NVDA Blackwell→Rubin product cycle sustains 60%+ data center revenue growth through 2027 ($194B FY2026 DC revenue, $75B+ quarterly run-rate), while the CUDA software ecosystem — 18 years of developer lock-in across PyTorch/TensorFlow/JAX — prevents any competitor from capturing material training workload share even as custom ASICs nibble at inference.
Research snapshot · 6/22/26
NVDANVIDIA Corporation
GAA and backside power delivery | Digital CMOS LiDAR | Machine vision and LiDAR-perception
Open NVDA in Robinhood ↗Conviction●●●●●5 of 5
Research target294.97Snapshot target
Thesis statusNEEDS_MORE_DATALast reviewed 6/22/26
Market cap4.82TSnapshot value
(1) Rubin (Vera Rubin) architecture production ramp H2 2026 — Jensen Huang cited ~$1T in demand through CY2027. (2) Hyperscaler capex continues at $700B+ aggregate — Microsoft/Google/Amazon/Meta all committed to NVIDIA-based infrastructure expansion. (3) Next earnings (Q2 FY27 expected ~Aug 2026) likely to show data center revenue >$75B quarterly with Blackwell at scale + early Rubin contributions.
(1) Hyperscalers meaningfully shift >30% of training capex to custom silicon (Google TPU v6, Amazon Trainium3, Microsoft Maia) — reduces NVIDIA share of the training TAM. (2) Rubin faces HBM4 supply shortage or power/thermal issues that delay the ramp by 6+ months. (3) AI model architectures shift away from dense transformer training toward sparse/MoE architectures that reduce GPU demand intensity per training run.
Overwhelmingly bullish consensus — "NVIDIA is the only game in town." CUDA moat universally acknowledged as unbreachable. Some concern about valuation ceiling and whether $200B quarterly DC revenue is already priced in. Minimal bearish voices — the most common critique is "priced for perfection" rather than "thesis is broken." Retail + institutional both long.
Snapshot · 6/22/26🟡 Mixed · ins-$267.4M · 13F 16+/9- · short↓0.16
Snapshot · 6/22/26