← Back to Ratio
Regulation · · 2 min read

DeepSeek and digital sovereignty

DeepSeek showed that the AI frontier fits within a regional budget. What this changes for digital sovereignty and regulation.

#deepseek #digital-sovereignty #regulation #china #open-source

I hate saying “I told you so.” But in this case, I will.

In October 2024, I presented a chapter in the volume coordinated by Saúl López Noriega, before Mexico’s Poder Judicial de la Federación. The month before, I had developed the same argument in a course with White Bx Project. The central thesis of both: the People’s Republic of China was consolidating its position as the world’s leading AI actor. This was empirical analysis, grounded in Kai-Fu Lee’s AI Super-powers and a comparative study of Chinese and Western regulatory frameworks. Industry debates revolved around the Silicon Valley narrative; the data pointed to China.

Today, the market is catching up, via DeepSeek.

What DeepSeek is and its market impact

DeepSeek is an open-source language model developed by a Chinese laboratory. It performs at the level of frontier US models, with training that demanded one-twentieth of the computational cost. The market reaction was immediate: exchanges entered a downturn and NVIDIA’s share price fell.

The reaction makes sense. The economic logic of recent years rested on a premise: the AI frontier required infrastructure at scale, and the providers of that infrastructure (chips, data centers, proprietary clouds) would capture the bulk of the value. If less digging is needed to find gold, fewer shovels are sold.

Open source as a regulatory instrument

What matters is what happened after training: all of DeepSeek’s details were made public. Architecture, weights, methodology. Any actor with the computational resources can replicate them, independent of the Stargate project announced by the United States.

This shifts the digital sovereignty debate. Tools exist to build AI autonomy outside of 500-billion-dollar budgets. The argument that frontier AI required investments only a handful of laboratories could finance was, in part, a market argument. DeepSeek eroded it.

A regulator’s access to a system’s architecture shapes the quality of the rules it can produce. With a public architecture, the regulator or judge works on the actual system. With a closed one, the work proceeds on the providers’ marketing representations.

A lawyer who understands the technical domain negotiates rules from a position of strength and litigates with discernment. Understanding the mechanisms also redefines the interlocutors that matter, both in rulemaking and in litigation.

It redefines, too, the grounds on which many Latin American governments had deferred regulation. DeepSeek erodes that asymmetry. The cracks in the premise were there from the start.


(In truth, everyone loves saying “I told you so.” It’s just that nobody admits it.)

Frequently asked questions

Why does DeepSeek matter for AI regulation?
DeepSeek released its full architecture and weights. Any actor with the computational resources can replicate them, independent of US laboratories. This shifts the digital sovereignty debate: the argument that only countries with multi-billion-dollar infrastructure can regulate with informed judgment loses its foundation.
What is digital sovereignty in the context of AI?
The capacity of a state, region, or organization to understand, audit, and, when necessary, reproduce the technologies it uses and regulates, free of structural dependence on external actors. It requires real alternatives, the ability to examine the systems being regulated, and the development of independent technical judgment.

Adapted to the site style. Substance preserved.

Original LinkedIn post — Originally published on January 27, 2025 · read the original