Bank of America’s chip analyst tells Wall Street the “next Nvidia” is still Nvidia

Wall Street may be scanning for the next artificial-intelligence winner, but Bank of America Securities’ top chip analyst says the answer is the same as it has been for most of the current AI cycle. In coverage of a client note published Monday, Bank of America semiconductor analyst Vivek Arya reiterated a Buy rating on NVIDIA and maintained a $350 price target, describing NVDA as the firm’s “top sector pick.” The note also leaned on recent interactions with NVIDIA executives, including CFO Colette Kress and management commentary around the company’s GTC Taipei programming, according to the reporting.

Arya’s core message was that NVIDIA’s advantage is not static. Management, as relayed in the report, is trying to “attack more of the AI rack” with each product generation, keeping NVIDIA positioned as a supplier across compute, networking, and system-level components rather than only at the accelerator layer. The note characterized NVIDIA as a “king of diversity,” supplying major hyperscalers and model builders while also maintaining incumbency in enterprise and industrial workloads. It also pointed to a growing portion of AI infrastructure spending that is tied to what NVIDIA describes as rack-scale systems.

A key part of the argument is that Wall Street may be underestimating the company’s expanding CPU ambitions. Bank of America reported that management is targeting a roughly $20 billion “Vera CPU” opportunity in the second half of fiscal 2027, with that figure split between traditional AI server head-node processors and standalone CPUs intended for emerging agentic AI and reinforcement-learning workloads. The note further said NVIDIA estimates its AI-related CPU total addressable market at about $200 billion, framing CPUs as another lever for monetizing AI buildouts.

In NVIDIA’s own materials, Vera is described as a CPU purpose-built for agentic AI and reinforcement learning, which are workloads where models move beyond answering questions and instead take actions, use tools, run code, and validate results. NVIDIA said Vera delivers results with twice the efficiency and 50% faster performance than traditional rack-scale CPUs, and that it is designed for organizations building “AI factories” at scale. NVIDIA also said Vera serves as the host CPU for NVIDIA Vera Rubin platforms through second-generation NVLink-C2C interconnect technology, offering up to 1.8 terabytes per second of coherent bandwidth between CPU and GPU.

The “AI factory” framing extends beyond CPUs. NVIDIA’s technical description of the Vera Rubin platform emphasizes co-design across compute, networking, power delivery, cooling, and system architecture, aiming to support sustained intelligence production at large deployments. NVIDIA said the Vera Rubin platform’s flagship rack-scale system, the Vera Rubin NVL72, is engineered to function as a rack-scale accelerator within a larger AI factory, and that platform-level components include a next-generation NVLink with multi-terabyte-per-second scale-up bandwidth. Separately, NVIDIA said Vera Rubin is moving into full production, highlighting network fabric elements such as Spectrum-X Ethernet Photonics, co-packaged optics switching now in production.

On valuation, Bank of America’s pitch in the reporting is that NVDA remains attractively priced relative to its growth prospects. The note reportedly said NVIDIA trades at roughly 16 times expected calendar 2027 earnings, with a PEG ratio around 0.4, below an average cited for the “Magnificent Seven.” Taken together with its system-level thesis, Bank of America kept the company as its top AI compute pick in the coverage.

What is less clear from the available reporting is how every assumption in Bank of America’s model is constructed, including the detailed sensitivity analysis behind longer-range targets referenced in the note. Bank of America did not publish the full analyst report in the coverage, and some future-oriented items discussed by management in public forums, such as how much “content per gigawatt” could rise across multiple architecture generations, remain dependent on product ramp schedules and customer adoption.

Investors who are trying to identify “the next Nvidia” may still be forced to watch NVDA closely, not only for GPU shipments but also for whether its newer CPU and rack-scale positioning converts into durable revenue. Near-term indicates to watch include the timing and adoption of Vera CPU systems, production ramps for Vera Rubin rack components, and whether hyperscalers continue sustaining AI infrastructure spend at a pace that can absorb new capacity without forcing pricing pressure.