AI2Day Daily Brief — 6 Sept 2026

Good morning — it is Sunday the sixth of September, twenty twenty-five, and Google just had one of its best single days in a rough summer, while Washington is quietly waking up to how much of America's AI infrastructure runs on Chinese hardware.

Google released Gemini three point eight Flash on Wednesday, its third Flash model in six weeks. The focus is coding and multi-step reasoning tasks, and Google says it outperforms its predecessor, three point seven Flash, on both. Pricing stays the same: seventy-five cents per million input tokens, three dollars seventy-five per million output. That same day, a federal judge ruled Google does not have to sell its AdX advertising exchange — its second courtroom win against a government breakup attempt in two years. Berkshire Hathaway's incoming chief executive Greg Abel also called Alphabet a winner in AI. Worth noting: nearly three-quarters of Google Cloud customers already use its AI products, and those customers are spending roughly fifty percent above their original commitments, according to Google's own figures.

Hard to overstate how much the timing mattered. Google spent the summer losing ground in the model race, watching senior researchers walk out, and reshuffling leadership at DeepMind. Its stock posted its longest monthly losing streak in over a decade. Wednesday was a different story.

Now to a supply-chain story that gets less attention than it deserves. A significant share of the hardware keeping American AI data centers running comes from China. Research from Johns Hopkins University puts Chinese suppliers at roughly thirty percent of certain transformer and switchgear categories used in US facilities. Chinese firms account for more than forty percent of US battery imports and around two-thirds of global optical transceiver supply by unit volume. That matters because US data center capacity is forecast to nearly triple by twenty thirty, from sixty-two gigawatts to a hundred and fifty-two gigawatts, according to S and P Global. President Trump has already signed an executive order declaring a national emergency over foreign-made power equipment. The problem is the market is already short — transformer and substation supply faces estimated shortfalls of fifteen and eight percent respectively in twenty twenty-six, before any new import restrictions take effect.

So the question is not just whether to decouple — it is whether the domestic supply chain can physically keep up with the build-out even without new restrictions in place.

On the defence side, the ARM Institute — that stands for Advanced Robotics for Manufacturing, a Pittsburgh-based consortium linking companies, universities and government — has been awarded nearly ninety million dollars to modernise military manufacturing sites. The money comes through the Organic Industrial Base Modernisation Challenge, run by the Office of the Undersecretary of War Manufacturing Technology Office. Fifteen member organisations will deploy robotics and AI solutions across twelve government-owned facilities — arsenals, ammunition plants, maintenance depots — within two years. Every single project is required to include a workforce training plan alongside the technology. That condition is notable: the funding explicitly treats upskilling workers as part of the deliverable, not an afterthought.

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. It runs in your browser on any computer. Windows, Mac, Linux, even a Chromebook. Nothing to install. Lifetime access is $99, in US dollars, until 15 September, at zero, the number two, hero, dot A I.

Next, a policy argument that keeps surfacing around social media regulation. Governments keep reaching for the same two levers: ban under-sixteens, or give users a switch to turn off algorithmic recommendations. Campaigners and researchers are pushing back hard on both. The Australian under-sixteens ban, which other governments have cited as a model, is widely described by researchers as unenforceable and easy to get around. And letting users opt out of recommendation, the critics argue, treats a symptom rather than the cause. The core demand is a binding digital duty of care — a legal obligation, already familiar from medicine and construction — that would hold platforms liable for the safety of the environments they build, not just the content that appears inside them. As of now, no government has passed such a law.

Finally, two more American newspapers have sued OpenAI and Microsoft. The Seattle Times and Newsday filed together, alleging their journalism was used without permission to train AI systems. Their legal complaint describes generative AI as, in their words, a snake eating its own tail: train a model on journalism, the argument runs, and the model starts replacing the need to read journalism, which kills the revenue that funds journalists, which eventually dries up the content future models depend on. The suit warns the industry could be broken beyond repair. What makes the Seattle Times case particularly striking is that Microsoft had previously funded journalism projects and fellowships at that paper. A Microsoft spokesperson said the company is surprised by the lawsuit but willing to explore a resolution. The New York Times opened this wave of copyright litigation against OpenAI and Microsoft back in twenty twenty-three.

For the full week in context, catch AI Today Weekly — the longer Monday show where we go deeper on everything that matters. That is your AI briefing for today. Every story is at A I two 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.