Good morning. It's Friday, June 19, and we're covering Bernie’s bold plan to spread AI wealth, Midjourney's pivot into healthcare, and Amazon's bid to challenge NVIDIA in the AI chip race.

Plus: Friday Fact on the surprising neuroscience of London's cabbies, and a guest essay exploring the ideas behind Anthropic's AI fears.

Amazon is eyeing AI chip sales outside its own cloud business, betting demand is large enough to avoid cannibalizing AWS. The company is exploring direct sales of its Trainium chips to third-party data centers, challenging NVIDIA's dominance. AI chief Peter DeSantis cited global underconsumption, while CEO Andy Jassy previously signaled rack sales were possible. Google's pursuing a similar strategy.

Google loses one of its top AI leaders just months after bringing him back to strengthen Gemini. Noam Shazeer, Gemini co-lead and vice president of engineering, is leaving to join OpenAI after rejoining Google in 2024 through the Character.AI partnership. Shazeer co-founded Character.AI after leaving Google in 2021, and his exit highlights the escalating battle for elite AI talent.

One of AI's founding researchers argues the industry's biggest threat is economics, not the technology itself. AMI Labs founder Yann LeCun called xAI a "failure" and said it cannot keep pace with OpenAI and Anthropic after key talent departures. He warned AI firms must raise prices or cut costs as investor-funded losses mount, risking a "big bubble explosion."

U.S. regulators are reshaping the electric grid to accommodate AI's surging energy demands. The Federal Energy Regulatory Commission unanimously ordered six regional grid operators to speed connections for large power users, including AI data centers. The facilities already consume about 5% of U.S. electricity demand and could triple that share by 2035, while operators have 30 to 60 days to submit compliance plans.

FRIDAY FACTS

Memorizing a city can permanently reshape a piece of your brain — find out which one, and how scientists proved it, at the bottom of this newsletter.

AI coding experts automate everything with agents, skills, loops, and cloud workflows to build, test, review, and deploy code faster.

_{^{Guest contributor: David Shapiro is an independent researcher, futurist, and writer exploring AI, systems thinking, and the societal impacts of emerging technologies. We're excited to include his insights.}}

Famous AI Doomer Eliezer Yudkowsky first wrote about “Recursive Self Improvement” back in December 2008 on LessWrong. For those who don’t know what this means, it is the hypothetical tipping point where AI systems are able to meaningfully contribute or even take over their own training and enhancement.

In short, one generation of AIs can give rise to their successors. Dario Amodei, and Anthropic more broadly, have bought this narrative, hook, line, and sinker. → Read the full article here.

Sen. Bernie Sanders unveiled legislation on June 17, 2026, that would create a U.S. sovereign wealth fund financed by a one-time 50% tax on the stock of the country's largest AI companies. Sanders estimates the plan would establish a nearly $7 trillion fund, with investment returns funding direct payments to Americans and public programs such as health care, education, and housing.

The proposal would also give the public a substantial ownership stake and influence over corporate decision-making, reflecting Sanders' view that AI-driven wealth should be shared broadly rather than concentrated among a small group of tech firms. The plan enters a growing debate over how the economic gains from artificial intelligence should be distributed as the industry rapidly expands. → Read the full article here.

Midjourney announced a new healthcare venture centered on a full-body ultrasonic scanner designed to produce detailed 3D images in about 60 seconds. The company envisions the system operating inside wellness-focused "Midjourney Spas," with the first location planned for San Francisco in 2027, combining routine health scans with amenities like hot tubs and saunas.

Using hundreds of thousands of ultrasound transducers and AI-powered image reconstruction, the scanner aims to generate body composition maps and eventually expand its diagnostic capabilities through regulatory approval. Midjourney says it plans to scale to more cities by 2028 and deploy more than 50,000 scanners globally by 2031, reflecting a broader push to make continuous health monitoring more accessible and data driven. → Read the full article here.

Google, Microsoft, Salesforce, Snowflake, ServiceNow, Databricks, and Hugging Face have joined forces to launch Agentic Resource Discovery (ARD), a new open protocol designed to let enterprise AI applications discover and use AI capabilities across workplace software. The initiative aims to turn platforms like GitHub Copilot and Gemini into centralized gateways for business AI, reducing the cost and complexity of connecting to multiple tools.

The effort also positions established software providers against OpenAI and Anthropic, which are pursuing their own visions of AI "superagents" but are not among ARD's initial supporters. → Read the full article here. (Paywall)

Microsoft Expands AI Sales in China: ByteDance is set to spend over $1B a year on Microsoft's cloud and OpenAI services, strengthening its AI footprint.

OpenAI Unveils LifeSciBench: The 750-task benchmark tests AI on real drug discovery workflows, measuring research skills beyond simple Q&A.

Space AI Data Centers Eye Insurance: Startups are seeking coverage for orbital data centers, a key step toward financing space-based AI infrastructure.

Google Launches Agent Discovery Standard: Agentic Resource Discovery is an open standard for finding, verifying, and connecting AI tools and agents.

Deutsche Bank Speeds Up Tech With AI: The bank says AI is cutting project timelines from years to months while keeping a close watch on costs.

FRIDAY FACTS

To become a licensed London taxi driver, trainees must pass "The Knowledge," a notoriously brutal exam requiring them to memorize roughly 25,000 streets and landmarks, a process that typically takes four years.

In 2011, neuroscientist Eleanor Maguire and colleagues scanned the brains of 79 trainees before they started studying, then scanned them again years later. The trainees who passed the exam showed measurable growth in their posterior hippocampus, the brain region tied to spatial memory. Those who dropped out showed no such change.

What makes this striking is what didn't grow that way: London bus drivers, who navigate similarly complex traffic but follow fixed routes, show no hippocampal enlargement. Neither do memory champions or doctors with vast factual knowledge. It's specifically the unstructured spatial mapping of the Knowledge that reshapes the brain, one of the clearest demonstrations that adult brains can physically restructure themselves in response to a specific cognitive demand.

Thanks for reading. See you next time!

— Matthew Berman, Nick Wentz & the Forward Future Team