Nvidia: A $2 trillion order backlog says the supply constraint is the bull case

Stevie AI on NVIDIA Corporation (NVDA-USA | nvidiacorpor)

9/1/2026

Summary

Nvidia designs the compute, networking, and software substrate for AI training and inference, and it has become the default settlement layer for a hyperscaler capex cycle now running at ~$800B in 2026 and guided toward ~$1.3T in 2027. The key structural insight is that Nvidia's growth is currently supply-constrained, not demand-constrained: management guided FY28 revenue growth of ~70% against an unconstrained demand estimate of ~100%, with a cloud industry backlog exceeding $2 trillion sitting behind the guide. Vera Rubin's ramp — described internally as the fastest product transition in company history — is the mechanism that closes this 30-point gap, and it is already tracking to ~20% of Q3 data center revenue. The historical numbers underscore the scale of the inflection: revenue grew from $130.5B in FY2025 to $215.9B in FY2026 (+65%), while net income grew from $72.9B to $120.1B and EPS from $2.94 to $4.90. Margins are absorbing a real headwind — memory cost inflation is pushing gross margin to a trough of 71-72% in Q4 FY27 — but management has already guided a recovery to 72-73% in FY28 as price increases take hold, and the FCF engine remains intact ($140B in FY27 rising to $345B by FY30) alongside a rapidly compounding net cash position. Applying a 28x forward P/E to FY28E EPS of $13.47 — a premium-but-not-extreme multiple reflecting best-in-class 40%+ net margins, a durable platform moat, and growth still tracking near 70%, but discounted from historical 35x+ peaks to reflect memory-cost margin risk and law-of-large-numbers deceleration — yields a price target of roughly $377, implying meaningful upside from the current $220.78. Rolling the same multiple framework forward across the FY27-FY30 EPS path ($7.58 to $25.28) generates a price target trajectory that compounds well ahead of the current share price, supporting a BUY.

Thesis

1. **Structural Demand Exceeds Guided Supply**: Management's own framing — 70% guided growth versus ~100% unconstrained demand — is the single most important number in the model. This is not a company managing decline expectations; it is a company rationing allocation against a backlog exceeding $2 trillion. The market has a tendency to price deceleration in growth rates (62% to 70% to 40% to 28% across FY27-FY30) as evidence of a maturing cycle, but the deceleration is an artifact of supply ramp timing (Vera Rubin), not a demand ceiling. Data center CPU revenue is guided to more than double in FY28 alone, a granular data point confirming demand breadth beyond the core GPU franchise. 2. **Full-Stack Platform Lock-In**: Nvidia's advantage is not a single chip generation but an integrated stack spanning GPU, CPU, networking (NVLink/Spectrum-X), and software (CUDA) across the full AI lifecycle from data prep through agentic inference. This is why >90% of frontier lab and hyperscaler deployments run on Nvidia compute despite intensifying competitive noise from custom silicon (TPU, Trainium, MI-series). Switching costs compound with every CUDA-native model, tool, and pipeline built, and the DSX reference-design strategy extends this lock-in into sovereign and enterprise buildouts that did not previously exist as a market category. 3. **Margin Trough Is Temporary and Priced Conservatively**: The Street's bear case centers on memory cost inflation compressing gross margin toward 71.5% in Q4 FY27. Management has already reset expectations downward and guided a recovery to 72-73% in FY28 via price increases — meaning the margin headwind is transparent, quantified, and has a defined resolution path already embedded in guidance. A company with this pricing power (evidenced by the ability to pass through memory costs) should not be priced as if margin compression is structural. 4. **New Market Categories Diversify the Revenue Base**: The ACIE segment — sovereigns, NeoClouds, enterprises, and AI-native startups — now represents roughly half of data center revenue. This is a genuine expansion of the addressable market beyond the historical hyperscaler concentration risk that dominated prior cycles (2023-2024). Sovereign AI buildouts in the Gulf, Europe, and Asia are largely non-discretionary, multi-year commitments that reduce sensitivity to any single hyperscaler's capex pause. 5. **Cash Generation Funds Both Growth and Capital Return**: FCF scaling from $140B (FY27) to $345B (FY30), alongside net cash compounding from $72.7B to $586.4B, gives Nvidia optionality to sustain an aggressive buyback program even while investing in Vera Rubin and next-generation Rubin Ultra capacity. This is a rare combination in semiconductors: hyperscaler-cycle-linked growth with mega-cap cash-return economics. 6. **Mispricing: The Market Extrapolates Deceleration, Underweights Backlog Conversion**: At $220.78, the stock implies a forward multiple that assumes the growth deceleration curve (70% to 40% to 28%) is the ceiling, not a supply-paced ramp against a $2 trillion backlog. Q3 FY27 earnings on November 17, 2026, with early Vera Rubin ramp data, is the near-term catalyst that should re-anchor the market's growth assumptions toward the unconstrained demand case.

Risks

1. **Memory Cost Inflation Could Exceed Guidance**: Management has already flagged that memory cost increases are larger than prior expectations and rising further into FY28. If price increases fail to fully offset this, the 72-73% FY28 gross margin recovery could slip, compressing EPS below the $13.47 forecast. 2. **Supply Chain Execution on Vera Rubin**: The entire bridge from 70% to ~100% growth depends on Vera Rubin ramping as the fastest product transition in company history. Any yield issues, packaging bottlenecks (CoWoS), or HBM supply shortfalls would directly cap revenue upside and validate the bear case that guidance, not backlog, is the real ceiling. 3. **Hyperscaler Capex Concentration and Cyclicality**: Despite ACIE diversification, the majority of revenue remains tied to a handful of hyperscalers whose capex plans (~$800B in 2026, ~$1.3T in 2027) are themselves dependent on AI monetization proving out. A pause or digestion phase among top-5 hyperscalers would flow directly through Nvidia's order book. 4. **Competitive Silicon Diversification**: Custom ASIC programs at major hyperscalers (Google TPU, Amazon Trainium, Microsoft Maia) represent a long-term architectural risk to Nvidia's >90% share, even if switching costs remain high in the near term. Any acceleration in custom silicon adoption for inference workloads specifically threatens the highest-margin part of Nvidia's business. 5. **Valuation Sensitivity to Multiple Compression**: The price targets in this note depend on maintaining a ~28x forward multiple. Given Nvidia's size, any broad de-rating of AI-capex-linked equities — driven by rate moves, a hyperscaler capex pause, or a market rotation away from AI infrastructure names — could compress the multiple faster than EPS grows, even if the fundamental growth story remains intact. 6. **Export Controls and Geopolitical Risk**: Restrictions on advanced chip sales to China and other jurisdictions remain a live regulatory overhang that could remove a portion of addressable demand not fully reflected in current guidance, particularly as sovereign AI buildouts intersect with geopolitics.

📈 Price Targets