
Tech • AI • Robotics
Nvidia is reportedly close to buying Hugging Face for about $13 billion, as a powerful new Chinese open model and fresh US rules on teen social media underscore a broader shift in the balance of tech power.
Hugging Face, founded by French entrepreneurs and now based in the United States, is reportedly being acquired by Nvidia for roughly $12.9 billion to $13 billion. The deal has not been officially confirmed, but the reported figure would represent an extraordinary multiple for a company that is still not fully profitable. The prospect is significant because Hugging Face has become a central hub for open AI models, datasets and deployable applications.
The reported move would give Nvidia influence over a platform often described as the GitHub of AI. Hugging Face hosts about 3 million models, 1 million datasets and 1 million applications in its Spaces environment. For Nvidia, the acquisition would extend its reach beyond chips into the infrastructure layer where developers publish, share and test open tools.
The possible takeover raises questions about concentration in AI. Nvidia already dominates critical hardware for model training and inference, and it has also been releasing its own open models, including the Nemotron family. Control of chips, models and a major distribution platform could attract scrutiny from US regulators concerned about excessive market power.
The development also revives debate over Europe’s ability to retain strategic AI assets. Hugging Face emerged from the French tech ecosystem before scaling in the US, and its possible sale would reinforce concerns that Europe struggles to keep control of major digital platforms even when they are built around values such as openness and community collaboration.
Another major AI development came from China, where the previously anonymous model OX Alpha was revealed to be GLM 5.3 Flash from Z.ai. The model had been released under a codename, allowing developers to test it without knowing its origin. That blind rollout appears to have worked as a marketing strategy, because the model gained attention on performance alone before its identity was disclosed.
GLM 5.3 Flash is a 320 billion-parameter mixture-of-experts model with about 18 billion active parameters per token, a design meant to reduce computing costs while preserving strong performance. It is multimodal, accepts inputs including text and images, and supports up to 1 million tokens of context. Benchmarks place it near leading US systems in coding, reasoning and agentic tasks, while operating at a far lower cost.
The Chinese model is being distributed under the permissive MIT license, allowing companies and governments to use, modify and commercialize it freely. That has implications for AI sovereignty: states and organizations outside the US and China can deploy advanced systems locally without depending on subscription access to American providers. It also weakens the strategic advantage of domestic model champions in smaller markets.
The most politically sensitive detail is that Z.ai said the model runs entirely on Chinese AI chips. That matters because US export controls were designed in part to slow the rise of Chinese frontier AI by restricting access to advanced Nvidia hardware. A strong open model trained and served on domestic Chinese infrastructure suggests those controls are no longer enough to preserve a decisive American lead.
Outside AI, US authorities reached a $17.1 billion settlement with Meta tied to child safety reforms on Instagram and Facebook. The measures include overnight limits for teen accounts, fewer notifications during school hours and usage restrictions aimed at reducing addictive design patterns. The case stands out because it combines financial penalties with mandatory product changes rather than relying only on fines.
In robotics, concern is rising over autonomous military systems after reports that 50,000 drones with autonomy software are being supplied to Ukraine, renewing warnings from the UN and the Red Cross over lethal autonomy. At the opposite end of the spectrum, Pollen Robotics, acquired by Hugging Face, is promoting open consumer robots including the Reachy Mini and a small duck-shaped robot priced around $399. The machines are designed as open platforms for experimentation, education and AI-driven behavior rather than closed products.
The reported Nvidia-Hugging Face deal, the rise of GLM 5.3 Flash and tougher US action on platform design all point to the same trend: control over AI is shifting from individual models to the ecosystems, hardware and rules that shape how the technology is used. Europe’s place in that new order remains uncertain.
Explain this