The Largest Untapped Market Inside U.S. Defense | Emil Michael

2026-01-15 - 44 min - source - Read full transcript

Sarah Guo (host)Elad Gil (host)Emil Michael

Key insights

The Department of War (formerly Defense) is deliberately opening its front door to startups instead of forcing them to sue for their first contract.

Michael cites SpaceX, Anduril, and Palantir as companies that historically had to sue the department to win initial work. He frames the current mandate as reversing that dynamic given a consolidated industrial base of roughly five prime contractors (down from about 50 in the 1980s) that leaves little room for new entrants without deliberate intervention.

defense-innovation-procurement

Michael collapsed 14 inherited 'critical technology areas' down to six action-oriented priorities because too many priorities functions as no priorities.

He compares it to a Series B company claiming 14 product lines being 'uninvestable.' The six areas are run with engineering-style sprints so people can hold them in their heads and execute daily instead of spreading resources too thin across dispersed goals.

defense-innovation-procurement

The department is not building its own foundation model; applied AI means adapting frontier commercial models to department use cases.

Michael argues hundreds of billions of private-sector dollars are already being spent on foundation models, so the department's job is integration and applied deployment rather than model-building from scratch - captured in GenAI.mil, which reached over a million unique users out of 3 million employees within 30 days of launch.

applied-ai-deployment

Standing up secure generative AI access for a 3-million-person organization required building an entirely separate data architecture, not just signing up for ChatGPT.

A tiger team of former Databricks, Meta, and AWS engineers built GenAI.mil in about 60 days with early collaboration from Google's Gemini team, specifically engineering data flows so nothing sent through the tool feeds back into public model training pools.

applied-ai-deployment

Applied AI inside the department is organized around three use-case categories: enterprise efficiency, intelligence analyst augmentation, and war-fighting simulation/planning.

For intelligence, Michael describes AI surfacing relevant signals to human analysts and fusing multiple intelligence types that a single analyst normally can't cross-reference alone, which he calls a force multiplier for an already strong US intelligence collection base.

applied-ai-deployment

Robots and drones are becoming the new front line in territorial warfare, reducing human casualties in land conflicts.

Michael points to the Russia-Ukraine war as a renaissance moment for drone and robot warfare, arguing that without it casualties would have been far higher, and predicts the shift toward autonomous systems will accelerate fastest in territorial/land battles versus sea power projection, which still requires carriers and human-crewed vessels for now.

autonomous-systems-warfare

Michael expects 20-30% of the defense budget to shift to autonomous systems within about a decade because they deliver more firepower per dollar.

He frames this as a straightforward cost/capability tradeoff: autonomous submarines, boats, and aircraft are cheaper to field than crewed equivalents while AI improves their sensing and independent decision-making, so budget allocation will track that improving capability curve year over year.

autonomous-systems-warfare

AI-native design, not incumbency, determined which robotics company progressed faster - Tesla's Optimus overtook Boston Dynamics because it was AI-infused from the start.

Michael uses this as evidence that the next wave of defense robotics winners will be those that build AI into the core system rather than bolting it onto legacy hardware platforms.

autonomous-systems-warfare

The US outsourced too much of its critical supply chain - pharmaceuticals, critical minerals, semiconductors, nuclear plant-building capability - and is now treating reshoring as a matter of national survival, not just economics.

Michael frames it as a Maslow's-hierarchy-style prioritization problem, citing a one-year self-sufficiency deadline implied by a critical-minerals export-control agreement with China, and notes roughly 200 pharmaceuticals currently sit on a critical shortage list due to import dependency.

supply-chain-security

China and other adversaries have distorted global commodity markets through subsidies and dumping, so US reshoring can't rely on true free-market price competition alone.

Michael and Gil discuss how subsidized rare-earth production undercut competing industries elsewhere, framing tariffs as a partial lever to offset that dumping dynamic rather than a rejection of free markets outright.

us-china-competition

The Office of Strategic Capital deploys roughly $200 billion in low-cost lending (about Treasuries plus 100 basis points) to expand production capacity at existing suppliers in brittle parts of the defense supply chain, such as germanium, brushless motors, and fiber optic cable manufacturing.

Michael describes identifying dozens of overlooked single-point-of-failure components - like one US nozzle maker supplying 20 different systems - and using lending plus signaling to attract private equity capital around those bottlenecks, aiming to fund two or three domestic suppliers per critical component instead of one.

supply-chain-security

China's military buildup is the largest in world history, and it is driving a step-change in required US defense investment and speed of execution.

Michael contrasts his first government stint (2009-2010, White House fellow during the global war on terror against low-tech insurgents) with the current environment, where the US faces a near-peer adversary with space capabilities, long-range strike, drones, and AI - a shift he says he understood theoretically before but now understands 'viscerally' after seeing the scale of supply-chain dependencies up close.

us-china-competition

Companies

SpaceX - Cited as a company that had to sue the Department of War to win its first contract, illustrating the old barrier-to-entry problem the department is now trying to remove.
Anduril - Cited alongside SpaceX and Palantir as having had to sue for its first defense contract.
Palantir - Cited alongside SpaceX and Anduril as having had to sue for its first defense contract.
Saronic - Named as a Defense Innovation Unit early-funding success story (autonomous surface vessels).
OpenAI - Referenced as the ChatGPT provider whose consumer product department staff are barred from using directly, driving the need for a walled-off GenAI.mil deployment.
Google DeepMind / Gemini - Gemini was the first frontier model integrated into GenAI.mil, credited with fast, hands-on collaboration to get the rollout live in about 60 days.
xAI / Grok - Named as the next model to be integrated into GenAI.mil after Gemini.
Anthropic / Claude - Named among the frontier-model providers the department has to architect data flows around so no department data leaks back into public training pools.
Boston Dynamics - Contrasted with Tesla Optimus as an example of a robotics company that progressed more slowly because it was less AI-infused from the start.
Tesla (Optimus) - Cited as outpacing Boston Dynamics because its robotics program was AI-infused from the beginning.
Andreessen Horowitz - Partner Scott Kupor is deploying the US Tech Force, a program aiming to place thousands of recent college graduates into two-year technologist tours at the Department of War.
Lattice Materials - Cited as a germanium and silicon crystal producer in Bozeman, Montana that received Department of War investment for defense optics supply chain security.

