Mexico 2025: The AI That Lives in PowerPoint

View on LinkedIn

Like a good politician on the campaign trail, making promises costs nothing.

And the Mexican government has been lavish with promises about artificial intelligence: national laboratories, sovereign language models, supercomputers with Aztec names, public AI schools. What is in short supply are finished products. As of December 2025, the inventory of what actually works fits on a cocktail napkin.

2025 Was a Fertile Year for Announcements.

In April, President Claudia Sheinbaum promised a National Artificial Intelligence Laboratory "by October." October came and went, and the laboratory did not. In July, Marcelo Ebrard announced a homegrown language model; in November he presented "KAL" — no technical documentation, no code, no benchmarks. Also in November came Coatlicue, the supercomputer that will be "the most powerful in Latin America" — when it is built, in 2026, if all goes according to plan. And so on: the Public AI Training Center issued a call for applications, but classes begin in January 2026. The National AI Policy is "under development."

The AI Law has multiple bills pending in Congress; none has been passed. What is actually running? Researchers from CIDE set out to find out. They found 119 AI applications reported across the three levels of government. The problem: opacity was widespread, many agencies did not even respond, and 223 supposed "AI" applications did not even qualify as such.

In the Mexican Government, Even the Definition of Artificial Intelligence Is Unclear.

To grasp the gap between announcement and reality, it is enough to review the 2025 timeline.

Ten major announcements in a single year. Zero finished products operating at scale. The pattern is consistent: a catchy name, a future date, and when that date arrives, a new announcement displaces the previous one.

While Mexico Accumulates Announcements, Other Countries Execute.

The United States, on the first day of the new administration, announced Stargate: $500 billion in AI infrastructure with data centers already under construction. But more revealing than the large numbers is something simple: in April, the White House issued two memoranda allowing federal agencies to procure AI technology from external vendors. No national laboratories, no sovereign language models, no supercomputers with Aztec names. Just one directive: you may buy AI — go ahead and do it. Sometimes the most effective public policy is the kind that removes obstacles rather than creating institutions.

And What Is Actually Running?

This is where the inventory gets complicated. MARCia, from the IMPI, exists, but intellectual property professionals prefer not to use it. Sor Juana, the Supreme Court's chatbot, was an experiment confined to a single chamber, with results that the tool itself warned could be inaccurate; the project was shut down. The IMSS has a chatbot pilot for pediatric oncology — valuable, but invisible at a national scale.

None of this transforms public administration. None of this competes with what other countries are already implementing. These are isolated, experimental projects with no continuity or scale. More proof-of-concept than public policy. The irony is that those who most need AI cannot use it. Judges in administrative tribunals and the federal judiciary describe backlogs years deep in their caseloads. When something as basic as jurimetrics is proposed to them — using AI for deadline calculations, damage quantification, statistical projections — the reception is mixed: genuine interest from some, institutional distrust from others, and zero infrastructure to implement anything. The problem is not the will; it is that no one has given them the tools.

In November 2025, federal judges were soliciting contributions from colleagues to buy paper and print rulings. It is hard to talk about artificial intelligence when there is no budget even for toner.

The Problem with Mexico's AI Strategy Is Not a Lack of Vision; It Is an Excess of It.

There is no shortage of Aztec names, national laboratories, sovereign language models, and promised supercomputers. What is missing are finished products, working implementations, tools in the hands of those who need them. While other countries remove obstacles so their institutions can adopt technology, Mexico keeps inaugurating initiatives that will live and die in a press release.

Making promises costs nothing, the saying goes. But it does not build infrastructure either.