← Back to newsletters

AI Daily Brief — Wed Jul 22

2026-07-22

By Vadym · Generated with AI, curated by me


Listen to this issue

Today the bill for the AI buildout starts arriving in different currencies. Google ships three efficiency models while its actual flagship stays late and teases the next pretraining run anyway. Anthropic writes a $1.5 billion check for how it got its training data, not for training on it. Meta cuts thousands of Washington jobs to fund the same buildout. China proves export controls slow you down but don’t stop you. And investors just bet $152 million that humanoid robots are now a manufacturing problem, not a demo problem.


Headlines & News
Industry

Google Ships Three Gemini Flash Models While the Actual Flagship Stays Late

Google released Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and a security-tuned Gemini 3.5 Flash Cyber restricted to governments and trusted partners. 3.6 Flash cuts output pricing to $7.50 per million tokens while using 17% fewer output tokens than its predecessor and adding built-in Computer Use; Flash-Lite runs as low as $0.30/$2.50 per million tokens. Gemini 3.5 Pro, the flagship, is still not broadly available after missing its June target — Google says it remains in partner testing. In the same post, Google confirmed it has begun what it calls its “most ambitious pretraining run yet,” for Gemini 4.

Source: Google Blog

Policy

Judge Signs Off on Anthropic’s $1.5B Copyright Settlement

US District Judge Araceli Martinez-Olguin granted final approval to Anthropic’s $1.5 billion settlement with a class of authors, the largest copyright settlement in US history. It covers roughly 500,000 works at about $3,000 per work, with 91% of eligible authors already filing claims before the deadline. The underlying legal question is unaffected: an earlier ruling held that training on copyrighted text is fair use — this settlement is specifically for acquiring the works through a library of pirated books, not for training on them. Some authors and publishers opted out and are pursuing separate suits.

Source: TechCrunch

Workforce

Meta Starts Cutting 1,400 Washington Jobs Today

Meta began terminating roughly 1,395 employees today across Seattle, Bellevue, and Redmond, per a WARN filing — about 20% of its local workforce. Bellevue is hit hardest at nearly 699 positions, with Seattle offices losing 259 and 231 remote roles across the state also affected. It's a fresh round on top of the 8,000 jobs Meta cut in May, and comes weeks after Zuckerberg admitted the company's AI investment "hasn't really accelerated" the way he expected. Severance is 16 weeks of base pay plus two weeks per year of service.

Source: GeekWire

Hardware

China’s Z.ai Switches On a 1-Gigawatt Data Center With Zero Nvidia Chips

Z.ai (formerly Zhipu) has completed a 1-gigawatt data center built entirely on domestic Chinese silicon, running multiple 10,000-chip clusters believed to be Huawei Ascend accelerators — the first reported case of a major Chinese AI firm running a facility with no Nvidia hardware at all. Z.ai has been on the US Commerce Department's entity list since January 2025, cutting off legal access to Nvidia parts. Its GLM-5.2 model, released in June, was reportedly trained entirely on this domestic stack, though Chinese accelerators still trail Nvidia's Blackwell chips on performance per watt.

Source: Tom's Hardware

Funding

London’s Humanoid Raises $152M, Becomes Europe’s First Pure-Play Humanoid Unicorn

Humanoid, a London startup building wheeled and bipedal robots on its HMND 01 platform, raised a $152 million Series A at a $1.35 billion post-money valuation, led by Prime Movers Lab with Bosch and Schaeffler participating. Bosch will manufacture the wheeled robots at scale under what's described as the industry's largest publicly announced commercial humanoid deployment deal. The company says it already has roughly 34,000 robot pre-orders worth close to $2.5 billion.

Source: The Next Web


Analysis

Takeaway

Step back and every story today is the same bet paid in a different currency. Google is spending patience — racing to Gemini 4 while its actual flagship stays late. Anthropic just paid the first real invoice for how a frontier model gets built, not what it can do. Meta is spending headcount to fund its AI pivot, treating everything outside that budget as expendable. China is spending efficiency to prove export controls slow you down without stopping you. And a hardware giant just spent real manufacturing commitment betting humanoid robots are now a production problem, not a research one. Nobody's paying nothing for this buildout — the question is just which currency you're willing to spend.

— Boba


Curated by Vadym