熊

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

LongbridgeAII'm LongbridgeAI, I can summarize articles.

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.

The copyright of this article belongs to the original author/organization.

The views expressed herein are solely those of the author and do not reflect the stance of the platform. The content is intended for investment reference purposes only and shall not be considered as investment advice. Please contact us if you have any questions or suggestions regarding the content services provided by the platform.

Author's posts

View more
  • 从美光$mu 最近的发展谈”惊喜”

    首先,说件非常重要的事:我因私人原因,最近很少贴文,可是冒充我的骗子号并没有中断偷盗别人的内容发文有人朋友会以为我换号了,并且还有朋友被骗子私下联络了。这里,镇重的申明:我在脸书只有一个有蓝勾认证过的号,没有其他号,也没有在其他平台有别的号。简易逻辑:我连贴文都没时间了...

  • AeroVironment: Contract suspension, is it a 'trap' or a 'golden trap'?

    1. Key event: The U.S. Department of Defense issued a "Stop Work Order" The U.S. Department of Defense (DOD) issued a "Stop Work Order" to $ Aerovironment.US. This primarily targets the BADGER phased array antenna system under the SCAR (Squad Common Equipment Transport) program. Reason: Not due to product failure, but to "renegotiate contract terms." The government aims to shift the contract from a "Cost-Plus" model to a "Fixed-Price" model...

  • Meta 收购 Manus 的根本原因是什么?为什么网上的质疑声这么大?

    这事吧,大多数人第一眼就看错了。大家都在说:Manus 是 Agent、是生态位、是套壳。但说实话,Meta 根本不是冲这些去的。你站在扎克伯格的位置想一秒就明白了:他现在最焦虑的根本不是 “有没有一个好看的 Agent 产品”,而是——AI 战争,Meta 在数据和实战经验上落后了。Manus 真正值钱的就两样东西。第一是 工程环境。不是 PPT 里的 demo...