David Ulevitch on American Dynamism: AI will compress nuclear regulatory timelines and public safety is the most underrated opportunity
May 14, 2025 with David Ulevitch
Key Points
- AI language models tuned to NRC regulations can compress nuclear reactor approval timelines from months to minutes by instantly mapping design changes against hundreds of thousands of regulatory pages.
- Defense procurement is opening to early-stage startups because the DoD faces real capability gaps that only AI-native founders can address faster than traditional primes.
- Public safety remains venture's most underrated category, with gaps in 911 dispatch automation and air traffic control infrastructure that extend well beyond Andreessen Horowitz's current portfolio.
Summary
David Ulevitch, general partner at Andreessen Horowitz on the American Dynamism team, makes the case that the current moment is genuinely different for defense and national-interest investing — and that public safety and nuclear energy are the two most underappreciated opportunities in the portfolio.
Defense procurement is changing
Ulevitch argues the DoD knows it has a problem: it isn't procuring fast enough, isn't manufacturing fast enough, and is attriting systems faster than expected. The enthusiasm he hears from generals and civilian leaders isn't performative — it's driven by real capability gaps. The pushback will come from the primes, who have "had a delicious steak lunch every day for the last 25 to 30 years," but Ulevitch thinks this is the moment early-stage startups break through.
On why venture leads rather than private equity: technology transformation cycles favor startups because only they can attract the talent willing to move fast on AI and computer vision. PE firms manage to the bottom line and won't pay for 10x engineers. That said, Ulevitch expects PE to follow quickly as sub-primes get rolled up to meet manufacturing demand.
Public safety is the most underrated category
Ulevitch claims to be the largest public safety venture investor in the country, if not the world. His portfolio includes Flock Safety, but he's explicit that Flock Safety's roadmap doesn't cover everything. 911 services are one gap — high-turnover, exhausting work that he argues needs to be reimagined with AI operators triaging calls so human dispatchers and first responders get better information faster. Air traffic control is another: the Newark FAA breakdown is an obvious opening for a venture-backed startup to replace 30-year-old radar infrastructure. Neither Andreessen Horowitz nor Flock Safety will build those, so the surface area keeps expanding.
AI compresses nuclear regulatory timelines
Ulevitch is invested in Radiant, which is building one-megawatt reactors designed to replace diesel generators. His broader nuclear thesis is that there are only two viable scales: one megawatt and gigawatt. He frames the failure to replicate Vogtle Units 3 and 4 across the country — taking the same workers, supply chains, and components and stamping them out — as "the biggest miss of our lifetimes."
The regulatory insight is specific. When a reactor designer makes a change to an application, today they have to map that change against hundreds of thousands of pages of NRC regulations, a process that can take three months and cost $10 million in consultant fees. AI changes that calculation: a language model tuned to the regulatory code can identify every affected statute in minutes. The same tool works for regulators reviewing applications. Ulevitch argues the entire nuclear approval timeline should compress dramatically as a result — something that wasn't possible before the current AI wave. The dependency runs both ways: the grid needs nuclear to power AI data centers, and AI can now help unlock nuclear capacity faster.
Space
Ulevitch defers to his partner Katherine on launch cost economics, noting that her expertise covers what happens after launch costs fall to commodity levels — specifically, which companies emerge once reliable Earth return is achievable.