Why did semiconductor giants AMD and Intel surge? The key lies in the CPU

The hottest stocks in the semiconductor sector these two days are undoubtedly $AMD(AMD.US) and $Intel(INTC.US). They share one thing in common: both are CPU giants.

Recently, the well-known institution KeyBanc released a research report, revealing a bombshell: AMD and Intel's server CPU production capacity for 2026 has basically been sold out!

This means market demand is insanely strong. More importantly, when supply can't meet demand, sellers gain "pricing power." The report mentions that both companies are considering raising server CPU prices by 10% to 15% in Q1 2026.

In the past, AMD mainly relied on "winning by volume" to grab market share; now, it has finally entered the advanced stage of "rising volume and price together." This is exactly the script Wall Street loves most.

You might wonder: Isn’t the protagonist of the AI era the GPU from $NVIDIA(NVDA.US)? Why is there a shortage of server CPUs?

▋ The Overlooked "Inference" Demand Explosion

Many people think CPUs are unimportant in the AI era, which is actually a big misconception. The role of CPUs undergoes a fundamental shift depending on the two different stages of AI—"training" and "inference."

1. CPU in the Training Phase: The "Construction Director" of an Airport

You can think of AI model "training" as building a massive international airport.

GPU is the core construction team: Responsible for laying runways and building terminals. Progress mainly depends on the team's hard skills.

CPU is the construction director: It doesn’t lay bricks but manages material scheduling (data preprocessing) and progress management (setting checkpoints).

Summary: At this stage, the CPU is a support-type commander. If it’s not up to par, the project will be chaotic, but the final speed still depends on the GPU.

2. CPU in the Inference Phase: The "Air Traffic Controller" in the Tower

After the model is built, it enters the "inference" stage, meaning the airport starts operating 24/7. At this point, the battlefield changes completely.

CPU becomes the tower controller: Every user request is like a plane waiting to land.

Its job: Receive requests, prioritize, and distribute them. To keep the runway (GPU) busy, it must cleverly bundle several small planes (small requests) together for processing.

Bottleneck effect: If the "controller" (CPU) is slow, even if the runway (GPU) is wide and long, planes can only circle in the air. This leads to low GPU utilization, causing users to experience lag and delays when using AI.

Data shows: Simply switching to a higher-frequency CPU can reduce the average latency of the entire AI inference system by 8% to 9%. The time saved is entirely squeezed out by improving the "controller’s" efficiency.

▋ Why is understanding this distinction important?

Simple summary:

During training: The CPU ensures the GPU has work to do.

During inference: The CPU determines how efficiently the GPU works.

If you only stack expensive GPUs but pair them with a garbage CPU, it’s like giving an eight-runway super airport a sluggish control tower—a pure waste of resources. As AI fully enters the era where "inference" determines victory, the status of CPUs is making a strong comeback.

▋ The "Dual Narrative" Under the Noise

Looking at AMD’s development, there are actually two storylines:

Visible line (GPU): Challenging NVIDIA—that’s the future’s vast frontier. Though exciting, it’s still playing catch-up.

Hidden line (CPU): Undermining Intel—this is the current foundation. The profits and cash flow from CPUs support AMD’s bold investments in the GPU battlefield.

To sum up: While the whole market is watching that dazzling GPU showdown, the CPU power shift and price surge happening inside server racks are the key pieces that make up these companies’ real value.