Local AI that threatens the hyperscalers: the scenario nobody wants to see

The end of data centers: the AI that will run on your laptop

What if the future of artificial intelligence lay not in Microsoft's or Google's enormous data centers, but in the chip of your own computer? The thesis is as simple as it is uncomfortable for the tech giants: processors will become cheaper and more efficient, and the AI models good enough for 95% of everyday tasks will end up being free. The result would be a world where every user has their own "tiny data center" and the big cloud providers lose their central role.

The trade war that is accelerating the democratization of AI

In the context of competition between China and the United States, the strategy of giving away AI models at the user level has a clear goal: to attack the revenues of the large American companies that underpin the country's technological superiority. If Chinese models are offered for free and are good enough for most needs, the paid subscription business starts to wobble. The logic is that of attacking the adversary's weak points, and the most vulnerable point is AI monetization.

The Jevons paradox: the uncomfortable counterargument

There is a nuance that unsettles the thesis of total decentralization. The Jevons paradox states that when a resource becomes more efficient and cheaper, its total consumption increases rather than decreases. If chips become more powerful and accessible, AI usage will skyrocket, and with it the need for infrastructure. Falling prices do not kill the hyperscalers; they turn them into a commodity consumed at a much larger scale.

The telecom bubble as a historical mirror

The analogy with the telecom bubble of three decades ago is revealing. Back then, technological disruption generated a price bubble for the companies that peered into the revolution. The business model changed, the required capex soared, and the players were reshuffled. In Spain, the name Julio Linares is cited as the great champion of the information society. The question is whether AI will trinc the same pattern: an initial bubble, a reshaping of the sector, and the disappearance of those who fail to adapt.

The survivors of the ecosystem

If AI moves to a local model, not all players suffer equally. Manufacturers of powerful processors such as TSMC, Arm or Qualcomm would keep a guaranteed market. Lithography companies like ASML would continue to enjoy a captive market. RAM and storage manufacturers would still have room to grow, though with smaller margins. Nvidia, which today sells GB200 and 300 cards at exorbitant prices, would not earn the same in a consumer world. The open question is what happens to companies like TeamViewer, Gartner or Oracle in this new scenario.

The OPEX problem: AI as an expense, not an investment

There is an aspect few analyses consider: AI will become an operating expense (OPEX) for companies. If it replaces workers, it is a cost that states will have to regulate. If it replaces no one, the business is not sustainable. The crossroads is obvious. AI will enable new business models, but it will also make companies disappear, replaced by more competitive ones. The transition will not be painless.

---

Dependence on AI is already brutal, as the two-hour outage that made billions of people tremble demonstrated. The technology that promises to free us from centralized data centers could, paradoxically, increase our dependence on them. Lateral thinking has its merit, but the market has its own laws. And history shows that tech bubbles do not always end where the most creative analysts predict.