The UN puts numbers on Latin America's absence from AI

A UN scientific panel confirms that 90% of the compute that trains the most capable systems is in two countries, and the region does not appear on the map.

Generated automatically · sources linked · no prior human review

UN Secretary-General António Guterres, together with Yoshua Bengio and Nobel laureate Maria Ressa, presented the preliminary report of the Independent International Scientific Panel on AI, a group of 40 experts whose secretariat is shared by UNESCO and the International Telecommunication Union. Its central conclusion is twofold: current safeguards are not keeping pace with the growth of AI capabilities, and the power to build these systems is extraordinarily concentrated. The United States controls about 75% and China around 15% of the compute (the processing capacity) that trains the world’s most advanced models. Together they account for 90%.

In that snapshot of power, no Latin American country appears. For the region, the figure matters beyond the statistic: it is the first time a multilateral scientific panel, backed by the UN, has put a number on a pattern that had been sensed. The report feeds into the first Global Dialogue on AI Governance, held in Geneva on July 6 and 7 alongside the ITU’s AI for Good summit. The uncomfortable question is what the region brings to that table when the infrastructure that gives a voice in these discussions is almost entirely in another hemisphere.

Mexico’s attempted answer came this same week. The government presented 13 technological sovereignty projects to its Congress, with the Coatlicue supercomputer as the centerpiece: 314 petaFLOPS, 14,480 GPUs and an investment of close to 6 billion pesos. It is the most detailed compute sovereignty package a country in the region has brought before a legislature. By contrast, a column drawing on MIT data points out that only 23% of Latin American organizations capture any economic value from AI, compared with 39% globally: building infrastructure is only half the challenge.

Also today

In the region

The week brings together three layers of the same problem. At the top, the UN report quantifies for the first time, with scientific weight, Latin America’s absence from the compute map. At the national level, Mexico formalizes before its Congress the region’s most ambitious technological sovereignty bet, while Brazil takes a step that is more symbolic than substantive with its new Parliamentary Front on AI. And underneath, the adoption gap persists: fewer than one in four organizations in the region turns AI into economic value. The regional conversation is no longer only about rules and is starting to be, also, about who has the machines.

Launches

  • Claude Sonnet 5 — Anthropic makes it the free default model for its Free and Pro plans, with performance close to its top of the line at lower cost and standard multilingual support, Spanish included.
  • Grok Voice Agent Builder — A no-code platform from xAI for building phone voice agents in minutes, at $0.05 per minute of audio and with support for more than 25 languages. An aggressive price compared with alternatives such as ElevenLabs, with a clear use for small businesses’ Spanish-language call centers.
  • Gemini 3.1 Flash Image and Gemini 3 Pro Image — Nicknamed “Nano Banana 2” and “Nano Banana Pro,” they are Google’s pair of image generation and editing models at two levels of price and quality.

Threads we’re following

The UN’s diagnosis does not arrive in a vacuum. In recent weeks we have been following how Latin America is left on the sidelines of global AI governance decisions that directly affect it: first with the export ban that took the Fable 5 model out of circulation for almost three weeks, without the region taking part in either the shutdown or the reopening. The report presented at the UN is the chapter that puts all those episodes in context: these are not isolated cases but a pattern of concentration of compute power that a scientific panel has just documented with figures.


If infrastructure is what gives a voice in these discussions, are a national supercomputer and good will enough for the region to arrive in Geneva with something more than a symbolic seat?

About this entry. It is generated automatically from public sources, without human review before publication. It may contain errors of interpretation or summary; please check each story against its original source (the links lead there) before citing it or making decisions based on it.

Doble Click is written with Anthropic models.

Spotted an error? Report it

Tell us what's wrong, quoting the sentence if you can and, if you have it, the source that corrects it. An automated process reviews reports every night: if the error is verified, the page is corrected and a correction note is added at the bottom.

Your email is optional: we only use it if we need more context about the report. It doesn't subscribe you to the newsletter.