Bank of America identifies $170 billion growth opportunity in Nvidia's optical networking and custom silicon platform.
Bank of America sees Nvidia's AI CPU strategy supporting substantial long-term upside. While Nvidia's momentum has been driven by GPU dominance, the investment bank argues the company's next major opportunity extends well beyond that category.
Google parent Alphabet recently strengthened the case for continued AI infrastructure spending by raising its 2026 capital expenditure forecast to $195 billion to $205 billion, underscoring that demand remains intense. Nvidia closed on Wednesday, July 22, at $212.06, up 2.3%, and has delivered a 14.7% gain over the past six months compared with the S&P 500's 8.5% gain.
Bank of America analysts, led by veteran analyst Vivek Arya, argue that the server CPU market for agentic AI represents Nvidia's next major frontier. They estimate this specific market could expand nearly fourfold to $170 billion by 2030—a substantial opportunity beyond Nvidia's current GPU-centric business.
Nvidia is pushing into the CPU market, long dominated by Advanced Micro Devices (AMD) and Intel (INTC), with its Vera CPU platform. Vera specializes in single-threaded CPU performance, delivering quick task completion for individual processors. CEO Jensen Huang has directly linked Vera to agentic workloads, stating that "AI agents will be the largest users of computing. Vera is the first CPU designed for that future."
Agentic AI workloads are sequential in nature, with each action undertaken by an AI agent depending on the previous one being completed. By shortening the latency in each CPU step, the total time to complete an agent's task improves substantially. Vera's monolithic compute die is designed to improve coherence and prevent data-access bottlenecks while keeping communications on a single chip to substantially reduce delays.
Beyond CPU performance alone, Nvidia is pursuing a system-level integration strategy. Vera is being co-designed with multiple other Nvidia technologies, including BlueField networking and storage products. Customers purchasing an Nvidia stack receive CPUs, GPUs, networking, interconnects, software, and orchestration technologies designed to work together—substantially improving performance while raising switching costs and positioning Nvidia's architecture as the default platform for large-scale AI factories.
Bank of America maintains a Buy rating on Nvidia stock with a $350 price target, implying roughly 68.8% upside from $207.29. The target equates to 26 times 2027 estimated earnings adjusted for the company's net cash position, sitting near the bottom of its historical forward PE range of 25 to 56 times. Nvidia stock is currently trading at 23.6 times forward non-GAAP earnings, 46% below its five-year average. With the company's leadership in both AI computing and data-center networking, Bank of America analysts believe it deserves a premium valuation.
Three primary concerns moderate this multiple: AI project lumpiness, as large sovereign, hyperscale, and enterprise projects can be delayed or resized; gaming cyclicality, given Nvidia's exposure to consumer markets vulnerable to inventory corrections; and power constraints, as customer demand may exceed the industry's capacity to build, power, and cool new data centers.
AMD presents a different architectural vision. The company argues that Nvidia is optimizing for the wrong constraint—focusing on faster task completion for individual AI agents rather than overall rack-level efficiency. AMD contends that commercial AI resembles an enterprise software platform with multiple moving parts, including APIs, databases, middleware, and caching layers. Under this model, what matters is how many agents a rack can support within a fixed budget. AMD's EPYC 9965 Turin processor is estimated to deliver nearly 2.4 times the rack-level throughput of Nvidia's Vera baseline in a modeled 100-kilowatt deployment, with its next-generation Venice platform expected to increase that advantage to 3.3 times.
The competitive debate also extends to ARM versus x86 architecture. Nvidia argues that the underlying chip standard matters less if its processors and software work together to deliver better overall AI performance. AMD and Intel, however, have a significant advantage: most business software was built around x86 architecture. For decades, companies have used x86 systems to run databases, security tools, operating systems, and internal applications—meaning the software is already tested, familiar, and widely supported. This compatibility advantage may prove valuable as AI agents work directly with corporate systems.
Bank of America does not declare a winner in this competition, noting that the agentic AI market remains in its early innings. Nvidia's bull case centers on Vera strengthening its full-system strategy; if customers prioritize lower latency, tighter CPU-GPU integration, and better accelerator utilization, Nvidia could substantially widen its competitive moat. AMD and Intel, meanwhile, maintain a practical advantage through their deeply embedded x86 architecture across enterprise software, lowering migration costs and easing adoption for traditional businesses.
The market may ultimately not produce a single winner across every workload. Nvidia could dominate latency-sensitive AI factories while AMD and Intel remain competitive in enterprise environments built around scale, compatibility, and parallel processing. Real-world customer deployments over time will provide greater clarity on where the market is heading and how much of that opportunity is already priced into Nvidia's valuation.