NVIDIA Corp.
Rating
Accumulate
Adding on Dips — Active Accumulation
Combined average of Moat (AI Resilience), Growth, and Valuation scores.
Moat Score
CUDA software ecosystem and 10-year hardware lead in AI compute.
Nvidia's moat isn't just "fast chips", it's the Full-Stack Software Advantage:
- CUDA Software Ecosystem: With over 4 million developers, CUDA is the industry standard. Moving to another hardware provider requires rewriting massive amounts of code.
- Innovation Velocity: Nvidia has moved to a 1-year product cycle (Hopper → Blackwell → Rubin), staying ahead of competitors who are still catching up to the last generation.
- Infiniband Networking: Their integration of networking (Mellanox) allows them to sell high-margin full-racks, not just individual GPUs.
Ten Moats Verdict
NVIDIA's moat is almost entirely AI-resilient. The CUDA network effect and proprietary compute standard deepen as AI infrastructure spending grows — making NVIDIA the infrastructure of the AI economy.
CUDA is the canonical learned-interfaces moat in semiconductors — 15+ years of developer mindshare, every ML PhD candidate learns CUDA, every major ML framework (PyTorch, TensorFlow, JAX) is CUDA-first by default. Switching to ROCm or any alternative is a multi-year rewrite. AI demand strengthens this moat rather than commoditising it: more AI workloads to write means more CUDA-coupled code, wider switching costs. Routed to resilient via aiExposure override — the AI wave is protecting this interface, not threatening it.
Customer ML training and inference pipelines are deeply embedded against CUDA-specific business logic — Megatron, DeepSpeed, vLLM, NCCL, cuDNN, cuBLAS are not portable abstractions. Every AI lab's production training stack is CUDA-coupled workflow at the code level. AI demand strengthens this lock-in by adding more CUDA-specific framework code to every codebase; routed to resilient via aiExposure override.
Not applicable — NVIDIA does not derive moat from public data access.
GPU chip architects, CUDA kernel engineers, and AI systems researchers remain extraordinarily scarce and cannot be AI-replaced.
CUDA + hardware + Mellanox networking + NIM microservices = a full-stack AI infrastructure solution competitors cannot match.
Millions of CUDA training runs generate proprietary AI workload optimization insights unavailable to competitors.
Export control whipsaw persists: H20 ban (April 2025) caused a $4.5B charge, reversed July 2025; H200 approved December 2025; 400K+ units cleared for China in April 2026. Policy risk remains structurally elevated despite the near-term reopening.
4M+ CUDA developers create the largest and most entrenched AI developer community — switching has a multi-year rewrite cost.
Every major AI training and inference workload is embedded in NVIDIA infrastructure at the infrastructure layer.
CUDA is the de facto standard platform for AI compute — the PyTorch/TF ecosystem is CUDA-first by default.
Combined average of Moat (AI Resilience), Growth, and Valuation scores.
Moat Score
CUDA software ecosystem and 10-year hardware lead in AI compute.
Growth Score
Q1 FY2027 (reported May 20 2026) blew past the Street: revenue $81.6B (+85% YoY) vs. $78.8B consensus, EPS $1.87 vs. $1.76. Data center hit $75B (+92% YoY, +21% QoQ) with networking a record $14.8B (+199% YoY). Critically, management guided Q2 FY27 to $91B — well above the ~$86B Street had modelled and a clean rejection of the pre-print 'in-line/miss' bear scenario. The board added $80B to buyback authorization and raised the dividend 25×. China H200 approvals (400K+ units cleared April 2026) and Vera Rubin volume production (H2 2026, on track) remain the key FY2027 swing factors. FY2027 EPS consensus has been revised up toward ~$9.00.
Valuation Score
At ~$250 following the post-print rally on the Q1 FY27 beat and $91B Q2 guide, NVDA sits roughly at the revised $260 base case. Forward P/E is ~28× on upgraded NTM EPS of ~$9.00 (PEG ~0.4× against ~75% FY27 EPS growth) — still inexpensive on growth-adjusted multiples, but the easy re-rating from the print has been realised. The binary May 20 catalyst is now behind the stock, leaving the next checkpoint the Q2 FY27 print and any hyperscaler capex revision. Fairly valued vs. the 12-month base case, with the bear ($130) far below.
The Ecosystem Moat (CUDA)
Nvidia's moat isn't just "fast chips", it's the Full-Stack Software Advantage:
- CUDA Software Ecosystem: With over 4 million developers, CUDA is the industry standard. Moving to another hardware provider requires rewriting massive amounts of code.
- Innovation Velocity: Nvidia has moved to a 1-year product cycle (Hopper → Blackwell → Rubin), staying ahead of competitors who are still catching up to the last generation.
- Infiniband Networking: Their integration of networking (Mellanox) allows them to sell high-margin full-racks, not just individual GPUs.
Ten Moats Verdict
NVIDIA's moat is almost entirely AI-resilient. The CUDA network effect and proprietary compute standard deepen as AI infrastructure spending grows — making NVIDIA the infrastructure of the AI economy.
CUDA is the canonical learned-interfaces moat in semiconductors — 15+ years of developer mindshare, every ML PhD candidate learns CUDA, every major ML framework (PyTorch, TensorFlow, JAX) is CUDA-first by default. Switching to ROCm or any alternative is a multi-year rewrite. AI demand strengthens this moat rather than commoditising it: more AI workloads to write means more CUDA-coupled code, wider switching costs. Routed to resilient via aiExposure override — the AI wave is protecting this interface, not threatening it.
