Perplexity says it plans to use Nvidia’s new CPU

Perplexity confirmed it plans to use Nvidia’s new “Vera” central processing units, citing AI agent coding tasks running about 1.5x faster than traditional CPUs. Nvidia expects Vera to drive $20B in sales by the end of its fiscal year as the company diversifies beyond AI-specific chips. Despite an overall market drop tied to AI trade concerns and oil rising after ship attacks, the incremental validation from Perplexity is a modest positive for Nvidia’s CPU roadmap.

Analysis

This is more important as a platform-expansion signal than as a near-term revenue event. If Nvidia can sell CPU sockets into AI-agent stacks, it raises the probability that the company captures a larger share of the rack budget, not just the accelerator budget; that is structurally negative for Intel and, to a lesser extent, AMD because it shifts enterprise/server procurement toward a bundled vendor with higher software and systems leverage. Oracle is a modest relative winner only if it can market a tighter NVDA-based cloud stack that improves utilization and differentiation, but the direct P&L impact is likely small versus the sentiment boost.

The market is likely to overreact to the concept and underreact to execution risk. The key question over the next 1-3 months is whether this becomes a meaningful design-win pipeline or just another proof-point with limited volume; absent disclosed unit demand, this remains a narrative trade. Falsifiers are simple: no follow-through in Nvidia datacenter commentary, or AMD/Intel showing stable server share and pricing discipline on their next prints.

Over 6-18 months, the bullish case is that AI agents increase always-on compute intensity, driving more CPU content per workload and making the control plane as strategic as the GPU. The contrarian risk is that hyperscalers and frontier-model labs keep pushing custom silicon, capping Nvidia’s attach-rate and preventing Vera from becoming a large standalone franchise. In that case, the move is overdone and the best short-term setup is mean reversion in the names most exposed to server-CPU share loss.

Key Decisions for Investors

• Long NVDA vs short INTC as a 1-3 month relative-value pair: asymmetric upside if Vera becomes a real platform attach, with INTC most exposed to share loss and multiple compression if the market extrapolates too far.
• Use AMD as the cleaner hedge against the AI-CPU narrative only if next channel checks show enterprise/server share pressure; otherwise keep the short modest because AMD’s mix is less levered than INTC’s.
• Add ORCL on weakness only as a tactical sentiment trade, not a fundamental catalyst trade; thesis is limited to higher AI-cloud stickiness if Nvidia-standard CPU stacks gain traction.
• If NVDA rallies >5% on the story without a volume or ASP disclosure, fade via short-dated call overwrites or a put spread; that would suggest the market has priced in more revenue than the data supports.