Good morning. It's Friday, August 28, and we're covering Meta’s surprising reliance on a rival AI lab, a world-first AI-assisted brain surgery, and NVIDIA’s growing AI ecosystem.

Plus: yesterday's AI risk poll results, and a Friday Fact about a strange number.

NVIDIA says its investments in AI customers create multiple paths to growth rather than “circular financing.” CFO Colette Kress argues the company is supporting a once-in-a-generation computing shift while expanding its customer base. AI labs receiving NVIDIA financial support could account for about 25% of its business next year, raising questions about the ecosystem’s interconnected risks.

Meta’s AI tools produced 220% more code changes but only a 36% increase in features reaching users. The company explored cutting some teams by up to 60% under Project OT before employee resistance and technical problems forced a retreat. Major technical and security incidents reportedly rose 40%, while employee response time increased 70%.

NVIDIA is optimizing its hardware for Chinese models even as Washington considers restricting support for them. The company highlighted DeepSeek V4 Flash and Alibaba’s Qwen 3.8 as part of a local AI initiative. An SEC filing warns that limits on supporting Chinese-origin models could materially affect NVIDIA’s business amid rising U.S. concerns over their adoption.

A mathematician needed an upper bound for a geometry problem — and ended up creating a number too large for the universe itself to hold. Get the answer at the bottom.

Most brands have no idea whether AI search engines recommend them or a competitor. Trendos.com tracks your brand across ChatGPT, Perplexity, Gemini, Google AI Overviews, and Claude with real conversational prompts, not keyword proxies. Free to start.

See if AI recommends your brand. Try Trendos.com free.

Meta has become one of Anthropic’s largest customers despite Mark Zuckerberg’s public criticism of leading AI labs, with the company at one point projecting it could spend as much as $10 billion annually on Anthropic’s models. Meta has used Anthropic’s tools, including Claude Code, to support internal development and test its forthcoming AI product Hatch, though it has since reduced some spending and is increasingly using its own tools.

The relationship underscores the AI industry’s unusual mix of competition and commercial dependence, as companies build rival models while also buying technology from one another. Meta is developing its own advanced model, internally called Watermelon, but its release has reportedly been delayed until at least October 2026. → Read the full article here. (Paywall)

Surgeons at University College London Hospital have removed a patient’s brain tumor using an artificial intelligence tool that analysed the operation’s live video feed in real time, in what the team describes as a world-first procedure. The system helped surgeons identify hidden blood vessels and nerves while removing a non-cancerous pituitary tumor from Rhys Hibbert, a 48-year-old father of two whose worsening vision had put him at risk of blindness.

Researchers trained the tool on hundreds of videos of similar operations, but surgeons retained full control throughout the procedure. The team is now preparing a larger trial, testing whether real-time AI can become a reliable second set of eyes in the operating room. → Read the full article here.

CrowdStrike and Okta shares jumped 20% and nearly 29%, respectively, on August 27, 2026, after both cybersecurity companies beat Wall Street expectations and raised their forecasts. Executives said wider adoption of AI is increasing cyber risks and prompting businesses to spend more on security.

→ Continue reading here.

Young Workers Turn to Traditional Crafts: Graduates are pursuing hands-on trades like bookbinding and metalwork that resist automation.

Alphabet Sheds $692 Billion: Shares fell 15% from their May peak as investors question Google's AI edge, spending and delayed Gemini release.

Salesforce and Anthropic Launch Claudeforce: Claude gains 37 Salesforce sales skills for live data, pipeline updates and governed actions.

Anthropic Plans Flexible IPO: Reports say Anthropic may let existing shareholders sell in its IPO while using longer lockups to limit volatility.

Z.ai Rallies on Chinese-Chip Model: Shares jumped 8% after Z.ai launched an AI model it says runs entirely on Chinese chips.

In 1971, Ronald Graham and Bruce Rothschild studied a Ramsey-theory problem involving red-and-blue connections between the corners of a high-dimensional cube. They proved that, in a sufficiently high dimension, four corners must form a monochromatic flat square.

Graham later gave the problem an enormous upper bound, popularized by Martin Gardner as Graham’s number. It appeared in the 1980 Guinness Book of World Records as the largest number used in a serious mathematical proof.

The true answer is still unknown: it is at least 13 but far smaller than Graham’s number. Graham’s number is finite and precisely defined, but writing out its digits—even one per Planck-sized volume—would require more space than the observable universe.

Here's how you voted: AI-enabled bioterrorism topped the list of concerns, while mass job losses and industry secrecy followed closely behind.

Thanks for reading. See you next time!

— Matthew Berman, Nick Wentz & the Forward Future Team