Nvidia’s AI CPU Push Threatens Intel and AMD

• Meta’s Muse became the No. 1 free iPhone app in the United States within 10 days, providing an early consumer-scale example of how AI agents can increase demand for cloud CPUs as well as GPUs.
• One agent does not require one physical CPU, but millions of agents can create substantial demand for virtual CPU hours, memory, storage, browser sessions, and secure cloud environments.
• Nvidia and Arm may capture more of this growth than many investors expect, while Intel and AMD can increase CPU revenue and still lose share of the expanding market.

For most investors, the artificial intelligence infrastructure boom has been a GPU story. Meta Platforms (NASDAQ: META | META Price Prediction) just gave the market a reason to look again at the ordinary central processor.

Meta’s new Muse personal agent reached the top of Apple’s U.S. free-app chart only 10 days after its launch. Muse can research products, fill out forms, handle email, make reservations, and continue working after the user closes the app. That makes it different from a chatbot that waits for a question, generates an answer, and then stops.

The important detail for semiconductor investors is where that work takes place. Meta says Muse operates in a secure virtual machine that isolates the agent’s activity and the user’s data. The environment supplies the browser, processing, memory, and storage needed to perform a series of tasks. The large language model still depends heavily on GPUs for reasoning and inference, but the agent also needs CPU resources to run software, direct the browser, move data, manage permissions, call tools, and keep the job operating.

Muse therefore puts a consumer product around a change already taking place inside the data center. AI is moving from models that answer questions to agents that perform work. That transition expands the infrastructure bill beyond accelerators and creates a second source of growth in general-purpose computing.

This does not mean Intel and AMD automatically collect the new revenue. Nvidia (NASDAQ: NVDA) is preparing its Vera CPU for precisely these workloads, cloud providers are deploying processors of their own, and Arm Holdings (NASDAQ: ARM) supplies the architecture behind many of the challengers. The CPU market is growing, but its competitive structure is changing at the same time.

An AI Agent Is More Than a Model

A conventional chatbot is relatively easy to understand. A user enters a prompt, a model processes it, and the application returns an answer. Most of the demanding mathematics occurs on an accelerator, which is why Nvidia’s GPUs became the economic center of the first phase of generative AI.

An agent adds an execution layer around the model. It may open a browser, consult several websites, retrieve information from email, update a calendar, run code, compare results, and ask the user to approve a purchase. Each step requires coordination. Some steps return to the AI model, while others are handled by a CPU because they resemble the mixed, sequential work performed by a conventional server.

Muse makes this distinction visible to an ordinary smartphone user. The phone is the interface, but the work is taking place in the cloud. According to reports describing Meta’s architecture, a secure virtual machine separates the agent’s activity from other users and limits access to sensitive information. That virtual environment can continue operating after the app has been closed.

It would be incorrect to conclude that one billion agents require one billion physical processors. Cloud systems divide a server into many virtual environments, and the number of agents that can share a processor varies with the complexity and duration of their assignments. The useful measurements are concurrent agent sessions, virtual CPU hours, memory consumption, storage activity, and the amount of software work performed around each model call.

That qualification does not weaken the investment case. A popular consumer agent can generate repeated, persistent computing activity from each user. An agent that works for 20 minutes across several websites consumes a different mix of infrastructure than a chatbot that produces a response in seconds. As more agents monitor inboxes, manage schedules, conduct research, write and test software, or negotiate purchases, their cumulative demand becomes meaningful even when processors are shared efficiently.

Nvidia Is Selling the CPU With the AI System

Nvidia introduced the Grace data-center CPU in 2021, but Vera represents a more ambitious move. Grace was commonly sold as part of Nvidia systems. Vera is being positioned both within the Vera Rubin platform and as a stand-alone server processor.

Vera contains 88 custom Olympus cores built on the Arm architecture. Nvidia designed it for the work surrounding accelerated computing, including agent orchestration, data processing, analytics, tool use, and sandboxed code execution. Those are the same categories of work that a product such as Muse brings into focus.

Management has cited visibility to approximately $20 billion in server CPU demand and has said CPU revenue could more than double in fiscal 2028. Investors should treat the $20 billion as a stated opportunity rather than booked revenue. Even with that distinction, the figure is striking because Intel (NASDAQ: INTC) and Advanced Micro Devices (NASDAQ: AMD) do not separately report their server CPU sales, and their data-center segments also include accelerators and other products.

Nvidia’s advantage is not based on Vera having the industry’s highest core count. AMD’s sixth-generation EPYC 9006 family, code-named Venice, offers as many as 256 cores and 512 threads. Intel retains a large installed base, mature software compatibility, and deep enterprise relationships. Both companies can make persuasive technical cases for their processors.

