Nvidia Customer Concentration: The Chart That Should Keep You Up at Night

Here is the sentence buried in Nvidia's fiscal 2026 second-quarter filing that should have gotten more attention than any earnings beat. In the quarter ended July 2025, one customer was 23% of total revenue and another was 16%. Two buyers. Nvidia won't even tell you their names. It calls them Customer A and Customer B. Together they were 39% of everything the most valuable company on Earth sold that quarter, up from a combined 25% in the same quarter a year earlier.

Let that sit for a second. The entire AI trade, the thing dragging up the S&P, the thing your index fund is now heavily exposed to whether you meant to be or not, runs through a company whose revenue rests on a number of customers you could count on one hand. That is not a footnote. That is the whole risk profile, printed in the disclosures, and mostly ignored because the growth numbers are so loud.

The chart that changed in about eighteen months

Concentration risk is one of those phrases that sounds abstract until you watch it happen quarter by quarter. So here is the actual progression, pulled straight from the filings, because the speed of it is the whole story.

As recently as fiscal 2023, Nvidia had no single customer at 10% or more of revenue. None. It was a chip company with a broad book of gamers, workstation buyers, carmakers, and a growing but still diffuse data center business. A healthy, diversified customer base by any normal standard.

By fiscal 2025, that had already shifted. One direct customer was 12% of revenue and two others were 11% each. Still spread out, but the shape was starting to lean.

Then the AI buildout went vertical, and so did the concentration. In the July 2025 quarter it was those two customers at 39%. In the October 2025 quarter, four direct customers each cleared the 10% bar, at 22%, 15%, 13%, and 11%, adding up to about 61% of Nvidia's roughly $57 billion in sales for the quarter. Four buyers, six of every ten dollars. For the April 2026 quarter, three customers came in at 21%, 17%, and 16%, roughly 54% of revenue between them.

Zoom out to the full year and the picture holds. For fiscal 2026, which ended in January, Nvidia's 10-K disclosed one direct customer at 22% of total revenue and another at 14%, both concentrated in the Compute & Networking segment. Compare that to fiscal 2025, where the top customer was 12% and two others were 11%. In a single year, the top-of-book concentration nearly doubled.

Who these customers actually are (and why Nvidia won't say)

This is where it gets genuinely interesting, and where a lot of the commentary gets it half wrong. The customers Nvidia discloses as Customer A, Customer B, and so on are its direct customers. Those are not necessarily the companies you think of when you think of AI.

Nvidia's direct customers are OEMs, system integrators, and hyperscale companies that physically assemble AI servers. Think Foxconn, Wistron, Super Micro, Quanta. They buy the GPUs, bolt them into racks, and ship finished systems. So when a filing says Customer A was 23% of revenue, that might be a contract manufacturer building boxes for several end buyers at once.

The actual demand, the money behind the money, comes from indirect customers: Microsoft, Amazon, Google, and Meta. The hyperscalers. They are the ones deciding how many GPUs get built into how many data centers. And Nvidia quietly notes something important in its filings, that some individual indirect customers also represent 10% or more of revenue. So the concentration exists at both layers. A few assemblers on the direct side, a few cloud giants on the demand side.

Why the anonymity? Partly competitive sensitivity, partly that the direct customer list is a moving target of contract manufacturers. But the effect on you as an investor is that you can't fully audit the risk. You know two customers are 39% of revenue. You can't cleanly map that back to which hyperscaler's budget is really carrying the quarter.

It all comes down to a few cloud capex budgets

Strip away the layers and here is the machine. Data center made up about 88% to 90% of Nvidia's total revenue in fiscal 2026. And Nvidia states that large cloud service providers were roughly 50% of data center revenue. Multiply it out and something like 45% of the entire company rides on a handful of cloud capex budgets.

Those budgets are, right now, enormous and growing. The four largest hyperscalers are estimated to spend around $725 billion combined on AI capital expenditure in 2026: roughly $200 billion at Amazon, $185 billion at Google, $125 billion at Meta, and $120 billion at Microsoft. That is up about 77% from an estimated $410 billion in 2025. As long as that curve keeps bending up, Nvidia's concentration looks like a feature. Its biggest customers are its fastest-growing ones.

