AI’s Next Venture Frontier Is Domain-Specific Software for Physical Systems

Nina Achadjian frames the next frontier for venture investment as a move from software that augments knowledge workers to AI that operates in the physical world. At Index Ventures, she said, much of the firm’s recent AI work has been in the stack built around large language models: foundation models such as Anthropic, inference infrastructure such as Fireworks, and application companies built on top.

The distinction matters because physical AI is not simply another application layer. Ed Ludlow put the shift in terms of the old venture maxim that “software is eating the world,” and the newer claim that hardware is eating the world. Achadjian’s answer was that the opportunity is the combination: better software embedded into domains where physical systems, engineering constraints, and failure modes are much less forgiving than in digital software.

She pointed to electrical engineers, mechanical engineers, and aerospace engineers as examples of technical workers who have used largely the same software stack for decades. Many of the incumbents serving those functions, she said, were built in the 1980s. That creates room for new companies, but only if they can solve “deeply technical, very domain-specific” problems.

The stakes are different from software that never touches the physical world. Achadjian contrasted software errors in digital products with engineering errors in physical systems: if one character is wrong in code tied to a nuclear reactor or a rocket, “something could actually blow up.” That makes the opportunity attractive, but also narrows the kind of founder and product that can credibly pursue it.

Ed Ludlow raised a practical problem for investors in humanoid robotics and other physical-AI categories: public-market comparables are scarce. In software, he said, venture investors often point to an existing public company as a proxy for valuation or market size. In robotics and AI grounded in real-world physics, the proxies are harder to find.

Public-market weakness sat in the background of that question. Ludlow opened by noting that there was “a lot going on” in public markets while private markets in physical AI remained active. On-screen market graphics showed the Nasdaq 100 down 2.64%, the PHLX Semiconductor index down 6.04%, and the S&P 500 down 1.40% intraday.

That is why Ludlow asked whether SpaceX would be a good proxy for the category. Achadjian said “all eyes are on the SpaceX IPO” because, in her view, it has opened investors’ eyes to how compelling and lucrative the market could be. She said that attention is drawing more dollars into private companies doing similar things.

Ludlow then played a clip of Trae Stephens, the Anduril co-founder and executive chairman, giving a concise version of Founders Fund’s SpaceX exposure and lesson from the investment.

Ludlow summarized that as a “pretty simple” rule: do not bet against Elon Musk. Achadjian’s answer focused instead on SpaceX as a talent source. Her version was operational: Elon Musk and SpaceX have become powerful talent magnets, and the people trained inside SpaceX are now founding companies that address specific infrastructure gaps in hardware development.

Nina Achadjian said two of her most recent investments were in companies founded by former SpaceX employees. The point was not merely that SpaceX alumni carry prestige. It was that their work inside SpaceX exposed them to concrete bottlenecks in building hardware-heavy systems.

One example was Scott Morton, founder of Revel. Achadjian said Morton spent 10 years at SpaceX working on launch control. That role, in her account, gave him a direct view into “how big of a need there is for better software for hardware,” which became the basis for Revel.

The second example was Sergiy Nesterenko, founder of Quilter. Achadjian said he spent six years at SpaceX and became frustrated that he had to wait for a human to manually lay out a printed circuit board. Printed circuit boards, she noted, are important to “everything that’s happening in chips.” Quilter’s response is to use AI to autonomously lay out PCBs.

These examples clarify what Index is looking for in the category. The target is not a vague “hardware plus AI” company. It is a company built around a specific engineering workflow where existing tools or manual processes create a constraint the founder has personally experienced in a demanding technical environment.

The broader claim is that AI’s physical-world opportunity depends on pairing software advances with domain expertise. Achadjian’s examples sit in engineering workflows rather than consumer-facing demonstrations: launch-control-adjacent hardware software, PCB layout, chips, and systems where errors can have physical consequences.

The SpaceX IPO question also carries a more general venture-capital implication: whether large listings can reopen paths for companies that have stayed private. Ludlow framed the public-market side as potentially volatile “one way or another,” then asked whether the venture industry needs big IPOs to get the market moving.

Nina Achadjian answered from the industry’s point of view rather than speaking about any one portfolio company. Venture capital, she said, is “always rooting for the IPO window to be open” because it creates more options for founders: more access to capital and more paths to liquidity.

That answer keeps the SpaceX discussion tied to the rest of the thesis without making the IPO do all the work. In Achadjian’s telling, investor attention around SpaceX can help draw capital toward physical-world technology companies, but her underlying investment logic rested on a narrower claim: the software stack for engineers building physical systems is outdated, AI can improve domain-specific technical workflows, and SpaceX has trained founders who know where those workflows break.