Customer ML training and inference pipelines are deeply embedded against CUDA-specific business logic — Megatron, DeepSpeed, vLLM, NCCL, cuDNN, cuBLAS are not portable abstractions. Every AI lab's production training stack is CUDA-coupled workflow at the code level. AI demand strengthens this lock-in by adding more CUDA-specific framework code to every codebase; routed to resilient via aiExposure override.
Not applicable — NVIDIA does not derive moat from public data access.
GPU chip architects, CUDA kernel engineers, and AI systems researchers remain extraordinarily scarce and cannot be AI-replaced.
CUDA + hardware + Mellanox networking + NIM microservices = a full-stack AI infrastructure solution competitors cannot match.
Millions of CUDA training runs generate proprietary AI workload optimization insights unavailable to competitors.
Export control whipsaw persists: H20 ban (April 2025) caused a $4.5B charge, reversed July 2025; H200 approved December 2025; 400K+ units cleared for China in April 2026. Policy risk remains structurally elevated despite the near-term reopening.
4M+ CUDA developers create the largest and most entrenched AI developer community — switching has a multi-year rewrite cost.
Every major AI training and inference workload is embedded in NVIDIA infrastructure at the infrastructure layer.
CUDA is the de facto standard platform for AI compute — the PyTorch/TF ecosystem is CUDA-first by default.
Growth Analysis
Growth Drivers
Key Risk
The near-term 'bubble crystallises' test resolved favourably: Q1 FY27 beat and the Q2 guide ($91B) came in well above the $86B Street line, so the in-line/miss scenario did not trigger. The structural concentration risk persists, however — the ~$2.1T cloud backlog remains roughly half-anchored to OpenAI + Anthropic (both deeply FCF-negative), Project Stargate ($500B) layers in financing-coupled commitments, and SOX still trades richly vs. its 200-day MA. Revised falsifiable test: if any top-3 hyperscaler cuts FY2027 AI capex guidance by >10% on a single earnings print between now and Jan 2027, the demand-durability thesis weakens and multiple compression dominates the EPS ramp.
Score Derivation
Base 90 (30%+ CAGR on $68B quarterly base) + 5 TAM expansion (sovereign AI, Rubin architecture) + 5 platform stickiness (CUDA ecosystem NRR equivalent) − 5 AI-capex demand concentration (OpenAI/Anthropic anchor half of $2.1T cloud backlog; both FCF-negative) = 95
Price Scenarios (12–24 Months)
Valuation Multiples
| Trailing P/E (GAAP) | ~42× |
| Forward P/E (NTM) | ~28× |
| PEG Ratio | ~0.4× |
| Price / Sales (NTM) | ~15× |
| Price / FCF | ~55× |
At 27× forward P/E with a PEG of ~0.6×, NVDA still screens cheap on growth-adjusted multiples — but the absolute price/sales (~15×) and price/FCF (~55×) sit well above sector medians, leaving the stock more dependent on the Q1 FY27 print and Q2 guide to validate the FY27 ramp. The SOX trades ~62% above its 200-day MA — higher than the Nasdaq's 55% deviation at the March 2000 peak — meaning NVDA's PEG support depends on EPS growth actually compounding, not on the multiple absorbing any miss.
Approximate figures as of May 2026.
Where We Are vs Targets
Loading live price…
Export controls re-escalate targeting Blackwell/Rubin-class chips; hyperscaler in-house ASICs capture 20%+ of AI training workloads; an OpenAI/Anthropic funding stumble forces backlog write-downs and CUDA lock-in erodes faster than expected.
- U.S. imposes new export restrictions on Blackwell/Rubin-class chips to allied nations, removing $15B+ in annual revenue
- Google TPU v6 and Amazon Trainium3 capture 20%+ of hyperscaler AI training by end of 2026, pressuring NVIDIA market share below 75%
- Hardware-agnostic tooling (OpenAI Triton, JAX) achieves broad adoption, weakening CUDA switching costs and forcing ASP compression
FY2027 tracks toward ~$390B+ after the Q1 beat and $91B Q2 guide; Vera Rubin ramps into H2 2026 on schedule; China H200 contributes $15-20B incremental revenue; NVIDIA Enterprise software reaches $5B+ ARR.
- Q2 FY2027 revenue lands near the $91B guide, confirming data center demand durability through the Blackwell-to-Rubin transition
- China H200 shipments (400K+ units cleared in April 2026) contribute $15-20B incremental FY2027 revenue, lifting the full-year total toward $390B+
- Vera Rubin NVL72 volume production commences H2 2026 at major hyperscalers, extending the $1T order book into FY2028
Vera Rubin cycle exceeds $1T order estimate; sovereign AI buildout accelerates to $150B+ globally; China becomes 15%+ of revenue; software inflects above $10B ARR.
- Sovereign AI infrastructure spending accelerates to $150B+ as 50+ nations deploy domestic GPU capacity, adding a recurring government revenue layer
- Vera Rubin yields exceed roadmap targets, enabling 3× performance-per-dollar vs. Blackwell and driving ASP expansion to $75K+ per rack unit
- NIM microservices and NVIDIA AI Enterprise scale to $10B+ ARR, re-rating the stock toward software multiples on a higher-margin revenue mix