The European Commission opened the call to build seven AI Gigafactories, an instrument that mobilizes more than 30 billion euros: up to 10 billion in European and national public funds that leverage some 20 billion in private money. Each facility is required to house at least 100,000 latest-generation chips (roughly four times the largest data center currently operating in the European Union) and to guarantee access for startups, small and medium-sized businesses and industry to train, fine-tune and use frontier models. Applications close on November 12, awards are expected in early 2027 and construction begins that same year.
What is interesting for Latin America is not the scale, which is out of reach, but the instrument. Europe turned a vague aspiration, “computational sovereignty,” into something that can be audited: an amount, a timeline, an award mechanism and an explicit obligation that the capacity be available to small companies and not only to whoever financed it. The region, by contrast, negotiates data center by data center with large cloud providers that arrive with their own hardware and keep control of the installed capacity. A single figure captures the gap: set against one European gigafactory is the supercomputer of about $5 million on which Latam-GPT was trained at the University of Tarapacá.
The contradiction is also worth stating, because it is the same one any country in the region would face: European sovereignty is being built with letters of intent signed with AMD, Nvidia and Qualcomm. The bloc that invented AI regulation cannot manufacture the silicon it needs either. That does not cancel out the exercise (having a compute policy with figures and deadlines is still better than not having one), but it does change the conversation: the useful question is not how to stop depending on others, but what specific capacity you want to buy, with what budget and under what conditions.
Also today
- Amazon raises its 2026 capex to $220 billion and AWS grows 37% — The large cloud providers are raising spending in the same week the market is punishing them; the chips each cloud designs in-house already bring in more than 25 billion annualized.
- Meta reports nearly $700 billion in future spending commitments — Of that total, 349.3 billion is contractual and non-cancelable: de facto debt even if demand is slow to arrive.
- Leopold Aschenbrenner’s fund liquidates its entire public portfolio and sells it to Citadel — Margin calls from Goldman Sachs, JPMorgan and Bank of America on four-times leverage in memory and chips. AI risk reaches the financial system through credit, not through the technology.
- Nscale buys Anyscale, the company behind Ray — About $1.65 billion for the open framework that distributes AI workloads across GPUs: consolidation is moving up from hardware to the software that orchestrates it.
- Google DeepMind disbands the AlphaFold team — About 25% of the full-time authors of the original papers have already left the company; John Jumper, 2024 Nobel laureate in Chemistry, left for Anthropic in June.
- LinkedIn adds a button to report posts that “seem like AI slop” — The platform that pushed AI-assisted writing for two years now asks users to flag the result, and adopts the word people use to mock it.
In the region
Panama was the regional move of the day, and it was not a small one: President José Raúl Mulino led the launch of the National Advanced Technologies Agenda, which brings together in a single instrument the National Artificial Intelligence Strategy (digital and physical infrastructure, talent development, data governance and cybersecurity, ethics and regulation, investment attraction and sectoral application) and a National Semiconductor and Microelectronics Strategy that commits $105 million, an Advanced Semiconductor Technology Center at the Technological University of Panama, master’s and doctoral scholarships in Arizona and 10,000 digital training scholarships. It is the first country in the region to publish its AI strategy and its semiconductor strategy on the same day, which is to say, to accept that AI policy is industrial policy. The sequence of the past month speaks for itself: joining Pax Silica on July 1 as a pilot of the U.S. supply chain credentialing platform, the installation of the National Commission on Critical and Emerging Technologies on the 8th, the formalization of the AI Subcommission on the 9th and the National Agenda on the 30th. Declared technological sovereignty inside an architecture designed elsewhere, with no regional body anywhere in the scaffolding. Outside the region but with a direct effect on it, on August 2 the transparency obligations of Article 50 of the European AI regulation take effect (disclosing that a person is talking to an AI, marking synthetic content in a machine-readable way, labeling deepfakes), together with sanctioning powers over general-purpose models, with a ceiling of 15 million euros or 3% of global turnover; it is worth clarifying that the obligations for high-risk systems do not arrive on that date, but in December 2027 and August 2028. No new legislative action in Brazil, Chile or Colombia during the week.
Launches
- Gemini Robotics 2 — Three models that for the first time control a full humanoid under a single learned policy. The spatial reasoning model, ER 2, is available to any developer through the Gemini API and Google AI Studio; the motor control and on-device execution models require applying to the Trusted Tester Program.
- New GPT-5.6 pricing — OpenAI cuts the price of Luna by 80%, down to $0.20 and $1.20 per million tokens, and that of Terra by 20%, which now stands at $2 and $12; Sol stays the same. It comes three weeks after launch and with no application required: the cheap tier, where Chinese open models compete and where the region actually builds, is also the most volatile.
Threads we’re following
A few days ago we described how Nvidia had become at the same time supplier, creditor and shareholder of the labs that buy its chips, and how that meant the price of compute was no longer set in anything resembling a market. Today’s news is the other end of the same thread: Amazon raising its capex to 220 billion, Meta acknowledging nearly 700 billion in future commitments and an AI-focused fund liquidating its portfolio after margin calls from three global banks. Infrastructure is increasingly financed with debt and leverage, and that is exactly why a public tender with an amount and a timeline, like the European one, stands out so much: it is the version of this same business in which someone can read the terms before signing.
If the bloc that wrote the rules for AI needs letters of intent with three U.S. companies to get chips, perhaps “sovereignty” is the wrong word. Wouldn’t it be more honest to discuss what specific capacity we want, with what money and under what conditions?
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.