Nvidia approaches the purchase from another direction. A customer selecting a Vera Rubin rack has already chosen Nvidia GPUs, networking, interconnects, software, and system architecture. The CPU arrives as part of that design. Intel and AMD traditionally competed for an individual processor socket; Nvidia is attempting to make the socket one element of a larger platform decision.

That difference matters because AI systems are becoming more tightly integrated. Performance depends on how quickly data moves among the CPU, accelerator, memory, and network, not simply on the specification of one chip. Nvidia can optimize those connections across the rack and use CUDA and its networking products to keep the customer within its platform.

The strategy does not require Vera to displace every Xeon or EPYC processor. Nvidia only needs to increase the amount of semiconductor content it captures from customers already buying its accelerators. Every Vera placed beside a Rubin GPU represents CPU revenue that was previously available to another supplier.

Nvidia’s Arm Strategy Survived a Failed Acquisition

Nvidia’s move into CPUs did not begin with Muse or Vera. The company has used Arm technology for years in products serving automotive, mobile, robotics, and embedded markets. In September 2020, Nvidia agreed to acquire Arm from SoftBank for $40 billion.

Arm develops processor architectures and licenses them to companies that design chips. Since many Nvidia competitors also rely on those licenses, regulators opposed placing Arm under Nvidia’s control. The companies terminated the acquisition in February 2022. Nvidia forfeited the $1.25 billion prepaid at signing and recorded acquisition-related costs, but it retained a 20-year Arm architecture license.

Nvidia later invested approximately $100 million in Arm’s 2023 public offering, purchasing about 1.96 million shares. It eventually sold the equity position, yet the strategic commitment continued through proprietary processors. Grace and Vera show that Nvidia did not need to own Arm to build a CPU business around the architecture.

The failed transaction changed the route, not the objective. Nvidia moved from an effort to own the architectural supplier to designing Arm-based processors and controlling how they operate inside Nvidia systems. That history helps explain why Vera should not be dismissed as a side product created after the GPU boom. It is the latest step in a long campaign to capture more of the computing system.

The CPU Market Can Triple Without Preserving Share

The Information Network estimates that AI data-center systems revenue will increase from approximately $560 billion in 2026 to $1.68 trillion in 2030, a compound annual growth rate of 31.6%. Accelerators remain the largest component, rising from an estimated $350 billion to $970 billion, while AI CPU revenue increases from $38 billion to $155 billion. That produces a faster 42.1% annual growth rate for AI CPUs.

The difference in growth rates reflects a broader mix of AI workloads. Training remains accelerator-intensive. Inference adds more interaction with applications and databases. Agentic AI adds persistent software execution, orchestration, retrieval, memory management, and secure virtual environments. CPUs do not replace GPUs in this model; they perform the work that surrounds and feeds them.

According to Table 1, The Information Network estimates that the overall server CPU market will grow from approximately $60 billion in 2026 to $195 billion in 2030. The supplier figures are estimates, not company guidance, because Intel and AMD do not disclose stand-alone server CPU revenue and Nvidia’s stated demand should not be confused with recognized sales.

Table 1: Estimated Server CPU Revenue by Supplier, 2026E–2030E

Source: The Information Network

The table does not forecast the disappearance of Intel or AMD. Intel’s estimated server CPU revenue more than doubles by 2030, but its share falls from 33.3% to 22.1% because it grows more slowly than the market. AMD is better positioned, reaching an estimated $69 billion and remaining the largest individual supplier in the forecast.

The largest structural change comes from Arm-based processors. Nvidia and other Arm suppliers, including processors designed internally by cloud companies, represent approximately 42.6% of server CPU revenue by 2030. Intel and AMD retain the other 57.4% through x86 products. A larger market allows all four categories to grow, while the allocation of new revenue shifts sharply.

The Cloud Companies Can Win Twice

Amazon (NASDAQ: AMZN), Microsoft (NASDAQ: MSFT), and Alphabet (NASDAQ: GOOGL) initially captured AI infrastructure demand by renting access to GPUs and other accelerators. Agents give the cloud providers another way to earn revenue from the same customer.

An agent may call an AI model for reasoning and then spend additional time in a CPU-based virtual machine running the resulting plan. It may also use databases, storage, networking, identity services, and security tools. The cloud provider can charge for both the accelerator work and the conventional computing services wrapped around it.

The cloud companies also design their own Arm-based CPUs. Amazon has Graviton, Google has Axion, and Microsoft has Cobalt. Internal processors allow them to reduce costs, tune hardware for their workloads, and retain economics that would otherwise go to Intel or AMD. Agentic AI therefore expands the market for CPUs while strengthening some of the largest buyers’ incentive to build those CPUs themselves.

