An Nvidia Move Just Put AMD and Intel on the Defensive

Nvidia expects to sell $20 billion of Arm-based Vera server CPUs this year, with revenue projected to double next year, as deployments expand to every major hyperscaler, neocloud, AI lab, and system OEM. Vera is claimed to complete agentic tasks 1.8x faster and provide 5x more bandwidth per watt than competing data-center CPUs, positioning Nvidia to gain share from Intel and AMD as Arm-based servers reach 45% of data-center revenue. The article argues Nvidia's 29x P/E, versus AMD's 122x and Intel's 88x, leaves it comparatively attractive despite Nvidia shares rising only 21% in 2026 while its rivals posted triple-digit gains.

Analysis

The relevant mechanism is not CPU revenue alone but control of the node architecture: an Nvidia CPU can raise GPU attach rates, networking content, and software lock-in by making the full rack a validated Nvidia design. That shifts purchasing power away from merchant x86 vendors and toward platform vendors, with Intel most exposed because server CPUs remain central to its gross-profit recovery and foundry-utilization narrative. AMD is less cleanly impaired: EPYC can remain the host CPU in heterogeneous AI clusters, and AMD’s own accelerator roadmap gives hyperscalers a strategic reason to avoid a single-vendor Nvidia stack.

The stated adoption and revenue outlook should be treated as management guidance rather than evidence of displacement. The near-term constraint is software and fleet migration: enterprise inference deployments generally retain x86 compatibility, while hyperscalers can optimize Arm internally; therefore, 1-3 month equity impact will depend on disclosed design wins and whether Vera is sold as a standalone SKU versus bundled system content. ARM is a second-order beneficiary only if its royalty economics capture Nvidia’s volume growth; that depends on the specific licensing structure and is not inferable from unit adoption alone.

Consensus may overstate direct CPU share loss while underestimating Nvidia’s ability to monetize a higher-value rack. A CPU substitution thesis becomes materially investable if Intel or AMD guide datacenter CPU ASPs, unit growth, or gross margin below expectations while Nvidia reports rising CPU/networking contribution without GPU bundle discounting. Over 6-18 months, the larger risk is customer pushback: hyperscalers may expand custom Arm silicon and AMD deployments precisely to cap Nvidia’s system-level pricing power.

Key Decisions for Investors

• Prefer a 3-6 month long NVDA / short INTC pair rather than an outright NVDA long: it targets platform-content gains against Intel’s higher operating-leverage exposure to server CPU share and pricing. Size modestly until Nvidia discloses standalone CPU revenue/margins; exit the short if Intel guides Data Center AI revenue and gross margin materially above consensus.
• Do not short AMD solely on this development. Maintain AMD as a watch item for next earnings: a cut to EPYC growth, server CPU ASPs, or AI accelerator attach would validate a short; stable EPYC share or improving MI-series demand falsifies the displacement case.
• Use ARM only as an event-driven long watch, not a direct read-through: initiate after evidence that Vera volume carries royalty-bearing ARM IP economics. Risk is that existing architectural licenses leave incremental royalties de minimis despite strong deployment commentary.
• For NVDA longs, require confirmation that incremental CPU adoption expands total system gross profit rather than discounts GPU bundles. A decline in datacenter gross margin or evidence of hyperscaler custom-silicon substitution is the key 6-18 month thesis break.