AI enters the era of CPUs. To no one’s surprise, this is bad news for the consumer.

The current state of data centers is redefining the production lines of the main technological players. What seemed like a specific crisis in the price of RAM and SSDs has ended up becoming a tsunami that is sweeping away products and the consumer market. The data centers They need the same components as consumers and the rest of the industry, and there was one component that seemed safe: the processors.

It’s over.

Danger! During his presentation of results In the first quarter of 2026, Intel gave a worrying piece of information: the ratio of CPUs and GPUs in data centers could soon reach 1:1. So far, we’ve talked about memory and GPUs as the primary hardware in data centers, but there needs to be a CPU running the show, and currently, there was one CPU for every eight GPUs. However, things are starting to change due to agentic AI.

Training is still important, but now we seem to be entering the era of inference, and that’s where CPUs excel. David Zinsner, Intel’s chief financial officer, described in the call with investors that the CPU/GPU ratio had already gone from 1:8 to 1:4 (one CPU for every four GPUs), but that this agentic AI was exploding the memory of the CPUs, approaching the aforementioned 1:1.

Change of course. I think you can now anticipate where the shots are going to go. As we read in Tom’s Hardwarethis action has caused a reaction: that Intel begins to move its production lines to begin to reduce its capacity in consumer products and increase the output of Xeon processors, which are indicated for servers and continuous work.

All because, currently, delivery times for server CPUs are about six months away and they cannot allow AMD, which also has its products for servers, to beat them in that race.

Consequences. Price increases. Without palliatives. Xeon CPU prices are estimated to have risen in March by 10% to 20% due to that shortage, but consumer CPUs have also increased in price by 5% to 10%. It is not going to stop there, as another 10% price increase is expected for the second half of the year. In fact, it is the same as with mobile phones, with Intel pointing out that the consumer PC market will decrease by double digits this year.

Here it happens as in the rest of the segments: manufacturers are raising prices for everyone (hyperscalers and consumers), but they allocate their production to data centers because they are the ones that will buy the most volume from them. Intel doesn’t care if there is no CPU for users because it is not the bulk of its current market, but what it cannot afford is to have hyperscalers for more than six months expectingespecially if data centers are expected to have to mount more CPUs in the short term.

The objective: the great foundry. Intel has been in the doldrums for a few years. In the consumer segment, AMD’s Ryzens have beaten them to the punch, the ARC GPUs have not finished coming together and things were not going well. A series of poor results led the United States government to invest $2 billion to ‘rescue’ the company with one goal: to make it the largest American foundry. Because, even if things did not go well, they are still one of the few companies in the world with the capacity to create chips, like TSMC or Samsung.

They have some of the best machines on the market and that government bailout soon began to bear fruit with clients such as Apple and Nvidia. During the presentation of results A few days ago, Intel declared a net loss of $3.7 billion, but something happened nonetheless: the stock rose 20%. The reason is that investors are not looking at Intel’s present, but rather its future, and the changes applied in recent months seem to be going in the right direction.

They are not the only ones. This change of direction and production lines is not exclusive to Intel. We have seen it in other companies, but it is true that here they directly advocate leaving out the consumer to prioritize the large customer: Big Tech. Something similar happened with Samsung a few days ago, when it was reported that the company had begun to move your LPDDR4 memory production lines to LPDDR5.

This type of memory is better, but also more expensive, which will cause devices that previously mounted LPDDR4 memory (low-end miniPC or entry- and mid-range mobile phones) to have to go directly for LPDDR5 memory that is faster, but also more expensive.

In the end, the translation is the same as always: as users, we are going to have to tighten our belts and hold on with the devices we have for a while longer. How long? Until 2027 if you ask some, 2028 if you ask others or if that… 2030.

Images | Intel

In Xataka | There is no energy for so many data centers and the consequence is clear: half of those planned for 2026 in the US are in danger