This creates a complicated result for merchant chip companies. Growing demand for cloud CPU hours benefits the data-center industry, but it does not guarantee a proportional increase in processors purchased from Intel or AMD. The hyperscaler may satisfy part of the growth with its own silicon, purchase Nvidia Vera for systems tied to Nvidia accelerators, and reserve x86 processors for applications that require compatibility or a particular performance profile.

The supply environment already suggests that general-purpose compute is no longer an afterthought. Reuters reported earlier in 2026 that Intel warned some Chinese customers of server CPU lead times reaching six months, while AMD cited delays of eight to 10 weeks for certain products. Prices in China had increased by more than 10%, according to that report. Those figures are not proof that AI agents alone caused the shortage, but they show that CPU demand was tightening before Muse became a hit consumer application.

Arm Is Becoming a Competitor as Well as a Licensor

Arm benefits whenever Nvidia, Amazon, Google, Microsoft, or another customer ships a processor using its architecture. It receives licensing fees and royalties, not the full selling price of the finished chip. Investors should not mistake the total value of Arm-based CPUs for Arm’s own revenue opportunity.

The company is nevertheless moving closer to the silicon market. Arm has discussed an AGI CPU sold as a finished product, and management has expressed confidence in a $2 billion AI-chip sales target after demand exceeded the manufacturing capacity it initially secured. That move can lift Arm’s revenue per system, although it also places the company in more direct competition with customers that historically licensed its technology.

For Nvidia, Arm’s expansion is both validation and risk. It supports the thesis that agentic AI creates a large market for power-efficient CPUs, but it adds another supplier to a field that already includes three hyperscalers with internal designs. Nvidia’s defense is system integration. Vera can be sold alongside GPUs, networking, and software in a configuration built for customers that value a complete AI platform.

Muse Also Reveals the Limits of the Thesis

Muse’s rapid rise demonstrates consumer interest, not yet the durability of the business. Agents must earn users’ trust because they may need access to email, calendars, financial information, and shopping accounts. Amazon has already blocked Muse from accessing its platform, citing unauthorized AI access and a failure to identify itself properly. That dispute shows that agents cannot assume permanent access to every website or service they are asked to use.

Security can also increase the cost of each agent session. Isolated virtual machines, approval controls, monitoring, and protection against malicious instructions require more infrastructure. Those expenses support the CPU demand thesis, but they can pressure the economics of a consumer service if most users remain on a free plan.

Meta has the advertising base and distribution to subsidize adoption while it develops subscriptions and commerce revenue. Smaller agent providers may not have that luxury. If customers resist paying, or if websites restrict automated access, usage can grow without producing an attractive return on the data-center investment.

Hardware utilization is another variable. Cloud operators continually improve scheduling so that more virtual agents share each physical processor. Better utilization reduces the number of servers required for a given amount of activity. The bullish case therefore depends on total agent usage growing faster than infrastructure efficiency, a plausible outcome but not a certainty.

The Bottom Line

Meta Muse has turned an abstract data-center forecast into a product that ordinary consumers can understand. An AI agent does more than generate language. It browses, executes, stores, retrieves, and remains active. GPUs perform the heaviest model calculations, while CPUs manage much of the work required to turn those calculations into action.

That division of labor is why CPU revenue can grow faster than accelerator revenue through 2030 without displacing GPUs. The Information Network expects AI CPU revenue to rise from $38 billion in 2026 to $155 billion in 2030, while the total server CPU market reaches approximately $195 billion. The expansion is large enough for Intel and AMD to grow, but growth alone will not preserve their market shares.

Nvidia holds the strongest strategic position among the merchant suppliers because it can sell Vera into an installed base already buying its GPUs, networking, and software. AMD remains the strongest stand-alone server CPU competitor and leads the 2030 supplier forecast. Intel retains a durable enterprise franchise, but it faces the greatest relative share pressure as new spending moves toward AI systems and Arm architectures.

Arm may be the broadest beneficiary because it participates in Nvidia Vera, hyperscaler processors, and other designs. Its decision to sell an AGI CPU directly could increase the amount of value it captures, while also changing its relationships with licensees.

Muse does not prove that every AI agent will become profitable or that every virtual session will require a new processor. It does establish that agentic AI is becoming a consumer workload rather than a laboratory concept. For semiconductor investors, the next phase of AI spending will be measured in more than GPUs and tokens. It will also be measured in virtual CPU hours, active software environments, and the amount of the complete system each supplier can capture.

Contact [email protected] for any questions or corrections.