Snippets.

// AI Briefing

July 17, 2026

AI Briefing

Today the center of gravity in AI tilts hard toward China. Moonshot dropped the largest open-weight model ever built, Apple was forced to route its Chinese AI through Alibaba's Qwen, and 29 countries signed onto a Beijing-led answer to Western AI governance, all as Shanghai's flagship conference kicks off with Xi in the room. Meanwhile TSMC's blowout numbers show the physical bill for all this ambition, and Anthropic keeps poaching the people who might build the next leap.

Moonshot Ships Kimi K3, the Largest Open-Weight Model Ever, and It's Beating the Frontier
01Open SourceCNBC

Moonshot Ships Kimi K3, the Largest Open-Weight Model Ever, and It's Beating the Frontier

China's Moonshot AI released Kimi K3, a 2.8-trillion-parameter open Mixture-of-Experts model with a 1M-token context and a new Kimi Delta Attention design that activates just 16 of 896 experts. It ranks third on the GDPval-AA occupational benchmark, trailing only Claude Fable 5 Max and GPT-5.6 Sol Max, and takes first place in Arena's blind Frontend Code test. It matters because a freely downloadable model is now trading blows with the most expensive closed systems on Earth, with full weights due July 27.

Apple Cleared to Launch Apple Intelligence in China, but Only by Running Alibaba's Qwen
02IndustryTechCrunch

Apple Cleared to Launch Apple Intelligence in China, but Only by Running Alibaba's Qwen

China's Cyberspace Administration approved Apple Intelligence for the mainland, with Alibaba's Qwen model powering its text and image features across iOS, iPadOS, macOS and visionOS, and Baidu also confirmed as a partner. Apple couldn't ship its own models and had to bolt on a domestic Chinese system to satisfy local rules. It's a landmark proof that even the world's most valuable company must hand the AI layer to local players to operate in China, setting a template every foreign tech firm will now study.

29 Countries Sign onto China's New World AI Cooperation Organization in Shanghai

29 Countries Sign onto China's New World AI Cooperation Organization in Shanghai

On the eve of the World Artificial Intelligence Conference, 29 countries including Russia, Brazil, Indonesia, Pakistan and Kazakhstan signed an agreement to establish the World Artificial Intelligence Cooperation Organization (WAICO), an intergovernmental body headquartered in Shanghai. Chinese Foreign Minister Wang Yi signed for Beijing, with UN Secretary-General Antonio Guterres present. It matters because it stands up a China-led alternative to Western-dominated AI governance forums, formalizing a bloc of Global South and non-aligned states around Beijing's rules.

TSMC's Blowout Quarter Shows the Physical Bill for the AI Boom
04IndustryQuartz

TSMC's Blowout Quarter Shows the Physical Bill for the AI Boom

TSMC posted record Q2 2026 revenue of roughly $40 billion, up 36% year over year, with June sales alone jumping 68% and breaking a four-year seasonal decline pattern. High-performance computing tied to AI accelerators now makes up 66% of wafer revenue, and both its N3 process and CoWoS packaging are sold out through year-end. It matters because TSMC is the clearest proxy for whether AI spending is real, and the numbers say demand is outrunning the world's ability to manufacture chips.

Anthropic Pulls Monzo Co-Founder Tom Blomfield onto Its Compute Team
05IndustryPYMNTS

Anthropic Pulls Monzo Co-Founder Tom Blomfield onto Its Compute Team

Y Combinator general partner and Monzo co-founder Tom Blomfield announced he's taking leave from YC to join Anthropic's compute team under Chief Compute Officer Tom Brown, citing compute as one of the industry's most important problems as it enters early recursive self-improvement. The move caps an aggressive 2026 recruiting run that already brought in Andrej Karpathy, Nobel laureate John Jumper and Berkeley's Jelani Nelson. It matters because the AI talent war is now being fought over the people who can scale compute, not just model research.

Test Your Understanding

Quiz

1 / 8

What makes Kimi K3 notable among AI models released to date?

Let's talk on WhatsApp
Today's AI Briefing5 stories
Aug 29, 2026

Summary

A federal judge just told the Pentagon it cannot punish an AI lab for refusing to hand over its models, which is the first time a court has drawn a hard line around what safety policies cost you in Washington. Elsewhere the theme is AI learning by watching rather than being told: a robot foundation model that copies a ten-minute task from one video, and Claude grinding through algebra that stumped a lab for eighteen months. Plus Meta patching the dumbest privacy hole in its glasses, and Pew's number on how many Americans now ask a chatbot about their symptoms.

Read full summary & take a quiz →

Top Stories

A Judge Just Ruled the Pentagon Illegally Punished Anthropic for Saying No to Mass Surveillance

Skild's Robot Model Watches One Video of You Making Pancakes and Then Makes Pancakes

Claude Did Eighteen Months of Algebra in Five Weeks, and Got Some of It Wrong Along the Way

Meta Is Finally Killing the Trick Where You Start Recording Then Cover the Light

A Third of Americans Now Ask a Chatbot About Their Symptoms, and Most Won't Share Their Health Data With It

9 quiz questions inside