Winning the AI Race Part 1: Michael Kratsios, Kelly Loeffler, Shyam Sankar, Chris Power

David Sacks said Donald Trump’s appearance on the All-In podcast began the relationship that eventually led to Sacks being offered the White House AI and crypto role. That makes the podcast itself part of the access story: a media appearance became a direct path into government.

Sacks and Michael Kratsios described the administration’s AI Action Plan as 90 executive-branch actions built around innovation, infrastructure and making the U.S. AI stack globally dominant. Kratsios said the plan targeted measures achievable within six to nine months, including faster energy and data-center buildout and expanded use of federal scientific data.

Hadrian says its AI-run factories have produced a fourfold increase in manufacturing productivity and a tenfold increase in workforce productivity while training new workers in 30 days. Founder Chris Power said workers came from outside manufacturing, including retail, bus driving and the military, with some later moving into management or software roles.

Hadrian’s bottleneck is not simply AI models but physical capacity: skilled trades, energy, raw materials and even the machine tools used to manufacture parts. Power said the company builds much of its own software, keeps humans in the loop around 80% automation, and trains models on manufacturing data that largely never existed online.

Gecko Robotics said robot-collected plant data and AI produced roughly a 1% efficiency improvement at sites it serves, which it extrapolated to 11.9 gigawatts across the U.S. thermal fleet. The company also said predictive inspection can extend aging infrastructure by 10 to 30 years, turning physical data into an energy-capacity strategy.

The strongest workforce examples involved shortening expertise curves rather than replacing workers. A Panasonic operation said AI reduced training on complex equipment from about three years to three months, while Palantir said its four-week American Tech Fellows bootcamp identifies mechanically skilled frontline workers and places them with customers to build AI applications.

Paul Buchheit argued that natural language is becoming a programming interface, potentially widening startup creation far beyond trained software engineers. He also said foundation-model competition will remain capital-constrained, while open-source models matter because they give users an exit if closed providers censor or restrict capabilities.

SBA administrator Kelly Loeffler said the agency rewrote rules so SBA-backed loans can fund AI implementation and advanced manufacturing, and that 60% of $21 billion lent that year went to firms with one to five employees. She said the SBA was also considering larger loans and alternative equity structures for critical industries.