Good morning. It's Tuesday, July 21, and we're covering Google's AI search squeeze on publishers, China's Kimi K3 fueling the open-source AI race, and how AI is reshaping China's booming microdrama industry.

Plus: a poll on the right China AI strategy and yesterday's zeitgeist on AI getting political, personal, and weird.

The Trump administration is reviving efforts to restrict advanced Chinese AI models as their growing competitiveness reshapes the market. Officials are reportedly considering measures ranging from Entity List designations to procurement rules and security requirements that could discourage U.S. adoption. The renewed push follows the rise of Moonshot AI's Kimi model, while critics, including David Sacks, warn such policies could strengthen OpenAI and Anthropic by limiting open-source competition.

The Information reports Google is developing "Frozen v2," a next-generation server chip targeted for 2028 that could deliver 6-10x more tokens per unit of power than its current AI chips when running Gemini. Google declined to confirm or deny the report but pointed to its "full stack approach" of co-designing hardware and software, and Alphabet stock jumped ~3% on the news heading into this week's earnings, easing investor unease about the $180-$190B Alphabet has committed to AI capex this year.

Microsoft is expanding its reliance on AMD across GPUs, CPUs and networking as demand for AI inference infrastructure accelerates. Azure will deploy AMD's Helios rackscale platform, powered by Instinct MI455X GPUs and EPYC "Venice" processors, to support frontier AI models, Azure AI services and customer workloads. Azure is also launching two new AMD-powered VM series and expanding Pensando DPU deployments, with planned Helios shipments.

Artificial intelligence is taking over tasks once handled by actors, writers and editors, cutting production costs across China's booming microdrama market. Producers increasingly use AI to write scripts, generate performances and speed postproduction as the industry expands internationally. China's microdrama sector is projected to reach about $14 billion in 2026, with platforms like Douyin driving rapid growth. Directors say AI is transforming nearly every stage of production while lowering barriers for new creators.

Moonshot AI’s open-source Kimi K3 challenges GPT-5.6 and Fable 5 with top coding benchmarks and fresh momentum in the U.S.-China AI race.

This week, the conversation kept returning to the same idea: AI becoming a political, economic, and cultural battleground. From open-source tensions and pricing myths to AI companions and internet celebrities, the debate shifted from what models can do to who controls them, how people use them, and what happens next.

▸ Why Chinese open-source AI is becoming a geopolitical flashpoint
▸ Why "cost per completed task" matters more than token pricing
▸ How AI companions, search, and streamers are reshaping the internet

Read yesterday's Zeitgeist →

A July 20, 2026, New York Times report examines how Google's AI-powered search features, including AI Mode and Gemini-generated responses, are changing how people use the web by keeping users inside Google's ecosystem rather than directing them to external websites. Studies cited in the report found users spend one to nine more minutes in AI Mode, and roughly 75% of AI Mode sessions end without a visit to another website.

Publishers, including Vox Media and Wikimedia, say declining search traffic is hurting their businesses, while Cloudflare estimates human visits to finance, publishing, and retail websites fell nearly 40% between June 2025 and April 2026. Google disputes claims that it is harming the open web, saying its AI search features still send billions of clicks to websites each week and has introduced new tools to highlight sources and give publishers more control over how their content appears. → Read the full article here. (Paywall)

Moonshot’s release of Kimi K3, a 2.8-trillion-parameter open-source AI model, has fueled debate over whether Chinese AI developers are closing the gap with leading U.S. models. The model topped Arena’s coding benchmark ahead of OpenAI’s GPT-5.6 and Anthropic’s Claude Fable 5, prompting praise from researchers and interest from companies including Microsoft, which is evaluating it for Azure and Copilot.

However, experts caution that benchmark performance does not always translate to real-world capability, noting Kimi K3’s high infrastructure requirements, greater token consumption, and estimated performance that still trails Anthropic’s latest models in underlying pre-training. The release underscores the growing competitiveness of China’s AI ecosystem while highlighting that the race increasingly depends on efficiency, deployment costs, and practical performance—not benchmark scores alone. → Read the full article here. (Paywall)

Washington can't decide what to do about Chinese open-source AI models.

Chinese robotics company Ubtech has unveiled the Uworld U1 series, which it describes as the world's first mass-produced, full-size humanoid companion robot. Available in multiple models priced from 119,800 RMB (about $17,700) to 990,000 RMB (about $145,700), the robots feature 88 servo joints, silicone skin, AI-powered conversation, and the ability to display 300 facial micro-expressions.

While Ubtech initially promoted the robots as capable of household tasks and caregiving, the commercial launch focused primarily on companionship after demonstrations highlighted slow response times and limited interactions. Ubtech says it received more than 13,000 orders on launch day, although many came from demonstration centers and government procurement programs rather than individual consumers. → Read the full article here.

AI Testing Chief Resigns: Chris Fall stepped down after three months leading the U.S. AI testing institute as Commerce searches for a new director.

Meta Shifts AI Priorities: Meta may reduce its $174B investment portfolio to help fund expanding artificial intelligence initiatives.

Infinity Raises $15M: AI startup Infinity raised $15M to build software that helps AI models run across different chips, challenging NVIDIA's ecosystem.

Adobe Adds AI Photo Critique: Adobe's Project Indigo app now uses AI to critique photos, suggest edits, and remove unwanted objects with one tap.

Spotify Purges AI Music Spam: Spotify removed 75 million AI-generated tracks in 2025 as it steps up efforts to curb spam and low-quality content.

Thanks for reading. See you next time!

— Matthew Berman, Nick Wentz & the Forward Future Team