custom chips Google designs in-house. Meta is reported to be in talks for multi-billion-dollar TPU deployments, though neither company has confirmed a signed deal. Google Cloud chief Thomas Kurian has said close to a hundred customers now run on the chips. And Google trains its own frontier Gemini models on that silicon instead of on Nvidia hardware.

The shift turns up in shipment data. Research firms tracking the market estimate custom AI-chip shipments are growing far faster than general-purpose GPUs, and New Street Research has projected Nvidia’s share of inference computing could fall toward 20% to 30% by 2028. Inference, the day-to-day work of running already-trained models, is the larger long-run market. The pool is narrowing.

A $107 billion quarter, almost all of it chips

Set against that is the scale Nvidia still commands. On November 17 the company reported record quarterly revenue of $115.8 billion, up 103% from a year earlier, with data-center sales of $107.6 billion, up 110%. Gross margin held near 75%. Scale like that buys time. Analysts still put Nvidia above 80% of AI-accelerator revenue, with most estimates showing the share slipping only a point or two a year.

The deeper defense is software. Nvidia’s CUDA platform, built over nearly two decades, is what the bulk of AI code is already written for, and moving a large training workload off it is slow and expensive. That lock-in is why even the challengers hedge their bets. Google still buys Nvidia chips for cloud customers who ask for them, and Anthropic runs on Amazon and Nvidia hardware alongside Google’s. Amazon and Nvidia, for their part, have outlined plans to put some two million additional Nvidia GPUs into AWS data centers through 2028.

What the data shows

As of Thursday’s close Nvidia was worth about $3.5 trillion, still narrowly ahead of Alphabet at $3.3 trillion; by Friday morning the order had flipped. Nvidia ended Thursday at $145.98, down 3.3% on the day and 40% below its 52-week high of $243.37.

The selling was broad. Nearly everything AI-linked fell together: the Nasdaq dropped 2.4% and the volatility index pushed above 30, with consumer-discretionary and communication-services names the weakest corners of the market. Alphabet fell too, down 3.5%. On a day the whole group dropped, the market-value reshuffle owed as much to a wholesale repricing of AI as to any single verdict on chips.

What would settle it

The tests are dated. Nvidia reports its next quarter in late February, and the data-center growth rate and forward guidance will show whether hyperscaler demand is still accelerating or beginning to flatten. The same hyperscalers, Alphabet, Microsoft, Amazon and Meta, give 2027 capital-spending guidance on their January and February earnings calls; the split those budgets reveal between in-house silicon and Nvidia orders is the number to read. Google’s next-generation TPUs are slated for broader availability in 2027, which will test whether the reported interest from Meta and others converts into deployed capacity.

What Friday repriced is not in dispute: the market now treats Alphabet and Nvidia as near-equals by value, something it did not do three months ago. What it has not resolved is whether that marks a durable change in who supplies AI computing or a fast crowd rotating between two expensive stocks. Nvidia’s revenue says its lead is intact; its customers’ chip orders say the lead is being tested. Both can be true at once, and the next two earnings seasons will do more to settle it than Friday’s tape did.

This is a general-market summary for information only — not investment advice, and not a recommendation regarding any security.

Filed under: analysis · semiconductors · artificial-intelligence · nvda · googl