Daily brief, November 5: data centers, Anthropic, ChatGPT GIFs

AI data centers made headlines at the American ballot box, tech giants keep spending wild amounts of money, and ChatGPT can now animate your images.

In the US, voters say no to data centers

On November 3, during the midterm elections, 41 local measures on data centers were put to a vote in 10 states. In 40 cases, voters chose to limit projects or reject measures that favored them. The only exception: Harwood, North Dakota. Two days later, OpenAI promised to stop signing nondisclosure agreements with local officials, to publish its construction plans and to pay for all of the electricity its sites use. Why it matters: without power and land, AI can't grow, and local residents are starting to have a say. Source: CNBC

Amazon, Alphabet, Microsoft and Meta: $207 billion in one quarter

In the third quarter of 2026, these four giants invested about $207 billion, mostly in AI infrastructure: Amazon $75.9 billion, Alphabet $51.4 billion, Microsoft $45.2 billion and Meta $34.9 billion. Heads up: this total adds up different spending programs. It's not an official "AI bill," since each company counts in its own way. Why it matters: this money ends up in the chips, servers and buildings mentioned above, and it explains why electricity is becoming a political issue. Source: Le Monde Informatique

Anthropic reveals its numbers ahead of its IPO

In a filing updated on November 2 with the US securities regulator, Anthropic reports nearly $17 billion in revenue for the third quarter, compared with $1.4 billion a year earlier. But the company posted an operating loss of $567 million: computing power is very expensive. The net loss of $23.3 billion comes mainly from the revaluation of financial instruments. The size of the offering and the valuation are circulating in the press, but they aren't confirmed. Why it matters: you finally get to see what cutting-edge AI really costs. We explain the news, we don't give advice. Source: Le Monde Informatique

Fal doubles its valuation to $16 billion

Fal, an American startup that runs video, image and audio AI models for other companies, announced on November 3 that it had raised $1.3 billion. Its valuation went from $8 billion to $16 billion in a few months, according to Bloomberg, which also mentions more than $800 million in annualized revenue. Fal doesn't build models: it hosts them, which is called inference. Why it matters: some of the money in AI is made by running everyone else's models, not just by creating them. Source: TechCrunch

ChatGPT can now make GIFs

With its new image model, announced on November 4, OpenAI lets you create a GIF from a description or a photo. You can adjust the speed and edit each frame. According to OpenAI, the model is also about 50% faster than the previous one. GIFs are limited to paid ChatGPT and Codex plans. Why it matters: making an animated meme no longer requires editing software, but there's no free version for now. Source: Blog du Modérateur

Iron powder to store summer energy

Experts at the University of Cologne propose storing solar and wind power in iron powder. It's burned without CO₂ in former coal plants, then "recharged" with green hydrogen. In their model, 37 GW of German power plants could be converted by 2045. Heads up: this is a simulation, and no pilot facility is mentioned. Why it matters: winter is solar power's real challenge, and this is one more idea for solving it, though it's far from ready. Source: Tech Xplore