AI compute keystone — GPUs + CUDA + networking; the ~90%-share accelerator the whole AI economy depends on
AI exposure 88% — derived from reported segment revenue, not asserted.
NVIDIA: The Demand Debate Is Over. The Credit Debate Has Begun — a single-company deep dive with its full slide deck, published on Frontier Macro & Markets, 2026-10-03.
Silicon Layer → Semiconductor Devices · figures rebased to Q2 FY2027, reported 2026-08-26
Compute — the Blackwell Ultra and Rubin accelerators — is about 77.5% of revenue and by far NVIDIA's largest business, so the primary sits in Semiconductor Devices. The rest is tagged separately: Network Switching & Hardware at 15.0%, which is the NVLink compute fabric plus InfiniBand and Spectrum-X Ethernet and would be a large networking company on its own; Robotics & Autonomous Systems at 0.8% for DRIVE and Isaac. Rebased 2026-09 to NVIDIA's last disclosed compute/networking ratio, 83.8/16.2 from the FY2026 10-K, after it retired that cut in Q1 FY2027 — the Atlas had been carrying networking at 10.6%.
Also tagged: emerging in Connectivity → Network Switching & Hardware; emerging in Frontier & Emerging Bets → Robotics & Autonomous Systems.
NVIDIA reports by market platform, and changed that presentation in Q1 FY2027. Q2 FY2027 (quarter ended July 26, 2026, $96,221M total): Data Center $89,023M = 92.5%, up 117% year on year, of which Hyperscale is $48,710M (50.6% of the company) and AI Clouds, Industrial & Enterprise $40,313M (41.9%); Edge Computing $7,198M = 7.5%. The Atlas splits Data Center by PRODUCT rather than by customer, because that is what the placement turns on, and NVIDIA no longer discloses that cut — so compute and networking are held at the last reported ratio, 83.8/16.2 from the FY2026 10-K ($162,361M and…
Moat. CUDA software ecosystem + full-stack (GPU+networking+systems) lock-in; ~90% AI-accelerator share. Switching costs are the moat, not just the silicon.
Bottleneck / pricing power. Pricing power is intact at extraordinary scale: operating margin was 69.2% in Q2 FY2027 on revenue that more than doubled year over year. The constraint on shipments is advanced packaging and HBM supply, not demand.
Role in the AI stack. The compute layer — the keystone supplier every hyperscaler depends on.
All 176 companies · What moved this quarter · How the map is built