The Demand Half Holds Up
Nvidia reported Q1 FY2027 on May 20. Total revenue of $81.6B, up 85% year over year and 20% sequentially. Data center revenue of $75.2B, up 92% year over year and 21% sequentially, driven by the ramp of Blackwell 300. GAAP gross margin 74.9%, non-GAAP 75.0% (NVIDIA official 8-K, 2026-05-20).
A company already clearing more than $80B in quarterly revenue is still growing its data center business at close to double. That is what the market bought: demand is no constraint, and the AI capex cycle has a long way to run.
Back in mid-April, our piece “AI’s Silicon Bottleneck” made one claim — demand is not the constraint, capital is not the constraint, silicon is. This quarter settles the first half. Demand really is not the problem. The half we actually came to check is the second one, and it shows up in a few numbers nobody was watching.
The $80B Buyback Is Nvidia Voting Against Itself
Three more numbers sit in the same release, and for all the noise around the print, nobody was watching them.
The board approved an additional $80B of share repurchase authorization on May 18. The quarter returned roughly $20B to shareholders through buybacks and dividends. And in the same period, Nvidia invested $18.6B in private companies and infrastructure funds (NVIDIA official 8-K, 2026-05-20).
Put those three figures next to the 92% growth and something stops adding up.
A company still growing its data center business near 100% has enough cash to top up buyback authorization by $80B, return $20B in a single quarter, and still put $18.6B into data centers and infrastructure funds itself. The logic is blunt: if pushing more capital into its own existing GPU capacity bought proportional growth in compute and revenue, the buyback and the self-funded infrastructure would both be losing trades, and the cash would stay locked in expansion. It did not. Nvidia handed a large slug back to shareholders and routed another into infrastructure upstream and down.
This is where the April line — rising capex does not equal rising compute — surfaces in an earnings report. The market assumes the transmission coefficient from dollars to FLOPs is one. Where Nvidia actually spends its own cash says otherwise: the marginal dollar thrown back into existing capacity does not buy a marginal unit of compute, because the binding constraint has moved somewhere it cannot reach — silicon upstream, power downstream. Nvidia, by the look of its own capital allocation, seems to agree.
The A-Share Mirror, Louder This Time
For readers outside China: A-share semiconductor names trade heavily as a “domestic substitution” and “supply-tightness” complex. When Nvidia validates AI demand, Chinese investors price the local foundries, memory, and equipment makers positioned to absorb the spillover and benefit from upstream pricing power — rather than buying Nvidia directly.
April’s A-share move was the market quietly self-pricing the supply constraint. This round was not quiet.
Starting the session after Nvidia’s May 20 report, A-share semiconductors surged, foundry and memory most of all. Over the one-week window from the May 18 close to the May 25 close (Eastmoney Choice, pulled 2026-05-26):
- Domestic foundry ran roughly +30% — the core exposure for absorbing demand spillover
- Memory interface and NOR moved about +29% — the HBM-adjacent pricing story
- Semiconductor equipment added about +17% — the picks-and-shovels play in an expansion cycle
- DDR5 and interconnect interface IP rose about +15%
The broad index fell about 2% on May 21, and these pockets mostly dropped with it before rebounding hard — a risk-on rising tide would not have them climbing against the market. So this is directional flow: money trading two threads, domestic substitution absorbing Nvidia’s demand spillover, and price increases from upstream tightness. Same logic as April, leaning on it harder.
US markets cheered Nvidia for validating demand; A-shares bet on the beneficiaries of supply tightness. Same capacity table, two markets — one reading the front, one reading the back.
The Other Side
Read the same logic in reverse and at least four spots deserve a discount.
Validated demand is not capped supply. This quarter’s growth ran on the Blackwell 300 ramp, and you cannot ramp something that is seizing up — the 21% sequential gain in data center is capacity getting delivered, not choked.
The $80B buyback need not mean a silicon ceiling. For a cash-rich, mature company, a buyback that size can be plain capital management and dilution offset, nothing more cryptic. Reading it as “capped marginal return” risks over-narrating. The $18.6B of self-funded infrastructure works just as well as strategic ecosystem positioning, not a confession that it cannot spend more on its own lines.
A +30% week is mostly sentiment, and everyone knows it. A one-week move of +30% in the foundry and memory-interface pockets cannot be cleanly stripped of theme and froth. Whatever the fundamentals deliver shows up later — utilization, gross margin, realized price increases in the next set of reports. Calling this “earnings delivery” today is premature.
A 75% gross margin is not the picture of a choked supplier. If silicon had hit a hard ceiling, pricing power and margins are usually the first things to bend. Nvidia held 75%, and that sits awkwardly with the “strangled from upstream” story.
All four hold. And they collapse into one question: is silicon really capped, or is this just expansion failing to keep pace?
Closing
The first half of the April thesis — demand is not the problem, capital is not the problem — this quarter settles it.
The second half gets only a sidelong glance: $80B of buyback plus $18.6B of self-funded infrastructure, with Nvidia itself apparently preferring to return the cash or build than to throw more at existing capacity. Corroboration for “transmission < 1,” not a verdict — and we are not treating it as one.
No prediction here, just a question we are leaving open: is this the physical ceiling of silicon and power finally arriving, or one more flare-up of the old timing mismatch between models that iterate by the quarter and fabs that expand by the year? Whichever side gets falsified is what the next two quarters are really about.
What to Watch
The following is an observation framework, not a trading signal.
- Whether Nvidia’s data center growth decelerates next quarter — a clear sequential slowdown that management attributes to supply would firm up the “silicon ceiling” narrative; sustained growth looks more like expansion catching up
- TSMC’s HPC revenue mix on its next call — a rising share signals the AI crowd-out on N3 is still intensifying
- HBM4 yield disclosures — every step up in yield effectively frees more wafer capacity, a direct variable for gauging supply tightness
- Whether A-share semiconductors rotate from sentiment to earnings — utilization, gross margin, and realized price increases in the next quarterly reports are the test of this +30% week’s quality
Data sources: Nvidia Q1 FY2027 figures from NVIDIA’s official 8-K (2026-05-20). Supply-side analysis carried over from SemiAnalysis “The Great AI Silicon Shortage” (2026-03-12). A-share equity data from Eastmoney Choice (pulled 2026-05-26; window from the 2026-05-18 close to the 2026-05-25 close).