Silicon Clips AI’s Wings

Silicon Clips AI’s Wings

The future of artificial intelligence is no longer being decided only in laboratories of algorithmic design. It is also being decided in wafer fabs, high-voltage grids and server rooms.

EM
Eduardo Martínez de la Fe
Sep 16, 2026
3 min read

For years, artificial intelligence was seen as a matter of brilliant algorithms and immeasurable quantities of data. Technological leadership belonged to whoever wrote the best code or discovered the most elegant mathematical architecture. But industrial reality has overturned that picture: the centre of gravity in contemporary technology has shifted from the logical layer to material infrastructure. Computing has a governance problem.

Wings clipped

It is, in a sense, a logical development. Training and deploying state-of-the-art models requires huge concentrations of computing power, vast electricity supplies and extraordinarily fragile microelectronics supply chains. But there is another side to the story. Advanced semiconductor fabs, the latest generation of graphics accelerators and gigawatt-scale data centres cannot be copied by downloading a file. Silicon therefore imposes physical limits that set the pace of software development: it has its wings clipped.

The security dilemma

This shift also transforms the security dilemma. Early theoretical models of technological competition, such as the foundational work of Naudé and Dimitri, warned that unregulated races encourage competitors to cut corners on safety in order to get there first. Yet the conventional regulatory response  (code audits and voluntary statements of principle) has shown very limited effectiveness in the face of the speed of the market. We see it every day.

Compute governance

Auditing every line of code or monitoring billions of parameters is technically unmanageable. Watching over physical infrastructure, by contrast, is feasible. Public safety institutes are therefore beginning to turn towards compute governance, because hardware has three features that software has never had: it is scarce, detectable and concentrated in the hands of a small number of actors and in a small number of places.

Compute thresholds

The control of compute thresholds, measured in floating-point operations per second (FLOPs), is thus emerging as a leading instrument of public oversight against biological and cyber risks, as well as monopolistic concentration. Regulatory debate is leaving behind the abstraction of a hypothetical superintelligence and moving towards material measures, including export licences for advanced chips, energy quotas and mandatory registers for large server installations.

The hardware economy

Institutions and companies must accept that those who wish to influence the direction of technology cannot simply regulate its final applications: they need to understand the political economy of hardware. It may not be obvious, but the future of artificial intelligence is no longer being decided only in laboratories of algorithmic design. It is also being decided in wafer fabs, high-voltage grids and server rooms. That is where computing capacity becomes strategic power.

Elon Musk, Sam Altman and other leaders in the sector retain their ambitions for a universal artificial mind, but those ambitions have been stripped of their metaphysical aura.

This article is republished from Futuribles. Here's the original article in Spanish and English.

About the Writer

More from Mindplex

Keep reading

Three more ideas worth your time.

Browse Community

Discussion

Join the discussion

Sign in to share a response with the community.

Type @ to mention someone Type / or use + to add a block Highlight text, then choose Link
Loading editor

Comments cannot be edited after posting because they become part of the reputation record. Give yours a quick review first.