Nvidia Guided $108B for One Quarter. Read Which Customers Are Carrying It

Nvidia guided $108 billion of revenue for its fiscal third quarter, its first quarterly forecast above $100 billion. The detail that matters sits one level down. Nvidia's CFO said the next quarter's growth comes primarily from AI clouds, sovereigns and enterprises, while hyperscaler growth reaccelerates later. For anyone reading Nvidia earnings as a gauge of AI infrastructure demand, the buyers holding up the guide have changed.

What Nvidia guided for Q3, and what it left out

Nvidia reported $96.2 billion of revenue for the quarter ended 26 July 2026, up 106% on a year earlier. Data center revenue was $89.0 billion of that. The company then guided the October quarter to $108.0 billion, plus or minus 2%, with no data center compute revenue from China assumed.

That guide implies roughly $11.8 billion of sequential growth. The press release does not say where it comes from. The answer is in the CFO's prepared remarks and in a disclosure Nvidia only started publishing this year.

Since fiscal Q1 2027, Nvidia has reported data center revenue in two lines. Hyperscale covers the largest cloud providers. ACIE stands for AI Clouds, Industrial and Enterprise, and covers the neoclouds, sovereign buyers and enterprises that buy outside the big four or five. Before this year, investors had to infer that split from scattered call commentary.

The split is the single most useful disclosure Nvidia has added in years, and most coverage of the quarter skipped past it. The headline revenue number tells you demand is large. The split tells you whose demand it is, which is the question that decides how durable it is.

Nvidia data center revenue, split by buyer

The 10-Q gives three comparable quarters. Read them as a sequence and the direction is plain. Hyperscale is still more than half of data center revenue, and its share is falling every quarter.

Dollar figures and growth rates are from Nvidia's Form 10-Q for the quarter ended 26 July 2026. Shares are calculated by Zan Digital. Nvidia reclassified one company from ACIE to hyperscale during the year and restated prior periods, so these are the restated figures.

Both halves more than doubled in a year. That is worth stating before anything else, because a story about shifting mix can read like a story about hyperscalers pulling back. They are not. Hyperscale revenue grew $5.7 billion in a single quarter.

What changed is the rate. ACIE grew almost twice as fast as hyperscale on the quarter. It took 59% of the sequential increase in data center revenue while being 43% of the base. If that gap holds for two more quarters, the two lines meet.

The table and chart cover only data center. Nvidia's other segment, Edge Computing, was $7.2 billion in Q2 and grew 13% on the quarter. It matters for the arithmetic of the guide, and it is not where the argument is.

Who carries the $108B Q3 guide

The CFO's own sentence

Nvidia did not leave this to inference. "We expect sequential growth to be driven primarily by ACIE with data center," Colette Kress said on the 26 August call, according to the published call transcript. The rest of her sentence covered hyperscale: "while growth in hyperscale is expected to reaccelerate in Q4 and into fiscal year '28, as supply of Vera Rubin grows over time".

The reason given is supply, not hyperscaler appetite. Kress said Vera Rubin production shipments began in August and that purchase orders had arrived from every major hyperscaler. She expects Vera Rubin to be about 20% of data center revenue in Q3. Hyperscalers are, in effect, queuing for the next generation while the newer buyers absorb what ships now.

The arithmetic it implies

Nvidia does not guide by customer group, so any split of the $108 billion is an estimate. The table below is ours, built on two stated assumptions. Edge Computing holds flat at $7.2 billion, leaving about $100.8 billion for data center. Then hyperscale growth is varied across a plausible range.

These are scenarios. They are not forecasts or reported figures. They assume revenue lands at the $108 billion midpoint and Edge Computing is flat. If Edge grows as it did in Q2, data center gets slightly less of the total and every ACIE figure falls by a few hundred million.

Two things fall out of it. First, "primarily ACIE" is consistent with hyperscale growing at almost any single-digit rate. Even at 10%, ACIE contributes the larger share of the dollar increase. Second, Q3 is the first quarter in which ACIE could plausibly match or pass hyperscale outright. It depends on a number Nvidia has not given.

What ACIE actually contains

ACIE is a residual category. It is everything in data center that is not a hyperscaler. That makes it easy to misread as "enterprise demand", which it mostly is not. The call and the 10-Q describe three different buyers inside it.

Neoclouds renting capacity to other people

The largest component is neoclouds: specialist GPU cloud operators that buy Nvidia systems and rent them by the hour or on contract. Kress said Nvidia's neocloud partners are expected to exit the year with 8 gigawatts of installed capacity, up from about 3 gigawatts at the end of 2025. A neocloud's demand for chips is derived demand. It buys because it has signed, or expects to sign, a customer for the capacity.

That makes the financing of neoclouds part of Nvidia's demand story. The debt structures behind the largest of them are examined in the breakdown of CoreWeave's debt stack and where it is stretched.

Sovereign buyers, mostly through regional clouds

Kress said Nvidia's sovereign AI business, which runs "primarily through the regional NeoClouds," grew 35% on the quarter and more than tripled on the year. Governments rarely buy racks directly. They fund or contract a national cloud operator, which then buys from Nvidia and shows up in ACIE. Where that money originates is covered in the analysis of sovereign AI funds as a source of compute demand.

Hyperscalers buying from someone else

This is the detail that changes how you read the split. The 10-Q says ACIE growth was driven by AI natives, enterprises and sovereigns, "as well as hyperscalers utilizing AI clouds". Kress described it on the call as "hyperscalers purchasing capacity to supplement their own build-outs".