Techniques and frameworks

Department of War critical technology areas (14 cut to 6) - Michael inherited 14 'critical technology areas,' concluded that many priorities meant no priorities, and cut the list to six action-oriented, sprint-managed areas including applied AI, scaled hypersonics, and scaled directed energy.
GenAI.mil - A department-wide, security-architected generative AI platform (currently on a version behind ChatGPT-equivalent capability, with 2.5 live and 3.0 imminent) built in about 60 days by a tiger team of former Databricks, Meta, and AWS engineers, reaching over a million unique users among 3 million department employees in 30 days.
Problem-statement RFPs vs. 300-page requirement documents - The department is shifting procurement from detailed spec documents that favor incumbent primes toward stating the problem and letting vendors propose novel solutions, which favors startups.
Valley of death capital stack - A layered set of funding mechanisms - the Defense Innovation Unit (~$1B for rapid contracting), AFIT (scale-up manufacturing funding), and the Office of Strategic Capital ($200B in low-cost lending authority at roughly Treasuries plus 100 basis points) - designed to keep startups funded between prototype and production scale.
Arsenal of Freedom tour - A joint tour with Secretary Hegseth visiting roughly 20 defense-industrial-base companies across the country to signal that the department wants industry to produce at scale and will cut bureaucracy to support it.

Summary

Emil Michael, the former Uber chief business officer now serving as under secretary for research and engineering (effectively CTO) at the renamed Department of War, walks through his first eight months reorganizing how the Pentagon buys and deploys technology. His portfolio spans DARPA, the department's chief AI office, the Defense Innovation Unit, and the Strategic Capability Office, unified around a roughly $150 billion annual technology budget. His first move was ruthless prioritization: he inherited 14 "critical technology areas," judged that too many priorities amounts to no priorities, and cut the list to six sprint-managed areas including applied AI, scaled hypersonics, and scaled directed energy.

The applied-AI story is the episode's centerpiece. Rather than building a foundation model in-house, Michael's team stood up GenAI.mil, a security-architected generative AI platform built in roughly 60 days by a tiger team of former Databricks, Meta, and AWS engineers, launching with Google's Gemini and later adding Grok, and reaching over a million unique users out of 3 million department employees within 30 days. The technical challenge wasn't model access but data architecture: ensuring nothing sent through the tool leaks back into any provider's public training pool. He frames the department's AI push around three use cases - enterprise efficiency, intelligence-analyst augmentation, and war-fighting simulation - and expects autonomous systems (drones, robots, autonomous ships and submarines) to claim 20-30% of the defense budget within a decade as AI-driven sensing and decision-making improve their cost-effectiveness relative to crewed platforms.

A recurring thread is opening the department to startups that historically couldn't get in the door. Michael notes that SpaceX, Anduril, and Palantir all had to sue the department to win their first contracts, a symptom of an industrial base consolidated from roughly 50 primes in the 1980s down to about five today. He describes shifting procurement from 300-page requirement documents - which favor incumbents skilled at filling out questionnaires - toward problem statements that let smaller companies propose novel solutions, alongside the "Arsenal of Freedom" tour with Secretary Hegseth visiting roughly 20 defense-industrial-base companies to signal the department wants their production and will cut bureaucratic friction to get it.

Supply-chain fragility is the other major theme. Michael describes treating reshoring of pharmaceuticals, critical minerals, semiconductors, and nuclear-plant-building capability as a matter of national survival rather than pure economics, citing a roughly one-year self-sufficiency deadline implied by a critical-minerals export-control arrangement with China and around 200 pharmaceuticals currently on a critical shortage list. The Office of Strategic Capital deploys about $200 billion in low-cost lending to expand capacity at existing suppliers of brittle single-point-of-failure components - citing examples like germanium, brushless motors for drones, and fiber optic cable - aiming to fund multiple domestic suppliers per critical part rather than leaving the country dependent on one.

Michael closes by contrasting his current role with his first government stint as a White House fellow in 2009-2010 during the war on terror, when the adversary was low-tech insurgents rather than a near-peer power with space, long-range strike, drone, and AI capabilities. He says the scale of China's military buildup and the depth of US supply-chain dependency - which he understood theoretically before taking the job - only hit him "viscerally" once inside, reinforcing his urgency around both technology adoption and industrial reshoring.

Notable Quotes

"It's crazy to me that SpaceX and Anduril and Palantir all had to sue the Department of War for their first contract. So the idea is you don't have to sue anymore. Come through the front door." - Emil Michael

"If there's that many, they can't be critical." - Emil Michael, on cutting 14 "critical technology areas" down to six

"Hardware is hard in Silicon Valley. Well, hardware is hard also in the defense industry." - Emil Michael

"The military buildup in China is the biggest military buildup in world history. And so there's a real urgency on our side to ensure that we are ahead, but we stay ahead." - Emil Michael