Evercore ISI reiterates Nvidia stock rating on AI demand outlook By Investing.com

Evercore reiterated an Outperform and $352 PT on Nvidia after CEO Jensen Huang’s GTC keynote, as firms highlight a revised data-center revenue outlook of $1 trillion for 2025–2027 (up from $0.5T prior). Nvidia reported $215.94B TTM revenue (+65% YoY) with a 71% gross profit margin and said it has shipped the first Vera Rubin platform and has $1T of cumulative purchase order visibility for Grace Blackwell and Vera Rubin through 2027. Multiple brokers (BofA, Goldman, Stifel, Keybanc, Wolfe) maintained Buy/Outperform stances, and Evercore projects Nvidia captures ~70% of value in the new Agentic-AI compute cycle.

Analysis

The current narrative prices Nvidia as the choke-point for the new AI compute stack; that creates concentrated second-order winners (server OEMs, power/cooling and high-density chassis suppliers) and losers (smaller accelerator vendors who can’t match scale). Expect supply-chain margin capture to bifurcate — firms that supply rack-level integration and logistics will see higher incremental gross margins than component suppliers unless component lead times re-price upward.

Networking choices (copper-first then hybrid/optical later) create a multi-year timing mismatch: copper-led ramps favor incumbents with mature copper PHY stacks in the next 12–36 months, while optical specialists should see a discrete step change in revenue and pricing power around the 2026 hybrid/optical transition. That calendar creates an investable corridor where OEMs and systems integrators can monetize now, with optical suppliers playing catch-up but likely benefiting from stronger multi-year capital commitments.

Concentration and policy are the two greatest tail-risks. A pause by a single hyperscaler, a macro ad-revenue shock, or renewed export controls could compress order visibility quickly and expose high multiple expectations; conversely, multi-year software-to-hardware migrations (on-prem inference, specialized agents) are structural upside. Watch quarterly guidebacks, PO cadence vs delivered units, and any public shift from copper to optical roadmaps as 6–18 month forward catalysts.

Consensus appears to underprice operational execution risk and inventory seasonality while over-indexing on perfect demand growth. That tilts the optimal approach toward convex, hedged exposure: capture upside to continuing AI capex but limit one-off downside should order visibility evaporate or geopolitical frictions force regional carve-outs.

Key Decisions for Investors

• Long NVDA with a hedged spread: allocate a tactical overweight via a 12–24 month call-debit spread (buy longer-dated call, sell a higher strike) sized to 3–5% portfolio risk. Rationale: captures multi-year AI upside while capping premium paid if cadence falters; target 2:1 upside/downside payoff to risk budget.
• Pair trade: long SMCI (6–12 month horizon) / short a generic hardware distributor or lower-margin OEM (equal notional, reduce beta). Rationale: server OEMs should capture rack-level pricing and services revenue; pair reduces market/systematic exposure and isolates systems integration premium.
• Event/hedge for hyperscaler-China risk: buy 3–12 month puts on BABA or use a low-cost put spread to protect EM hyperscaler exposure if export-control headlines re-intensify. Rationale: mitigates geopolitical downside that would materially dent near-term GPU demand from large China customers.
• Volatility play around guidance prints: sell near-dated NVDA calls funded by buying longer-dated protection (calendar call spread) into the next major guide/earnings cycle. Rationale: monetizes elevated short-term implied vol premia while retaining longer-term upside exposure in case execution continues.