Where the name sits. The operating story here is a data center infrastructure company running at a pace that has no precedent in the semiconductor industry. $NVDA's Q1 FY2027 print, released May 20, showed $81.6 billion in revenue, up 85% from the year prior and up 20% from the prior quarter. Data center revenue of $75.2 billion accounted for 92% of that total, split roughly evenly between hyperscale customers at $37.9 billion and AI cloud, industrial, and enterprise at $37.4 billion. The company guides to $91.0 billion for Q2. This is not a story about graphics cards anymore; it is a story about the acceleration of compute as a utility, and about whether one company can maintain a systems-level grip on that transition long enough to justify a $5.2 trillion market capitalization. The Blackwell 300 series is in full ramp. The Vera Rubin platform, promised at 5x the inference throughput of Blackwell, entered production earlier this year and is scheduled for customer delivery in the second half of 2026.
What the price has registered, and what it hasn't. $NVDA closed at an all-time high of $235.74 on May 14, twelve days before a print that delivered exactly what the consensus required. The stock pulled back on results day to the current range around $213.95, a roughly 9% retreat from peak despite guidance implying another 12% sequential revenue step in Q2. The 52-week range of $132.92 to $236.54 carries its own narrative: the low came in early 2025 when H20 export restrictions, combined with concerns about compute efficiency stemming from DeepSeek's inference benchmarks, produced a sentiment reset of consequence. The price has since roughly doubled from that floor. What the current multiple, approximately 33x trailing earnings, appears to credit: continued hyperscaler spending, CUDA ecosystem stickiness, and a reliable architecture cadence. What it does not appear to have fully absorbed: the revenue-sharing arrangement on China sales, where 25% of H200 revenues now go to the U.S. government under the current structure, and the earliest stages of credible in-house silicon development at the hyperscalers themselves.
What I'm watching. Three things draw most of my attention here. First, the China revenue structure. Jensen Huang announced at GTC 2026 in March that $NVDA has received purchase orders for H200 processors from Chinese customers and is restarting that manufacturing line. The economics of this channel have changed materially from the pre-2025 baseline: the government takes 25% of revenues from H200 sales, up from 15% on H20 sales under the initial arrangement. Management has not broken out China contribution explicitly in recent filings. The H20 episode in April 2025 produced an anticipated charge of $5.5 billion before the policy reversed three months later. The market appears to have treated the current arrangement as a settled matter, and that comfort may deserve closer examination.
Second, the Vera Rubin delivery and hyperscaler reception. The platform ships in H2 2026, and the gap between shipped and deployed at revenue-generating scale historically runs six to twelve months in this industry. The Q3 FY2027 earnings call, due roughly November 2026, will be the first substantive evidence of whether Rubin pull-through is consistent with the Blackwell ramp or more complicated. The customers who matter most here are $MSFT, $META, $GOOGL, and $AMZN, each of which is both a major $NVDA customer and an increasingly serious developer of in-house inference silicon.
Third, the competitive displacement pace. $AMD has the MI350 series in the 2026 market, and Meta has disclosed an arrangement for MI450-based deployment that could represent up to six gigawatts of capacity. $NVDA held approximately 86% of the AI GPU market in 2025. The question is not whether that share erodes; it is how quickly, and whether margin follows or leads.
The counter. The bear case rests on three pressures arriving together. The China revenue-sharing model is a structural margin tax of uncertain duration, open to restriction and relaxation at the discretion of an administration that has already moved in both directions within a single calendar year. The hyperscalers, who account for the majority of $NVDA's data center revenue, are also the parties best positioned to reduce their dependence on it over time: Google's TPU, Amazon's Trainium, and Microsoft's Maia programs represent serious, multi-year efforts to recapture margin from the AI infrastructure layer. And the annual architecture refresh cadence, Blackwell to Rubin to Rubin Ultra to Feynman across roughly four years, means enterprise customers are perpetually financing upgrades they may eventually resist or defer. I am not moving toward this view because CUDA's developer ecosystem, now at over five million contributors, has no credible replacement timeline in the near term; hyperscaler capital expenditure is running above $200 billion annually with no disclosed deceleration; and gross margins at 75% show no evidence of pricing pressure from any direction.
Where this leaves us. This piece carries no position. At $213.95 and $5.2 trillion in market capitalization, the name carries embedded expectations that do not fit the club's framework for considered entry at this moment. Those already in the name face a different set of questions, centered on the Rubin delivery timeline and the China revenue structure as Q3 and Q4 of this year unfold. For those outside, the work here is to understand what is already in the price before forming a view on what happens next. Each of the operating assumptions that sustains this multiple may prove correct. They are assumptions rather than certainties, and that distinction is the work. The piece is the position.