Delian Asparouhov on the DOD budget freeze, slop in defense investing, and Varda's factory growth

Summary

Delian Asparouhov joins from Varda Space's factory floor — with capsules four and five visible behind him — to walk through the defense tech investment landscape, the DOD budget freeze, and what OpenAI's image generation release means for mid-market AI companies.

Defense tech: a generational arc

Asparouhov frames the evolution of defense investing in three phases. From roughly 2003 to 2015, breaking into the sector required being a billionaire willing to sue your only customer — SpaceX and Palantir both did exactly that around 2010–2012 to win early contracts. From 2015 to 2020, you still needed to be a billionaire, but at least litigation was optional. By 2020, being friends with a billionaire was close enough. The 2022 FOMO wave extrapolated that trend to its logical absurdity: high school students raising $1M seed rounds for defense startups.

He expects a meaningful pullback. Several VCs who now loudly claim defense mandates had the sector explicitly listed in the vice clause of their LPAs as recently as 2017 — alongside gambling and pornography. Some of those LPs, he argues, will push defense back into the vice clause after watching their managers pile into deals with no program-of-record path and hardware that keeps failing in the field.

The CR and the budget freeze

On the continuing resolution, Asparouhov says most of the pushback he received on a prior clip came from junior companies with rose-colored glasses. The sophisticated operators — he names Tim Hughes and Matt Grimm at Anduril, and Shyam Sankar at Palantir — didn't dispute his framing. His position is that the legal text of the CR matters less than how warfighters and Congress will actually interpret reprogramming authority. The CR nominally allows some flexibility, but that flexibility is constrained by what the House and Senate passed in late February, and the political antibodies in the system are real.

The ISS administrator fight is his clearest illustration. Elon Musk pushed to deorbit the ISS ahead of schedule; Ted Cruz, who chairs the Senate science committee and whose Texas constituents staff Johnson Space Center, responded by stalling Jared Isaacman's NASA administrator confirmation hearing. Isaacman is broadly liked by the space community — he funded and flew his own SpaceX missions and became the first private citizen to perform a spacewalk — but his confirmation is now likely delayed until May. NASA's reauthorization bill, which sets the agency's budget for the next five to seven years, is frozen in the interim. The lesson Asparouhov draws applies directly to defense: even with tech's most powerful figures at the levers of government, institutional balances of power don't simply dissolve.

Slop in defense investing

Asparouhov's diligence edge in hard tech is walking the factory floor and talking to the thermal engineer, the firmware engineer, and the technician doing assembly. It won't make him a world-class engineer in any of those disciplines, but it lets him gauge whether the quality bar across a team holds. He describes watching investors write checks of $100M-plus into companies where satellites are consistently failing on orbit, no contracts have closed, and headcount growth is being used as a proxy for success. Headcount scaling at a company with a broken product culture, he argues, just means you've hired more bad people faster.

He puts the likely attrition rate for the current field at 50–60% over the next one to two years — dead or acquired for pennies on the dollar.

OpenAI's image generation release

On the GPT-4o image generation launch, Asparouhov's read is that a large cohort of standalone AI image generation companies just got Sherlocked by OpenAI simultaneously on brand and technology. The consumer play for those companies is effectively over. His tactical advice for a $50M Series B-stage image generation company is to abandon consumer entirely, go deep on enterprise, and build proprietary studio relationships — pitching Hollywood on the risk that feeding content into ChatGPT means training OpenAI's model with your proprietary assets. The broader structural point mirrors his defense view: be bullish on the dominant platform and bullish on the end user who gets new tools, and avoid everything in between.

Varda factory update

Varda now has five capsules in various stages. The first landed last year; the second landed in Australia two weeks ago and is currently at an air show; the third is in orbit; capsules four and five are on the factory floor behind Asparouhov. The factory is visibly packed.