NVIDIA Corporation

▲ Bull Case

• ◆Inference-driven TAM expansion sustains 20%+ revenue CAGR through FY2029. As AI deployment shifts from training to inference, the addressable market expands from ~$150B to $300B+. NVIDIA's full-stack advantage (hardware + CUDA + TensorRT optimization) allows it to defend 65–70%+ blended share across both workloads, driving FY2028E revenue to $300B+ and sustaining 55%+ FCF margins. → Target: $175–$220/share.
• ◆Blackwell and next-generation Rubin architectures deepen the moat. NVIDIA's annual cadence (new architecture every 1–2 years) continuously resets the performance frontier, forcing competitors into perpetual catch-up. CUDA 13+ and NVLink 6.0 create system-level lock-in that custom ASICs cannot replicate for multi-model, multi-tenant workloads. Gross margins stabilize at 74–76% as software attach rates grow.
• ◆Sovereign AI and enterprise adoption open a second growth wave. Government-funded AI infrastructure programs ($50B+ globally announced) and enterprise AI deployment (NVIDIA AI Enterprise at $4,500/GPU/year) create revenue diversification beyond hyperscaler capex, reducing customer concentration and smoothing cyclicality.

▼ Bear Case

• ◆Hyperscaler capex cycle peaks in CY2026 and contracts 20–30% through CY2028. Historical precedent shows cloud capex cycles last 2–3 years; the current cycle began in mid-CY2023, placing peak spend in CY2025–CY2026. A capex pause — even without demand destruction — would create a revenue air pocket of $30–50B. Top 5 customers controlling 50%+ of revenue amplifies this concentration risk. → Target: $80–$100/share.
• ◆Custom ASICs erode inference market share to below 50% by CY2028. Google (TPUv6), Amazon (Trainium3), Microsoft (Maia 2), and Meta (MTIA v2) are all investing $1B+ annually in custom silicon optimized for their specific inference workloads, where custom chips offer 2–4x better price/performance for targeted models. As inference grows to exceed training spend (~$90–120B by CY2028), NVIDIA's blended market share compresses from ~80% toward 55–60%.
• ◆Export controls remove $20–30B+ in revenue with no replacement market. Progressive U.S. restrictions have already eliminated NVIDIA's ability to sell cutting-edge GPUs to China. Further tightening to include Middle East, Southeast Asia, or lower-performance tiers would materially impair the top line. This risk is directionally one-way — no plausible scenario involves regulatory relaxation.

“Is NVIDIA a durable platform company (Microsoft/Google analog) or a cyclical hardware supplier at peak earnings (Intel 2000/Cisco 2000 analog)? The bull camp argues that CUDA creates a software-defined competitive moat with recurring characteristics — switching costs, ecosystem effects, and compounding developer adoption — that will sustain supranormal returns through multiple hardware cycles, pointing to ROIC of 118%, expanding software attach, and the structural shift of $1T+ in data center infrastructure toward accelerated computing. The bear camp argues that NVIDIA's revenue is transactional (GPU shipments, not subscriptions), its customer base is the most concentrated in the S&P 500 (top 5 = 50%+ revenue), and its gross margins of 73–75% are unsustainable once competition arrives at scale. The resolution depends on one empirical question: whether NVIDIA's software ecosystem revenue (NVAIE, DGX Cloud, CUDA enterprise licensing) reaches $10–15B+ by FY2028, confirming the platform thesis — or whether software remains a rounding error, confirming the hardware-cyclical classification. At current levels, the market is pricing in 'platform with some cyclicality' — a consensus that is correct on expected value but underweights the bimodality of possible outcomes.”

• ◆FY2026-Q4 earnings + FY2027 guidance (Late February 2026) — first full-quarter Blackwell revenue; sets the growth deceleration narrative
• ◆Hyperscaler CY2026 capex guidance from MSFT, AMZN, GOOG, META earnings calls (January–April 2026) — any sign of pause immediately reprices NVDA's revenue trajectory
• ◆U.S. Bureau of Industry & Security export control rulemaking (Ongoing; next major revision expected H2 2026) — each incremental restriction removes $5–15B in addressable revenue
• ◆Blackwell gross margin trajectory visible in FY2027-Q1 and Q2 filings (May–August 2026) — confirms or denies margin sustainability thesis
• ◆Rubin architecture announcement / GTC 2027 (March 2027) — validates NVIDIA's annual architecture cadence and next-gen competitiveness

• ◆Hyperscaler capex deceleration/pause — Critical severity; 25–30% probability over 18 months; removes $30–50B in revenue within 2–3 quarters
• ◆Export control escalation (China + broader restrictions) — Critical severity; 40–50% probability of incremental tightening; eliminates $15–30B in addressable revenue
• ◆TSMC/Taiwan supply disruption (geopolitical or natural disaster) — Critical severity; 5–10% annual probability; NVIDIA cannot ship product without TSMC for 18–24+ months
• ◆Custom ASIC competitive displacement in inference — High severity; 50–60% probability of gradual share loss; compresses blended margins and share over 3–5 years
• ◆Key-person risk (Jensen Huang departure) — High severity; <5% annual probability; no identified successor; stock would de-rate 15–25% immediately