What it takes to build in America again

She starts with why the capital showed up. The category was lonely well over five years ago and now the water is warm, on what she calls a few different tailwinds. The first is bipartisan recognition that the country has to build again, felt through an aging grid, rising electricity prices and the difficulty of siting any new industrial facility, and including the realization of dependence on foreign adversaries for the large majority of critical materials and components. The second is proof that the returns exist, which she attributes to Palantir’s share price, Anduril’s performance and a SpaceX public offering that gave a long-illiquid class of investors a large exit. To that she adds the strategic case: in the age of AI, physical plant is a moat precisely because it is expensive and slow, since it is not easy for OpenAI or Anthropic to release a new model that suddenly displaces a factory.

Her correction to the intuitive picture of the grid is the sharpest technical turn in the episode. The instinct is that bigger is better and the country needs enormous new plants, nuclear or gas. She says the bottleneck is not generation at all but getting power from where it is made to where it is used, which makes the real obstacle transmission and the grid interconnection. You could build as many large gas plants as you like and still be stuck. That reverses what counts as good technology. The distributed home batteries her firm backs in Base Power substitute thousands of small sites for a single point of failure. Smaller nuclear reactors may not be optimized for price per unit of electricity but are optimized for speed to power, built the way a car or a prefabricated home is built, and sited where the power is needed so that no interconnection queue applies at the point of generation. The same logic runs into hardware. The grid runs on technology invented in the 1940s and 1950s and deployed in the 1970s, now at end of life, heavy, and supplied by only a handful of makers worldwide. Her stronger claim is that this equipment is not merely old but mute, so the system is run essentially blind, with no ability to observe or control it. Her answer is Heron Power, Drew Baglino’s company, whose thesis she states as taking the technology developed at Tesla, first for the vehicle powertrain and then for Tesla Energy, and scaling it up for the grid itself, starting with how data centers and generation projects connect to it. Its solid-state transformers swap the large drums of oil for silicon carbide, which she describes as software instead of hardware for managing electricity flow.

The financing picture follows the same shape: no shortage of money, a shortage of the conditions money needs. The first plant, what she calls factory zero, is often funded with venture dollars, which are genuinely available right now. She is blunt that this is the most expensive capital a founder can use, since it trades away future cash flow and equity, and that it should not fund every subsequent dollar in the ground. Project finance arrives later, once there are customer deposits and a demonstrated ability to build to a timeline, and large credit funds are now buying into earlier equity rounds to position themselves as the long-term partner. The gap in the middle is the host’s framing: he puts almost unlimited capital behind the very large, utility-sized project with high-quality offtake and asks about the space in between, and she agrees in four words before putting it in her own weaker terms, that there is still a bit of a gap. Founders fill it with more venture, which she says is not necessarily the most efficient way to do it. Her diagnosis of why credit does not fill it is about credibility rather than risk appetite. Founders are telling conservative credit investors a story that sounds unbelievable against their experience: that AI can get a permit in six months instead of eighteen, that it can project-manage ten thousand contractors, that a project can come in ahead of schedule and under budget. She notes that the last phrase has dropped out of the American industrial vocabulary because the country has not done it in decades. Hence her view that a first finance chief who can translate technology-driven speed to a traditional finance audience is a very hard role to hire for.

That leads to the claim the episode is built around. Underwriting a business is mostly about certainty, and in these categories the uncertainty is regulatory. She is careful to say that regulations exist for a reason and that she is not arguing to get rid of all regulation. What she wants is that a business meeting the letter of the law gets a definitive answer in a reasonable period, and that the answer is final, with a limitation period so that it is not relitigated in court afterwards. Today, she says, a large industrial project can simply never get an answer, bounced between bodies for years, and that delay is what materially damages the return. Her conclusion is that this matters more than any particular financing tool the government holds, and that both the current and previous administrations have acted through executive authority when the durable version has to be codified in law so builders can plan across ten and twenty years rather than two. The extension beyond permits is the social license. Mariana Minerals, her mining portfolio company, worked with public high schools, local community colleges and community centers before it needed to, and she reframes the jobs comparison that gets made against it: the honest baseline is not the roughly 300 people the mine employed at its historical peak but the zero it employed after going bankrupt. Site selection follows the same logic. A factory zero gains from sitting next to its engineers, which for many of her companies still means California. But expansion is moving out of the state quickly and some companies have opted out of it altogether, Base Power building in Texas. The regions that solve grid modernization, cost allocation and training pipelines are the ones she expects to win the next twenty years.