11 Words From Nvidia CFO Colette Kress That Should Have AMD and Intel Investors Worried

Nvidia says its new Vera CPU opens a brand-new $200 billion TAM and could generate nearly $20 billion of CPU revenue this year, expanding its AI stack beyond GPUs. The platform is designed for agentic AI workloads and complements Blackwell and Rubin GPUs rather than cannibalizing them, potentially increasing Nvidia’s share of AI infrastructure spend. The article argues the market has not fully priced in this diversification, leaving room for valuation upside.

Analysis

Vera matters less as a standalone product than as a lever to widen NVDA’s attach rate across the entire AI data-center bill of materials. The second-order effect is that Nvidia can monetize workloads that are currently a handoff to x86 incumbents, which should improve wallet share even if unit GPU demand merely normalizes instead of re-accelerating. That also makes the business less cyclical: if the CPU becomes the control plane for agentic systems, customers are more likely to standardize on Nvidia’s stack and delay multi-vendor substitution.

The competitive read-through is negative for INTC and AMD, but not primarily because they lose share in general-purpose server CPUs; the bigger issue is that Nvidia is redefining the procurement decision around software-integrated architecture rather than isolated chip performance. In that framework, a CPU sold alongside GPUs, networking, and CUDA creates a bundle discount that rivals will struggle to match without sacrificing margin or ecosystem control. Suppliers tied to data-center buildouts should benefit from higher platform complexity, but the value capture should concentrate further inside Nvidia’s stack.

The market may be underestimating the timeline asymmetry. GPU demand can swing quarter to quarter, but CPU adoption in AI infrastructure is a multi-year conversion event; if management already has visibility to meaningful revenue, the first-order earnings impact is likely to be visible sooner than consensus expects. The bigger risk is that investors extrapolate the TAM before proving actual deployment economics, and that hyperscalers respond with custom silicon or internal orchestration layers that reduce the need for a bundled Nvidia CPU over 12-24 months.

Contrarian view: the opportunity is probably real, but the stock may already discount a lot of platform expansion through the multiple. The better debate is not whether Vera adds revenue, but whether it materially raises NVDA’s terminal margin and switching costs enough to justify multiple expansion from an already premium base. If agentic workloads stay niche, Vera becomes an incremental share grab; if they scale into production, it becomes the start of a much stickier infrastructure franchise.

Key Decisions for Investors

• Add to NVDA on any post-headline weakness over the next 1-3 weeks; the setup favors buying dips because the CPU story is an underappreciated free option on a larger platform multiple, not just a revenue line item.
• Underweight or short-basket INTC/AMD vs NVDA for a 3-6 month horizon; Vera increases the probability that AI server wins get decided at the platform level, where incumbents are weakest. Use a pair trade to reduce market beta.
• Buy NVDA call spreads 6-12 months out rather than outright calls; the catalyst is adoption visibility over several quarters, and spreads capture upside while limiting multiple-contraction risk if the market shrugs.
• Monitor hyperscaler capex commentary over the next 1-2 earnings cycles; if CPUs begin showing up in budget language alongside GPUs and networking, add to NVDA and reduce exposure to standalone CPU names.
• Avoid chasing the implied TAM into earnings unless management quantifies conversion rates; if Vera revenue is still a small fraction of the stated opportunity, the stock may revert to GPU-cycle trading despite the positive narrative.