So part of ACIE is hyperscaler demand, routed through a neocloud's balance sheet instead of the hyperscaler's own. The split shows who Nvidia invoices and who owns the asset. It does not show who ultimately consumes the compute. Our view is that the hyperscale share understates how dependent Nvidia remains on the largest clouds, and nobody outside Nvidia can say by how much.

How concentrated Nvidia's demand really is

Concentration in Nvidia's filings comes in two layers, and the headline number belongs to the layer that matters least.

Direct customers: falling, and mostly middlemen

For fiscal Q2 2027, one direct customer represented 16% of total revenue, according to the 10-Q. A year earlier, two direct customers made up 23% and 16%. For the first half of fiscal 2027, three direct customers each passed 10%, at 16%, 15% and 13%.

On the surface that is deconcentration. But Nvidia's direct customers are largely the original design manufacturers, OEMs and distributors that assemble and ship servers. They are the loading dock, not the buyer. A falling top-customer share can simply mean assembly work spread across more contract manufacturers.

Receivables tell the same story from the other side. Five direct customers accounted for 22%, 14%, 13%, 11% and 10% of accounts receivable at 26 July 2026, which is 70% combined. Days sales outstanding rose to 60, which Kress attributed to extended payment terms for large purchases shipping over several quarters.

Indirect customers: where the risk actually sits

Nvidia defines indirect customers as those who buy through its direct customers, including cloud providers, AI clouds, model makers, enterprises and public sector entities. The 10-Q states that revenue "is concentrated among a limited number of direct and indirect customers and this trend may continue". It notes that some indirect customers individually represent 10% or more of revenue. It adds that an AI research and deployment company contributed a meaningful amount of revenue by buying cloud services from Nvidia's customers.

That last line is the concentration that matters. A lab renting GPUs from a neocloud shows up nowhere in the direct customer list and possibly in ACIE. Kress went further on the call: Nvidia expects demand from AI labs it supports with its balance sheet "to contribute toward roughly a quarter of our business next year". That is a stated dependency on a handful of buyers, and it is larger than any direct customer figure in the filing.

The balance sheet behind the newer buyers

Hyperscalers pay for Nvidia systems from their own operating cash flow. Much of ACIE cannot. A neocloud or a lab needs lenders, and lenders need a contracted customer. Nvidia has started supplying part of that comfort itself, and the 10-Q quantifies it.

The new structure, introduced in Q2, works like this. Nvidia sells systems to an AI cloud and commits to buy cloud services back, a take-or-pay floor that lets the project raise debt. Nvidia then shares in rental revenue above that floor. The 10-Q puts these commitments at $36 billion, typically six years in duration. Kress said on the call, "We're not making loans."

The guarantees are larger. In August 2026 Nvidia agreed guarantees capped at $105 billion supporting leases by SB Energy for about 4.25 gigawatts at a campus in Pike County, Ohio, on behalf of an affiliate of OpenAI. With smaller land, power and shell guarantees for AI clouds, the 10-Q lists maximum gross exposure of $108.5 billion.

None of this is hidden. Nvidia disclosed it in detail and addressed it on the call: "we know some will call this circular financing. We see it differently". Kress argued the risk is limited because Nvidia compute "can be redeployed to support other customers". The mechanics of vendor support for customers, and the historical cases where it went badly, are covered in the breakdown of how the circular deals actually work.

Our position is narrower than the circular financing critique. The structures may be sound. But they mean the growth line in the Q3 guide is the line where Nvidia has the most of its own credit at stake. That should change how much weight you put on it.

Where this argument is weakest

The split is Nvidia's classification, and it moves

The 10-Q records one reclassification. Nvidia moved one company from ACIE to hyperscale "due to a change in their business model" and restated prior periods. One reclassification can move billions between the lines. The categories are Nvidia's own and are new, so a trend built on three quarters of a restated series is thinner than it looks. We are reading direction, not precision.

Supply, not demand, may explain the whole mix shift

The strongest counter-case is Nvidia's own. Kress tied hyperscale's slower Q3 to Vera Rubin supply, and said customer forecasts point to revenue doubling next year while Nvidia expects about 70% because it is supply constrained. On that reading, hyperscalers are not less committed. They are waiting for a specific product, and ACIE is buying what is available. If Q4 hyperscale growth reaccelerates as guided, this post's emphasis will look overdone. We would accept that outcome as evidence against our framing.

The hyperscalers' own plans support the counter-case too. Kress cited capex by the top five hyperscalers of nearly $800 billion in 2026 and an expected $1.3 trillion in 2027. Those are their plans as Nvidia relayed them, not audited outcomes, but they do not describe a customer group in retreat. Whether those budgets pay back is a separate question, taken up in the payback math behind hyperscaler AI capex.

What to watch in the Q3 report

Nvidia's fiscal Q3 report will arrive in November. Three lines in it settle most of the questions above, and none of them is the headline revenue figure.

The second row is the one we would read first. If ACIE grows and the commitments behind it grow faster, the growth is partly underwritten by the seller. If ACIE grows and commitments stay flat, the newer buyers are standing on their own, and the bull case gets materially stronger.

Frequently asked questions

Where to start

If you track AI infrastructure demand for a living, add one line to your model this week. Put ACIE revenue next to Nvidia's AI cloud commitments and guarantee exposure, quarter by quarter. Two quarters of that pairing will tell you more about the durability of the build than the headline guide.

If you buy compute rather than analyse it, ask your neocloud provider one question at renewal. Who else has contracted the capacity you are using, and on what term? A provider whose floor is guaranteed by its chip supplier carries a different risk from one whose floor is a signed customer, and you deserve to know which you are renting from.

Related reading