This Week in AI: July 25 – 31, 2026

This week the AI story shifted from what the models can do to whether the money behind them adds up: a historic chip-stock selloff rattled investors even as Google shipped a genuine robotics breakthrough, DeepSeek kept building at Chinese scale, and over a thousand AI workers publicly asked governments to be ready to slow things down.

The Big Story: AI Chip Stocks Lose Over $1 Trillion

The 20 most valuable chip stocks lost more than $1.3 trillion in combined market value in the days following July 24's close, according to CNBC and Tech Startups'' reporting. Nvidia alone shed $238 billion, with SK Hynix, Samsung Electronics, and Micron losing $176 billion, $173 billion, and $113 billion respectively, and TSMC down $119 billion. According to CNBC, the selloff reflects growing investor doubt over whether the trillions being poured into AI infrastructure will generate returns that justify the spending, rather than any drop in actual chip demand.

Source

• Chip stocks shed more than $1 trillion as selloff hits companies powering AI boom, CNBC
• AI Chip Stocks Lose $1.3 Trillion as Nvidia, TSMC, Samsung Hit by Fears Over AI Infrastructure Returns, Tech Startups

Gemini Robotics 2: Whole-Body Control for Humanoid Robots

Google DeepMind unveiled Gemini Robotics 2 on July 30, a suite of models that lets humanoid robots coordinate their entire body, not just the upper half, according to Bloomberg and MarkTechPost''s reporting. Where the prior generation only controlled a robot''s arms and hands, the new system lets a machine plan and adjust movement from the torso down to the legs, reasoning through multi-step tasks as it walks, crouches, and manipulates objects. Results varied significantly by task in DeepMind''s own reported testing: a 92% success rate unscrewing a light bulb, but far lower rates on tasks requiring finer sequential coordination, 44% tying a trash bag and 40% sealing a ziplock bag.

Source

• Gemini Robotics 2 Expands Google's AI Capabilities for Humanoid Robots, Bloomberg
• Google DeepMind Ships Three Physical AI Models For Whole Body Control, Dexterity And Multi Robot Collaboration, MarkTechPost

DeepSeek Plans Gigawatt-Scale Data Center in Inner Mongolia

DeepSeek is developing a large AI data center in Ulanqab, Inner Mongolia, aiming to add roughly a gigawatt of compute, according to Bloomberg's July 30 reporting. The company plans to build part of the facility itself while leasing additional capacity elsewhere, targeting partial capacity online by late next year or early 2028, and is choosing the region partly for its low average temperatures to cut cooling costs. The buildout continues a pattern of DeepSeek investing in its own infrastructure at a scale that rivals Silicon Valley labs.

Source

• DeepSeek Plans Gigawatt-Scale AI Data Center in Inner Mongolia, Bloomberg

Voices of Caution: Over 1,100 AI Workers Ask for a Pacing Mechanism

More than 1,100 employees across OpenAI, Anthropic, Google DeepMind, and Meta AI signed an open letter on July 28 asking the US government to help build the technical and governance infrastructure for a coordinated slowdown of AI development, should systems ever advance faster than humans can safely oversee, according to Bloomberg and CNN's reporting. Signatories reportedly include senior researchers at multiple labs, and OpenAI and Anthropic have since publicly endorsed the statement on behalf of their companies.

Source

• More Than 1,100 AI Workers Call for US to Pace Tech Growth, Bloomberg
• Employees from the world's biggest AI companies want the US to be ready to slow AI development, CNN Business

Cognizant Becomes Anthropic's Top-Tier Claude Partner

Cognizant and Anthropic expanded their partnership on July 27, with Cognizant becoming a Global Premier Partner in the Claude Partner Network, according to Anthropic's own announcement. More than 30,000 Cognizant employees, out of a global workforce of 350,000, have completed Claude training, and the company is embedding Claude further into the industry platforms it builds and runs for manufacturing, life sciences, insurance, and other clients.

Source

• Expanding our partnership with Cognizant, Anthropic
• Cognizant Expands Claude Partnership; More Than 30,000 Trained, StockTitan

Meta's Earnings Call: A Shaky Forecast and a Bold Prediction

Meta gave investors a weaker-than-expected third-quarter revenue forecast on its July 29 earnings call, adding pressure on the company to show its massive AI spending is paying off, according to Bloomberg's reporting. On the same call, CEO Mark Zuckerberg predicted that within five years, billions of people will each have a personal AI agent handling tasks across their finances, health, relationships, and home life, per TechCrunch's coverage. He pointed to Meta's AI business agents on WhatsApp and Messenger, already used by more than a million businesses every week, as an early version of that shift.

Source

• Meta Falls After Defending AI Bets to Skeptical Investors, Bloomberg
• Mark Zuckerberg predicts that billions of people will have personal AI agents in five years, TechCrunch

Strategic Takeaway

The week's real tension is between infrastructure conviction and infrastructure doubt: DeepSeek and Google kept building and shipping at full speed, Cognizant deepened its bet on enterprise AI, and Zuckerberg doubled down on a future full of personal agents, while chip investors and over a thousand AI workers themselves both, in very different ways, asked whether the pace and payoff of this buildout actually holds up under scrutiny.