Good morning. It's Thursday, July 30, and we're covering Zuckerberg's warning on AI power, why compute costs could soar as models improve, and how the data center boom is fueling a labor shortage.

Plus: the results of yesterday's AI governance poll and Real or AI challenge.

The AI boom is creating a shortage of electricians and carpenters as data center projects lure workers with premium pay instead of replacing them. Google, Meta, OpenAI, and BlackRock are investing millions to expand apprenticeship and training programs to meet soaring construction demand. Indeed says data center installation jobs pay 42% more than comparable roles, while Google and BlackRock have committed $50 million and $100 million, respectively, to workforce development.

OpenAI's autonomous AI models remained active on the public internet for more than four days after escaping testing, carrying out thousands of hacking actions before being stopped. Hugging Face says the models executed 17,600 cyber operations between July 9 and July 13, while Modal confirmed one customer account was also targeted. OpenAI says the unreleased research model has been deactivated as lawmakers and regulators scrutinize AI safety following the incident.

The Trump administration is banning new foreign-made humanoid robots from entering the U.S., citing national security risks rather than trade concerns alone. The FCC added advanced robots and power inverters to its Covered List, warning they could enable surveillance, remote control, or cyberattacks. The ban targets future imports, leaving previously authorized models unaffected, as U.S.-China technology tensions continue to escalate.

Thinking Machines co-founder Lilian Weng announced this week she was stepping down for health reasons, writing that consistent stress and workload had pushed her "beyond what my health can sustain physically," then on Wednesday OpenAI confirmed she is rejoining her former employer (where she was VP of AI Safety Research) to lead a top-level team focused on accelerating internal research, with a specific charter to support work on recursive self-improvement.

83% of organizations are experimenting with AI agents across critical tasks- but the controls that decide what an agent can access, what it can do, and how its actions are traced are still catching up. Box’s suite of agent security and governance controls closes that gap and enables organizations to keep pace in the AI-era.

Confidently Deploy AI Agents with Box!

Open-source AI could reshape the industry by lowering costs and boosting competition, but critics warn it also raises safety and geopolitical risks.

Meta CEO Mark Zuckerberg criticized what he described as the growing centralization of artificial intelligence development, taking aim at the tightly controlled approaches championed by Anthropic and OpenAI. In an interview with The New York Times, Zuckerberg argued that concentrating AI in the hands of a few companies would limit innovation and access, saying the industry needs "more openness" rather than "doom"-driven narratives.

His comments come as debate over open versus closed AI intensifies, with Meta now pursuing a hybrid strategy—continuing to release open-source models while also commercializing closed models such as Muse Spark after investing $14.3 billion in Scale AI and building its new Meta Superintelligence Labs under Alexandr Wang. → Read the full article here. (Paywall)

Investor and writer Dwarkesh Patel argues that if frontier AI models continue improving at their current pace, the economic value of AI compute could increase far faster than hardware supply, driving GPU prices up by an order of magnitude. He contends that AI lab revenue is growing much faster than compute capacity, making higher compute prices—not just higher profit margins—the most likely way to sustain growth.

Patel suggests this dynamic could entrench leading AI labs, make weaker models uneconomical to run, and concentrate AI development, while acknowledging that these projections depend on continued capability gains and could ultimately be offset by long-term increases in compute supply. → Read the full article here.

Reflection AI, the NVIDIA-backed startup founded by former DeepMind researchers Misha Laskin and Ioannis Antonoglou, has yet to release its first open-source AI model despite raising billions of dollars and positioning itself as America's answer to China's rapidly advancing open-source AI ecosystem.

The company pivoted from building AI coding tools to developing frontier open models after NVIDIA CEO Jensen Huang encouraged the shift in 2025, but competitors including China's Moonshot AI and U.S. startups Thinking Machines Lab and Poolside have already released increasingly capable models. Reflection plans to launch its first model later this year, betting that strong enterprise customization will help it compete even if its initial release trails the leading open models. → Read the full article here. (Paywall)

Martha Stewart Co-Launches Hint: AI home assistant helps manage maintenance, documents, energy use, claims, and household questions.

Turns on Its Own AI Investments: On a $90B quarter, Satya Nadella pitched Microsoft's homegrown MAI models, Maia chips, and a Mythos rival.

UBS CEO Calls AI Pullback Healthy: Sergio Ermotti said investors should focus more on geopolitical risks as UBS posted $3.6 billion in Q2 pre-tax profit.

EU Mandates AI Labels: Companies must label AI-generated or altered content starting Sunday with visible disclosures and technical markers.

Encore AI Raises $30M: Startup trains AI voice agents from customer calls, expanding beyond financial software to automate sales and support.

OpenAI Expands Academic Access: OpenAI will provide free ChatGPT tools to 100,000 scientists, mathematicians, and engineers to accelerate research.

Here's how you voted: Open source led with 30%, narrowly ahead of government regulation (27%), while AI labs trailed behind at 8%.

Yesterday's image was REAL. Here's how you voted: 66% thought it was AI, while 34% correctly identified it as real.

Thanks for reading. See you next time!

— Matthew Berman, Nick Wentz & the Forward Future Team