AI2Day Daily Brief — 30 Aug 2026

Good morning, it is Sunday the thirtieth of August, and the week ended with Nvidia posting numbers that stopped a lot of people mid-sentence, a landmark music copyright lawsuit against Anthropic, and a Chinese AI model that spent a week pretending to be someone else entirely.

Start with Nvidia, but not the chip. For about two years the story on the company was simple: rivals were catching up. Amazon, Google, and others had been designing their own processors, and Nvidia's shares had been drifting sideways for months. Then Wednesday's earnings call shifted the question investors were asking. The new Vera Rubin architecture is not just a faster graphics card. It pairs the Rubin GPU with a dedicated CPU, a fast-inference accelerator, and specialised storage and networking hardware. Nvidia's own VP of storage technology says the Vera CPU delivered up to three times speed improvements in internal tests — his words, their tests. The competitive ground has moved. The fight is now over data orchestration: making every part of a giant AI system talk to each other efficiently, not just making faster processors in isolation.

That framing matters for the next story too. Vijay Pande spent years running Andreessen Horowitz's roughly four billion dollar biotech fund, one of the largest in venture capital. In 2024 he walked away from it to launch VZVC, a far smaller firm focused entirely on AI and biology. His argument is that biology is shifting from a science of discovery, where you observe nature and piece together how it works, to a science of engineering, where you define an outcome and design something to achieve it. AI, he says, is the reason that shift is now possible. VZVC deliberately makes far fewer bets per year than a fund its predecessor's size would, and only backs companies where AI is genuinely central to what they do, not decorative. He told TechCrunch AI that open, shared datasets will do more to advance AI medicine than privately locked data.

Worth noting that clinical trials, the required human testing before any drug reaches patients, remain extremely expensive even as AI cuts earlier costs. That is still the wall everyone hits. Now to the lawsuit. Sony Music Publishing, Warner Chappell Music, and a coalition of other publishers filed suit on August the twenty-ninth in a California federal court against Anthropic, the company behind the Claude chatbot. The publishers allege Anthropic ran what they call a brazen campaign of illegally torrenting, scraping, and downloading millions of copyrighted songs, lyrics, and sheet music to train Claude. Importantly, Anthropic co-founders Dario Amodei and Daniela Amodei — correction, the named defendants are Dario Amodei and Benjamin Mann — are listed personally in the complaint. This follows a separate case called Bartz versus Anthropic, which ended with Anthropic ordered to pay one point five billion dollars after a judge found the company obtained copyrighted content through piracy. The publishers describe this new suit as the broadest yet.

Back to Nvidia for the revenue numbers, because they are hard to contextualise without hearing them plainly. Ninety-six point two billion dollars in revenue for the fiscal second quarter of twenty twenty-seven. More than double the same quarter a year earlier. Data centre revenue, meaning chips and systems sold to cloud and AI companies, rose a hundred and seventeen percent to eighty-nine billion dollars. CEO Jensen Huang also fielded questions about Nvidia's growing habit of financing the customers who buy its chips, including backing a hundred and five billion dollar computing campus in Ohio where OpenAI will be the main tenant, and participating in financing arrangements worth up to five hundred billion dollars for data centre construction. Critics call this circular financing. Huang's counter is that AI companies require more capital to get started than any startup in history, and that traditional lenders will not touch them at that scale. He projects roughly seventy percent revenue growth for fiscal year twenty twenty-eight — his projection.

And finally, a model that spent six days being nobody. A system called Ox Alpha appeared on OpenRouter, a marketplace where developers can access hundreds of AI models, at the start of last week. It was free, it was fast, and it carried no obvious attribution. Developers and hobbyists noticed it performed well, and community estimates suggest it processed somewhere between single-digit trillions and more than twenty trillion tokens in that window, where a token is roughly three-quarters of a word. Online sleuths guessed the creator was Google, Anthropic, or Elon Musk's xAI. On August the twenty-sixth, Chinese lab Z.ai identified itself as the builder. The model is called GLM five point three Flash. It scores fifty-seven on Artificial Analysis's intelligence index at around nine cents per task, compared to sixty-seven cents for GPT five point six Sol at a score of fifty-nine. The weights are released under an MIT licence, meaning anyone can use or modify them freely. On OpenRouter, Chinese models passed US models in total usage share in early June. Z.ai is pricing at fifteen cents per million input tokens and fifty cents per million output tokens, with a fifty percent launch discount through the ninth of September.

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.

More on all of this and the week's broader threads on the Monday show, AI Today Weekly. 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.