Invest1 publisher2 min readPublished
Van Espahbodi says frontier AI labs are now more capital-intensive than hardware startups
The investor sold out of the Starburst accelerator and has backed 14 industrial and hard-tech companies since January 2023, on a screen that pairs commoditized hardware with software talent and skips science.
The Investor · Invest desk
What happened
- Van Espahbodi has spent 25 years in and around aerospace and defense, starting as a congressional staffer and later working in Raytheon's CEO office on foreign military sales.
- A decade ago he co-founded the Starburst accelerator, and on the advice of friends at Founders Fund he opened an office in El Segundo near SpaceX as alumni left to start hard-tech companies.
- Generational Partners, which he founded after selling his Starburst stake, has backed 14 companies since making its first investment in January 2023.
- Crunchbase News headlined its interview with him around a warning about "tourists and FOMO" as software VCs chase SpaceX alumni.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- exposure Espahbodi advises federal agencies on working with nontraditional, venture-backed companies while running a fund invested in that cohort, so the agencies taking his advice are hearing from someone with a position in the outcome.
- constraint He said he does not necessarily see opportunities in quantum computing, nuclear fusion or laboratory spinouts, the categories that absorb the most science risk.
- contradiction The headline warning is about price behaviour among software investors; the reasoning Crunchbase published is about a definitional mix-up in what they are buying.
- capability If former SpaceX, Tesla and Rivian staff built digital foundations that AI now augments, factory-scaling teams become underwritable in industries they have never worked in.
The screen is four conditions wide. Espahbodi told Crunchbase he looks for "creative software talent combined with commoditized hardware, significant customer demand and a new business model" [12]. He also said, plainly, "I'm not looking to invest in science" [13]. The distinction he insists on is definitional: "People often confuse hard tech with deep tech, but nomenclature aside, you don't need to invest in science to win in these categories" [15].
That puts the capital question the other way up. "Another major component of the AI question is that frontier labs have become more expensive and capital-intensive than traditional hardware companies," he said [16]. He added that the success of those labs, together with the SpaceX IPO becoming an enormous wealth-creation event, raises questions about what is truly capital-intensive and what makes a product or its intellectual property defensible [17].
He dates the move to about four years before the interview, when he noticed friends leaving SpaceX and moving horizontally across physical industries [22]; he sold his equity in the accelerator and part of the investment team left with him [23]. The deployment record since is short. Fourteen companies from a first investment in January 2023, which went to a North Dakota drone company [7][8]. Across three years, that works out to a new name about every eleven weeks [25]. Crunchbase did not publish the fund's size, its check sizes, or a valuation for any of the 14 [26].
One portfolio company was founded by the team that built the factory for Starlink user terminals and scaled that assembly line at high volume [19]. While deploying those terminals globally, Espahbodi said, the team observed that poverty often stemmed from a lack of access to clean water [20].
The durable part of the argument is the capital comparison, because it changes who is qualified to underwrite what. If a hardware company can be built with commoditized parts and software people [12], then a software fund is not obviously the wrong owner of that risk. What the incoming money is bidding for is access to operators who have stood up a production line. It breaks in the numbers. If companies on his list come back for rounds at the scale he ascribes to frontier labs [16], the premise that hardware need not be capital-intensive has failed. A portfolio of 14 is small enough that one or two such rounds would show it [7].
What to watch
- Whether any of the 14 portfolio companies discloses a round large enough to sit beside frontier-lab budgets.
- Whether Generational Partners publishes a fund size, a check size, or a 15th investment.
- Whether the SpaceX IPO that Espahbodi cites delivers the wealth-creation event his environment argument rests on.