Physical AI Is Less Than 3% of Nvidia's Revenue. Here's How It Could Transform Nvidia By 2035.

Nvidia reported record fiscal 2026 metrics — $215.9 billion in revenue, $120.1 billion in net income, and $96.6 billion in free cash flow — with data centers accounting for 89.7% of revenue. The company unveiled its Rubin architecture (six new chips plus integrated systems such as the NVL72) aimed at agentic AI, while noting physical AI has already contributed north of $6 billion in fiscal 2026; management expects robotaxi and robotics scale to drive orders of magnitude more compute demand over the next decade. The piece argues Nvidia retains a long growth runway from generative and agentic AI within data centers and meaningful optionality if physical AI adoption expands, while highlighting current revenue concentration in data centers.

Analysis

Market structure: Nvidia (NVDA) is the primary beneficiary — data centers were 89.7% of FY26 revenue and Rubin gives Nvidia incremental vertical capture (chips+CPU+network) that increases ASPs and gross-margin leverage. Beneficiaries include hyperscalers (AMZN, GOOGL) for higher-performance instances, server/network OEMs that integrate Rubin stacks, and industrial OEMs (CAT, TSLA) that buy physical-AI compute; losers are legacy x86 incumbents (INTC) and undifferentiated GPU challengers as pricing power concentrates. Supply/demand looks tight for high-end HBM-backed accelerators: expect multi-quarter lead times unless foundry/memory capacity expands, implying sustained elevated pricing and input-cost pressure on suppliers.

Risk assessment: Tail risks include US export restrictions to China, a major wafer-capacity shock at TSMC, or rapid software commoditization that reduces GPU differentiation — any could cut revenue growth >30% QoQ in stress scenarios. Immediate (days) risk is IV and sentiment repricing after earnings; medium (3–12 months) depends on Rubin shipments/design wins; long term (3–5 years) hinges on robotaxi/robotics scale (management cites physical-AI TAM in the “hundreds of billions”). Hidden dependencies: hyperscaler capex cycles, software stack lock-in, and power/infrastructure constraints in data centers. Catalysts: announced design wins, ship volumes, GTC demos, and robotaxi deployment metrics.

Trade implications: For directional exposure, favor defined-risk bullish structures: establish a 2–4% portfolio notional long via 12–24 month NVDA LEAP call spreads (target +30–50% in 12 months, max loss = premium). Pair trade: long NVDA call spread vs short INTC (buy 6–12m NVDA calls, buy 3–6m INTC puts equal notional) to express GPU vs CPU divergence. If already long stock, sell 30–60 day 10–15% OTM covered calls to monetize near-term IV until Rubin order-book clarity.

Contrarian angles: Consensus underweights concentration risk — data centers ~90% revenue means a single hyperscaler slowdown could cut growth materially; the market may be underpricing the chance of margin compression if competitors emulate Rubin or memory prices fall. Historical parallels: CPU-to-GPU shifts looked permanent but were punctuated by cyclical capex and alternative architectures (e.g., FPGA, custom ASIC) emerging unexpectedly. Unintended consequence: vertical integration (Rubin stacks) could accelerate customer switching costs but also invite antitrust scrutiny or encourage hyperscalers to vertically insource over 3–5 years.

Key Decisions for Investors

• Establish a 2–4% portfolio notional long in NVDA via 12–24 month LEAP call spreads (buy-debit) targeting +30–50% in 12 months; size to risk no more than 1% of NAV loss, and cut if NVDA falls 20% from entry or if Rubin confirmed orders miss management guidance by >25%.
• Implement a relative-value pair: long NVDA (6–12m call spread) and buy 3–6m INTC puts equal notional (1–2% portfolio risk) to profit from GPU/accelerator share gains vs legacy CPU exposure; unwind if INTC shows sustained design-win momentum or NVDA gross margins contract >200bps QoQ.
• If long NVDA equity, sell 30–60 day covered calls 10–15% OTM to harvest IV premium while awaiting Rubin shipment/design-win data; roll or re-evaluate after each quarterly earnings/GTC update.
• Initiate small sector reallocation: add 1–2% each to AMZN and GOOGL (cloud demand plays) and 0.5–1% to CAT (industrial robotics exposure); trim if cloud capex guidance falls >20% QoQ or if robotaxi deployments miss projected milestones by >50% over 12 months.