The next OpenAI will be built in a factory, a data center, or a defense program

David George is making a specific, checkable call: the next company to reach the scale of SpaceX or OpenAI will come from the physical world, not from software. The domains he points to are defense, manufacturing, and data centers. The horizon is implied by the word “next,” which puts this in range of verification within a decade. If George is right, the generation of companies that build on artificial intelligence at civilizational scale will be the ones that moved atoms, not just bits.

The reasoning behind the call is not stated in detail, but the structure of it is clear. OpenAI and SpaceX share a specific profile: they entered domains where incumbent structures were slow, capital costs were high, and the gap between what was technically possible and what was being attempted was enormous. Physical-world sectors fit that description today in ways that pure software no longer does. Defense, manufacturing, and data center infrastructure all carry barriers that filter out casual entrants and reward organizations that can absorb complexity over long timelines.

The capital markets appear to be forming a similar view. Public reporting from Value Add Pulse shows more than $21 billion flowing into physical AI infrastructure in a single week in mid-2026, none of it directed at foundation-model labs. Hadrian, a company building AI-driven factories for defense and aerospace manufacturers, raised $1.37 billion at a valuation of $7.87 billion in a Series D round, according to reporting from Tech Funding News. That valuation represents nearly five times where investors priced the company in January of the same year. A separate company, Lumilens, emerged from two years of stealth with over $700 million at a $5.51 billion valuation to build optical interconnects for AI data centers.

The next SpaceX AI or OpenAI are probably going to get created. And they'll probably be in those kinds of domains. David George

None of this capital movement proves George’s call correct. A funding surge can precede a correction as easily as it can precede a category-defining company. The history of hard-tech investment cycles is full of well-capitalized efforts that did not produce an outlier at the scale George is describing. What the capital does show is that the conviction is not his alone, even if the framing is.

The distinction George is drawing matters because scale at the level of OpenAI or SpaceX is not just a financial outcome. Both companies restructured what their respective fields believed was possible, and they did it by operating in domains where the feedback loops between capital, technology, and institutional adoption were long enough to build durable moats. Pure software AI companies face a different dynamic: commoditization of models is moving faster than in any prior software cycle, and the application layer is crowded. Physical-world domains offer the opposite problem set: they are hard to enter, slow to iterate in, and difficult to displace once a company earns procurement relationships and operational trust.

George’s call will be testable. If the next decade produces a company at SpaceX or OpenAI scale, its domain of origin will either confirm or refute the prediction. The interesting question between now and then is whether the organizations best positioned to build in defense, manufacturing, and infrastructure are the ones currently receiving that capital, or whether the company that fits George’s description has not yet been founded.