AI Has Also Hoarded Processors, CPUs at Risk of Shortage

The hardware supply chain is facing a new critical phase: after the shortage of memory and GPUs, it's now processors' turn. The cause seems to be the spread of agentive AI models, which introduce workloads that are much more complex than in the past.

Data centers are experiencing increased demand for computational resources that exceed available production capacity. In this scenario, CPUs have taken on an increasingly central role and are becoming an evident bottleneck. The new models do not limit themselves to inference but require intricate operations that include access to databases, simulations, and management of physical processes—areas where the power of CPUs is crucial.

This change is also reflected in server design. Previously, a single processor managed multiple GPUs within the same stack, with a heavily unbalanced ratio. Today, the configuration is changing: CPUs and GPUs have reached a more balanced operational weight, signaling a rapid structural transformation of AI infrastructure.

The effects are often directly noticeable by users. Some digital services show instability and slowdowns, as reported on various software development platforms. These are indirect indicators of the growing pressure on cloud infrastructures.

This has clearly generated a new imbalance between supply and demand: CPU suppliers, especially at the enterprise level, cannot keep pace with demand. We're not just talking about AMD EPYC or Intel Xeon processors, but also about solutions based on ARM architecture (like Amazon Graviton) or RISC-V designed for specific applications.

OpenAI ported their entire codebase to ARM so they can use $AMZN graviton CPUs as CPU capacity is very short.

Many suppliers like Microsoft or the aforementioned Amazon have already allocated much of their CPU production to projects dedicated to artificial intelligence, which leaves other sectors underserved that have increased their demand. On the other hand, it is precisely the supply constraint that has led AI model providers to shift some workloads to architectures other than x86.

Needless to say, like with memory, the demand for CPUs is prompting various companies to reallocate production lines. This means a reduction in consumer production, leading to increased prices and reduced availability.

In this case, too, expanding production capacity represents an economic risk as well as a long maneuver to implement. The establishment of new foundries would take years and there's no guarantee that the investment will be recouped, at least in the short or medium term.

In short, it seems that AI is cannibalizing the entire hardware supply chain, and really difficult times are ahead for both consumer users and professionals. At present, there have been no comments from CPU suppliers regarding this situation, but it does not bode well.