Nvidia Moves From GPUs to CPUs as AI Agents Reshape a $200 Billion Chip Market
A customer's early test found Nvidia's Vera CPU handled about twice the AI agent workload of its existing chips. The bigger story is where the money in semiconductors is heading as AI shifts from answering questions to doing work.
1. A customer that used only Intel and AMD chose Nvidia
According to Reuters on Oct. 19, Modal, an AI cloud startup, will adopt Nvidia's Vera CPU in servers built for AI agents.
Modal had used only Intel and AMD CPUs until now. In early testing, a Vera-based system processed about twice as many agent tasks as its existing CPUs. The result matters because it comes from a customer running a live service, not from a benchmark Nvidia published itself.
Modal's growth is also notable. It recently raised $750 million at a valuation of $15.75 billion. Usage has tripled, and revenue has grown more than 15-fold from a year earlier. The company expects $1 billion in revenue over the next 12 months. That is the company's forecast, not realized annual revenue.
Some caution is warranted. The 2x throughput is a result from Modal's specific early test. It does not mean the same gap would appear in every CPU workload. Modal also does not plan to drop Intel and AMD. It intends to run CPUs from all three companies side by side.
2. AI agents make the CPU important again
So far, attention in AI chips has centered on GPUs, which do the heavy lifting in training models and generating large volumes of tokens.
AI agents work differently. They write code and then run it. They analyze files, access databases, link multiple programs and check the results.
- GPU: reasoning and task planning. Runs the language model and decides the next action.
- CPU: tool calls and task execution. Runs code, processes files and controls systems.
- GPU + CPU: verification and iteration. Repeats the loop dozens of times if needed.
If the CPU is slow in this chain, the whole job can stall even when the GPU is fast enough.
Nvidia is aiming at that bottleneck. Vera is designed with 88 Olympus CPU cores and up to 1.2 TB per second of memory bandwidth. The focus is single-thread performance, memory access and the efficiency of running many agents at once.
3. Nvidia is chasing more than one kind of CPU revenue
- Vera revenue expectation: $20 billion or more, Nvidia's outlook for the current fiscal year
- Related market estimate: $200 billion, Nvidia's own estimate
The Vera CPU revenue Nvidia has given investors is at least $20 billion this fiscal year. That is a future projection, not revenue already booked. The roughly $200 billion market size is also the company's own estimate.
The larger picture is the supply structure of the whole server.
| Area | Nvidia's offering |
|---|---|
| AI compute | GPUs, AI accelerators |
| Agent execution | Vera, Grace CPUs |
| Data movement | NVLink, networking |
| Development environment | CUDA, AI software |
Nvidia is widening its role from a supplier of individual chips to an AI infrastructure provider that integrates GPUs, CPUs, networking and software.
For customers, an integrated design could improve operating efficiency. But heavier reliance on one supplier carries a risk: less bargaining power on price later and less flexibility to switch systems.
4. A threat or an opportunity for Intel and AMD?
If AI agents spread, overall CPU demand could rise. Market growth itself is therefore a positive for Intel and AMD.
The question is who captures the higher share and profit in a growing market.
Intel and AMD have the established x86 server ecosystem, software compatibility and customer relationships. AMD, in particular, can respond with the competitiveness of its next-generation EPYC server CPUs.
Nvidia, for its part, can design data movement between GPU and CPU, memory architecture and software optimization as a single system.
Even if Vera is faster on certain tasks, real-world adoption still faces barriers: software switching costs and operational stability. That is also why Modal chose a mixed-server approach.
For now, the current test results alone do not support a conclusion that Intel's and AMD's competitiveness is weakening.
What it means
If the GPU drives the brain of AI, the CPU is the execution layer that lets AI do work in the real world. That is why Nvidia wants to expand into this market.
The real meaning of this news is not that Nvidia has beaten Intel and AMD today. It is the shift it points to: in the AI agent era, a company that finishes whole AI jobs more cheaply and quickly may capture more economic value than one that only builds good GPUs.
The next task for investors is to watch how far that possibility turns into actual revenue and profit.
Insight Times Editorial Desk
