Perplexity to use Nvidia’s new CPUs for AI agent tasks

Perplexity said it will use Nvidia’s new Vera CPU, citing ~1.5x faster performance versus traditional CPUs for AI-agent coding tasks. Nvidia projects Vera CPU sales of $20B by the end of its fiscal year, as AI firms shift toward more generic CPU designs alongside AI-specific chips. Despite the broader “stocks fall” backdrop and oil rising after ship attacks, this development is a positive read-through for Nvidia’s CPU demand, likely supporting modest single-stock movement.

Analysis

The strategic read-through is not "Nvidia sold a CPU" but that it is trying to widen the attach rate of its platform from accelerator-only spend into the control plane of AI infrastructure. That is a cleaner moat than GPU demand alone because once an enterprise standardizes on a vendor-managed stack for agentic workloads, the switching cost becomes architectural, not just performance-based. For AMD and especially INTC, the risk is less immediate unit displacement than margin pressure as buyers justify a premium for a vertically integrated AI server bill of materials.

Near term, the market may overreact to the competitive headline before there is any evidence of meaningful socket share or revenue mix change. The real catalyst path is 1-3 months: commentary from hyperscaler capex plans, OEM design wins, and whether Nvidia starts bundling CPUs into broader AI platform commitments that compress competitors' pricing. The structural read-through is 6-18 months: if AI agents truly keep servers hot continuously, CPU utilization rises and the value shifts toward vendors that can sell the full stack, which should be favorable for NVDA and neutral-to-negative for legacy x86 incumbents.

The contrarian point is that this is still a proof-of-concept customer count, not a demand curve. If NVDA's CPU revenues do not show up as meaningful incremental growth on the next two earnings calls, the market will likely fade the narrative and rotate back to GPU scarcity and export-control issues. For ORCL, the benefit is indirect: being an early customer can help cloud differentiation, but the financial impact is likely too small to justify a standalone long unless it coincides with accelerating OCI bookings.

Key Decisions for Investors

• Long NVDA / short AMD as a 1-3 month relative-value pair: thesis is platform expansion versus narrative compression in x86; stop if AMD server share commentary or guide-outs stabilize and NVDA CPU traction looks non-exclusive.
• Tactically fade any knee-jerk rally in INTC on the article with a short-lived short or put spread into the next earnings cycle; the setup only works if investors start pricing a meaningful AI-control-plane share loss, which is not yet proven.
• Add a small tactical long NVDA on pullbacks rather than strength; the best risk/reward is if the market waits for confirmation while management continues to seed design wins. Falsify if CPU revenue commentary remains immaterial after two reporting periods.
• Use ORCL as a secondary beneficiary only if OCI growth and AI capex commentary accelerate together; otherwise keep it on the watchlist, not in the basket, because the article alone does not justify a standalone fundamental re-rate.