Inside Palantir: Building Software That Matters | Shyam Sankar on a16z

Shyam Sankar's 2024 essay «First Breakfast» on defense reformation was an act of equal parts desperation and optimism. After nearly two decades watching the Pentagon from the inside, he witnessed a slow erosion of American deterrence — Crimea annexed, the Spratly Islands militarized, Iran's nuclear ambitions unchecked, and the October 7th pogrom in Israel. Each event in isolation could be dismissed, but together they revealed a loss of deterrence that had frog-boiled its way into crisis. The essay crystallized a diagnosis: America had accidentally turned its back on the very principles that once made it dominant. Yet Sankar saw a parallel resurgence outside the Pentagon — founders re-emerging, energy for building in the national interest, and a moment to reclaim past strengths with vigor. The shot clock was running out, but the opportunity for reformation had arrived.

The impetus for going public wasn't ideological performance; it was seeing what was possible. In the last year, Sankar has witnessed more change inside the Department of Defense than in the prior 19 years combined. Leadership — both inside and outside the building — has set conditions for heretics to thrive. Sankar points to a «conspiracy coalition of the willing» that recognized the urgency and acted. The election played a role, he acknowledges, not for partisan reasons but because it installed leaders who viewed the problem with clarity and empowered those capable of fixing it. The result is a mobilization of talent, ambition, and institutional courage that hadn't been seen in decades.

After October 7th, Israel mobilized 360,000 reservists — most with 20 years of industry experience. They were horrified at the state of IDF technology and modernized more in four months than Sankar had seen in 10 years of working with them. The implicit self-critique was clear: «At 20, I could code, but I didn't know what I was doing. At 40, I know how to build internet-scale systems.» America is drowning in that talent, Sankar argues, yet the defense establishment has made voluntary civil-military fusion nearly impossible. In World War II, the U.S. direct-commissioned 100,000 people into the military — authorities that still exist but lay dormant.

Sankar joined the Army alongside Bob McGrew (former Chief Research Officer at OpenAI), Boz (CTO of Meta), and Andrew Wheel (former Chief Product Officer at OpenAI). The program, launched by General George and Secretary Driscoll, places senior industry advisers inside the Army to accelerate software adoption, force structure planning, and operational data teams. The biggest surprise for Sankar has been the quality of talent among junior enlisted personnel. Green suiters with no formal CS training are building the most compelling AI applications he's seen anywhere — driven by existential stakes, not 10% efficiency gains. Ten years ago, these same soldiers would have made a PowerPoint and been dismissed by a program bureaucrat. Today, they spend two weeks building an empirical prototype, and the Army adopts it because everyone wants to win.

The AI value stack is undergoing violent reshuffling. Models are being commoditized and are under constant margin pressure. In response, model companies are expanding upward, building what they sometimes dismissively call a «harness» but is actually AI infrastructure — software for orchestration, memory, and tooling. Meanwhile, companies that started as vertical AI solutions are earning their way down the stack, realizing they need robust AI infrastructure to scale across use cases and customers. Sankar's thesis: value will accrue in two layers — chips at the bottom and AI infrastructure in the middle. These are defensible moats. At Palantir, that infrastructure layer is Ontology, the system that turns raw models into decision-making platforms. The application layer remains important, but without infrastructure, applications can't scale or adapt. This mirrors the cloud era, where infrastructure (AWS, Azure) captured enormous value even as applications flourished on top.

Hyman Rickover's story is the defense founder archetype Sankar wants Hollywood to tell. Born in a shtetl in Poland, Rickover arrived at Ellis Island at age six. His family nearly missed their pickup window — the telegram his mother sent was pocketed by a con man, and they were saved on day 11 by a chance encounter. Rickover became a notoriously difficult 5'2" admiral who drove a coal ship in World War II, an unglamorous post. After the war, he was sent to Oak Ridge, the vestige of the Manhattan Project, and conceived of putting nuclear power inside submarines. The Navy didn't want it. Oppenheimer thought it wouldn't work. Rickover's first office was a converted women's restroom — an intentional humiliation to make him quit.

Rickover built the USS Nautilus in five years. He demanded safety specifications 100 times stricter than the minimum standard, insisting it be «safe enough for my son.» Soviet submariners took six-month respites at Sochi to recover white blood cell counts from radiation exposure. The U.S. nuclear navy has had zero deaths due to radiation. Rickover served as a four-star admiral for 30 years, an tenure unthinkable today. Admiral Zumwalt, Chief of Naval Operations, said the Navy had three enemies: the Soviet Union, the Air Force, and Hyman Rickover. That level of institutional friction is the price of transformative innovation. Only a founder, Sankar argues, could sustain that level of abuse and still deliver a capability that defines American power for generations.