But run the logic the other way. The same thing that makes the numbers gorgeous on the way up makes them brutal on the way down. If even one or two of those four decides 2027 is a digestion year, that it has bought enough compute for now and wants to actually earn a return on the last $200 billion, the air comes out of Nvidia's biggest revenue line fast. You don't need a recession. You need four CFOs, and really only two of them, to independently decide to slow down.

The part that should actually worry you

Here is the twist that makes this more than a standard concentration story. The customers financing Nvidia's growth are the same ones building chips to escape it.

The bulk of that $725 billion in hyperscaler spending funds Nvidia GPUs, yes. But it also funds custom silicon. Amazon has Trainium and Inferentia. Google has TPUs, now many generations deep. Microsoft has Maia. Meta has its MTIA line. Every one of these is a direct attempt to do more AI compute without paying Nvidia's margin. And Nvidia's gross margins are the fattest in the industry precisely because, for now, nobody has a real alternative for frontier training at scale.

So the concentration and the competitive threat are the same companies. Your top four customers, the ones who are more than half your revenue, are also your four most capable and best-funded future competitors. They have the demand, the cash, the engineering talent, and every incentive to design you out. That is a genuinely uncomfortable structure, and it doesn't show up anywhere in a P/E ratio.

Nvidia isn't hiding any of this. Its filings warn plainly that revenue is concentrated among a limited number of direct and indirect customers, that the trend may continue, and that some individual indirect customers represent 10% or more of revenue. When a company puts that in black and white, it is not being modest. It is telling you where the fault line is.

Why the market shrugs, and why that's rational (for now)

If this risk is so obvious, why does the stock keep making new highs? Because the numbers are, frankly, absurd in the best way. Nvidia reported record first-quarter fiscal 2027 revenue of $81.6 billion, up 85% year over year. Data center revenue alone was $75.2 billion, up 92%. GAAP net income was $58.3 billion, up 211%. When a business is compounding like that, concentration reads as momentum, not fragility.

And there's a real argument that the concentration is temporary. In a land grab, the buyers with the deepest pockets move first and biggest, so of course they dominate the customer list early. As sovereign AI projects, enterprises, second-tier clouds, and startups build out, the base could broaden and the top-heavy shape could flatten on its own. That's a plausible bull case. It just isn't the current reality, and betting on it is betting on a shape the filings do not yet show.

The way I think about it: the concentration doesn't make Nvidia a bad business. It makes it a high-beta bet on a specific thesis, which is that a small number of hyperscalers keep spending like it's an arms race for several more years. If you believe that, you should own it knowing the risk is binary-ish and correlated. If you don't, no amount of great margins fixes the fact that you're underwriting four capex committees.

What to actually watch

You don't need to guess. The concentration means the leading indicators live on the customers' income statements, not Nvidia's. Watch these, in order.

• Hyperscaler capex guidance.Microsoft, Amazon, Google, and Meta each guide on capital spending every quarter. That guidance is Nvidia's revenue two to three quarters out. The word to fear is “discipline.”
• The custom-silicon roadmaps.Every generation of TPU, Trainium, Maia, and MTIA that hits volume is a slice of demand that doesn't route through Nvidia. Watch the share of internal workloads moving to in-house chips.
• The 10-Q customer table. This is the direct read. Every quarter Nvidia discloses which customers cleared 10%, and at what share. Whether the top line is climbing toward 25% or easing back tells you if the risk is intensifying or diffusing.
• Data center as a share of total.At about 88% to 90%, there's essentially no other engine. Any sign of the other segments (gaming, automotive, pro visualization) growing back into relevance is a diversification signal worth something.

The bottom line

Customer concentration is the kind of risk that is invisible right up until the day it isn't. It costs nothing while the top customers are growing and everything the moment one of them pauses. Nvidia went from no customer above 10% of revenue in fiscal 2023 to two customers at 39% in a single quarter of 2025. That is one of the fastest concentrations of revenue into a few hands I've ever seen in a company this size.

The AI boom is real and Nvidia is its best-positioned company. Both things can be true while the structure underneath stays fragile. The entire trade is, quite literally, a few CFOs' budget decisions away from a very different quarter. That's not a reason to run. It's a reason to know precisely what you're holding, and to watch the customers more closely than the company.