CPUs have suddenly emerged as a key focus for the market this year.

The main driver is that OpenAI and Anthropic have signed large-scale partnership agreements with cloud providers such as Amazon, Google, and Microsoft, leading them to consume significant amounts of cloud CPU capacity.
There are two main reasons why AI requires more CPUs.
First, the training paradigm is shifting from traditional pretraining toward reinforcement learning.
After a model generates code, it must run unit tests and validate the results in simulated websites or sandbox environments. These workloads rely heavily on CPUs.
Second, the inference paradigm is shifting from simple chat interfaces toward agents.
Models must call a wide range of tools, including search engines, databases, code execution environments, compilers, and deployment systems. Each of these tool calls requires substantial CPU processing in the background.
There is also an indirect source of demand that is easy to overlook.
The number of GitHub commits globally has already increased severalfold compared with last year. The enormous volume of code generated by AI ultimately needs to be deployed and executed in CPU-based environments.
Since this trend was first highlighted, the share prices of Arm and Intel have risen severalfold, while AMD has also benefited meaningfully.
NVIDIA has even begun selling its Vera CPU as a standalone product and has set a related revenue target of $20 billion.
However, CPUs are unlikely to become the next central pillar of the AI investment cycle.
For example, a single Blackwell GPU costs more than $50,000, while a CPU costs roughly $5,000.
Even assuming one CPU supports two GPUs, every $100,000 invested in GPUs would translate into only around $5,000 of corresponding CPU spending.
The CPU market was clearly undervalued in the past, but the current re-rating is better understood as a return toward a more reasonable valuation rather than evidence that CPUs will become the next major protagonist of the AI industry.