AI2Day Daily Brief — 25 Aug 2026

Good morning, it is Tuesday the twenty-fifth of August, and we are starting with Hugging Face, the open-source AI platform that may be about to get a very large price tag attached to it.

Hugging Face, the platform where hundreds of thousands of AI models live for anyone to download and use freely, has been approached by potential buyers at a valuation of thirteen billion dollars or more, according to TechCrunch AI. The company last raised money in 2023 at four-point-five billion, and earlier this year it turned down a five-hundred-million-dollar investment from Nvidia that would have valued it at seven billion. No deal is done and no buyer has been named. CEO Clem Delangue says the company is close to profitability and focused on long-term sustainability, which is his way of signalling the community-first mission is not going away. If you think of Hugging Face as GitHub for AI, you understand why so many parties want a piece of it.

Thirteen billion is quite the jump from four-point-five. Next, a political fight over data centres that comes down to electricity bills and who holds the planning pen.

Australian Prime Minister Anthony Albanese walked into a national cabinet meeting this week with a straightforward promise: new federal rules on where AI data centres can be built will sit alongside state planning laws, not override them. Queensland and the Northern Territory had already pushed back against federal oversight, and Albanese needed to ease that friction. The pressure is real. Australia's national electricity grid operator forecasts that data centre power consumption will grow seven times over in coming years. Albanese says he will introduce significant legislation next year to make sure the economic gains from AI spread widely rather than clustering in one or two places. States are not convinced yet.

Managing power demand at that scale while keeping state governments on board is genuinely hard policy work. Moving on: OpenAI wants software agents to run your working day, but the numbers suggest most users are not there yet.

OpenAI released ChatGPT Work last month, included in the twenty-dollar-a-month subscription. It is built on Codex, the agentic coding tool, and is designed to act on your behalf across your inbox, Slack, spreadsheets, and design tools. The catch: an OpenAI-backed study found that in June of this year, fewer than one percent of individual subscribers were using Codex at all. The desktop app has around twenty million users; the web version has more than one billion. OpenAI's own lead desktop engineer has handed the app access to his inbox, Slack, Notion, and Figma to stress-test it, and he has publicly acknowledged the risk that it might pull from a private message and not know it should not. Focused rivals like Harvey for lawyers and Clay for sales teams are already competing for the same professional users with a narrower pitch.

One percent adoption on the product your next big thing is built on is a real challenge. Now a startup betting that the next frontier for AI is not language, it is physical space.

Today's sponsor is Zero2Hero: an AI poker coach sits at your table, and you can ask it mid hand whether the player across from you is bluffing. The free tier is real: 50 hands a day against AI opponents, no card. Not a trial. Lifetime access is $99, in US dollars, until 15 September, at zero, the number two, hero, dot A I.

General Intuition is in talks to raise new funding at a pre-money valuation of six billion dollars. Pre-money means that is what investors agree the company is worth before any new cash arrives. Investors said to be joining the round include Valor Ventures, Point72 Ventures, and Seven Seven Six. What does General Intuition actually build? A foundation model designed to give AI agents a working sense of how space and time behave: left, right, near, far, before, after. The kind of spatial reasoning a toddler picks up by crawling around a room but that is genuinely hard to encode in software. That is the missing layer for useful robots and autonomous systems. If the round closes at the reported figure, the company joins a short list of AI startups valued in the billions before reaching mainstream commercial scale.

Physical reasoning as a fundable foundation model is a bet worth watching. Finally, Apple's machine learning researchers have published a method that speeds up how AI analyses video, by teaching it to think in pictures internally rather than drawing them out.

Apple ML Research has published a training framework called Internalized Visual Thinking, or IVT, aimed at video reasoning. Earlier approaches used something called Visual Chain-of-Thought, where an AI reasons step by step and actually generates intermediate images along the way, like a student sketching rough diagrams before writing an answer. Those in-between images help accuracy but slow everything down, especially across many video frames. IVT removes that step. Instead of generating images mid-analysis, the model learns to do that visual reasoning silently inside itself during training, so at deployment it outputs only text. The researchers describe it as a post-training framework, meaning it is applied after a model is already built. The goal is proactive video reasoning: the AI anticipates what happens next rather than just describing what it sees.

For more depth on all five stories and everything else moving in AI this week, catch AI Today Weekly on Monday. That is your AI briefing for today. Every story is at A-I-2-Day dot live. That is A, I, the number two, D-A-Y, dot live. We are back tomorrow morning. Do us a favour before you go: hit that thumbs up, and subscribe for next week.