Keith Rabois on why most defense startups will fail, Ramp's pitch to Congress, and OpenAI saving San Francisco
May 1, 2025 with Keith Rabois
Key Points
- Keith Rabois pitched Ramp's expense-management software to Congress as a federal budget fix, citing data that commercial customers save 5% to 15% on expenses and arguing the government could recover waste on 4.5 million credit card accounts without cutting programs.
- Rabois estimates only 40 people in the country can close defense contracts, making sales talent the structural constraint that will cause most defense startups to fail regardless of technical strength.
- Rabois credits OpenAI's success with saving San Francisco from continued decline, arguing the city would still rank outside the top 15 US metros without the company's breakthrough.
Summary
Keith Rabois appeared at a conference in Washington, DC, where he had just come off a panel with Eric Lyman and Senator Joni Ernst of Iowa making the case for applying Ramp's expense-management software to federal spending.
The pitch is straightforward: Ramp's commercial customers save between 5% and 15% on expenses through programmatic reporting, and the federal government, with 4.5 million government credit card holders against roughly 2 million employees, would likely save even more. Ernst cited 80,000 charges on federal credit cards at casinos as an example of the kind of waste that doesn't require cutting popular programs to fix. Rabois frames the macro payoff in interest-rate terms — cut 1% from the federal budget, and the Fed gains room to lower rates by 1%, which flows through to mortgage affordability and non-inflationary growth.
Defense tech
Rabois is skeptical of the defense startup wave. AI remains the dominant category at the conference, with defense tech second, and he thinks that positioning is a structural problem rather than an opportunity. The core constraint is sales talent, not technology — he estimates there are roughly 40 people in the country who can actually close contracts with three-letter agencies and defense buyers. Without one of them, a technically strong startup is still likely to fail.
He agrees with Delian Asparouhov's earlier comments that most defense companies won't make it. His fund is not actively deploying into government-only solutions, though he made one recent defense-oriented investment he describes as an outlier. The more durable shift he sees is companies adding federal revenue earlier in their trajectory — Scale AI being the clearest example — rather than building around government contracts from inception. Companies that survive will earn federal revenue as a supplement to commercial success, not as a substitute for it.
The historical bar is brutal: Rabois puts the odds of building an Oracle-scale government technology franchise at roughly one per decade.
OpenAI and San Francisco
Rabois's view on San Francisco's recovery is blunter than Jensen Huang's. His version: OpenAI saved SF. Without OpenAI's success, he argues, the city would still be in decline. The pattern is familiar — before the internet boom, SF ranked 12th to 15th among US cities by commercial activity, with Oakland outperforming it. The internet created enough spillover to rescue San Francisco without the city's culture being especially hospitable to company building.
He illustrates the point with a 2001 PayPal anecdote: Peter Thiel asked the team to name a successful San Francisco company before deciding whether to move the office there. The best answer anyone could produce was CNC. PayPal moved to Mountain View instead.