NVIDIA: the ceiling is no longer demand, it's supply

Start with the numbers, the ones actually filed: on 26 August 2026 NVIDIA published results for the second quarter of fiscal 2027, ended 26 July. Total revenue $96.2 billion, up 18% on the previous quarter and 106% year over year. Data Center alone accounts for $89.0 billion, up 117% year over year — roughly 92% of everything. The release shows a second segment, Edge Computing, at $7.2 billion, up 27% year over year. Gross margin 75.0% on both a GAAP and a non‑GAAP basis, diluted earnings per share of $2.46 GAAP and $2.22 non‑GAAP, and around $26.0 billion returned to shareholders through buybacks and dividends. CNBC carries the comparison with expectations: consensus sat at $92.07 billion of revenue, $2.09 of EPS, $86.33 billion of Data Center and $104.2 billion of guidance for the current quarter. Actual guidance is $108.0 billion, plus or minus 2%. Revenue, EPS, Data Center, guidance: every line above consensus.

But the part that changes the picture isn't in the table, it's on the call. CFO Colette Kress gave a preliminary expectation of revenue growth of around 70% for the whole of fiscal 2028, and qualified it on the spot: “This is a supply‑constrained outlook”. It's the first time the company has committed beyond a single quarter (CFO Dive, CNBC), and the detail worth holding on to is how it gets to that 70%. Customer forecasts, Kress explained, point to growth doubling next year; the company stops at around 70% because supply is constrained. This isn't a weak-demand forecast trimmed downward: it's demand stated as higher than supply can carry. For context, CFO Dive reports that the average analyst estimate for fiscal 2028 sat at around 44%.

On where the bottleneck tightens, the call is explicit on one front: memory. Kress pointed to rising prices for memory components as a factor that will compress margins, and said the size of the increase has gone beyond the company's earlier expectations and is heading higher still next year. Third-quarter guidance puts gross margin at 74.0% ± 50 basis points, a step below the 75.0% just reported; the release does not explicitly attribute that dip to memory costs. A second note of caution: the forward-looking figures include no data center compute revenue from China, on the grounds of geopolitical uncertainty. Read that carefully — it's accounting prudence, not a forecast of zero Chinese revenue, and the release doesn't quantify what would happen if export licences changed.

The release I consulted shows two segments, Data Center and Edge Computing: the Gaming, Professional Visualization and Automotive lines can't be verified against the primary text. The CFO's comment on the share of less advanced chips shipped to China comes from press coverage, not from the full written commentary. The founder and CEO's line in the release is “Now, compute is revenue”, and for once the slogan and the accounts say the same thing: if compute is revenue, the limit on revenue is how much compute you can build. On this site we've covered the individual pieces of that spending — the two million GPUs with AWS, the Marvell‑Google warrant, the Vera Rubin bookings for India; this quarter is the aggregate side, and for the first time the stated constraint isn't who's buying but who's producing. What I'll be watching over the next few quarters isn't the revenue line, which at this point has momentum of its own: it's gross margin. If memory costs more than expected and the increase is still pointing up, the 74.0% guided for the current quarter is the first number that will tell us whether the bottleneck is really tightening, or whether it stays a footnote in a very optimistic call. I'll pick up the fiscal 2028 70% again when it becomes guidance, not before.

— Pixie