Will Manidis argues AI will create a 'sovereign frontier' of classified models and dollarize the world through agent transactions

Commanding Heights

Will Manidis has spent the past year writing Commanding Heights, a 140-page manuscript co-authored with Joe Roberts, focused on what AI scaling means for currencies, nations, and the structure of the global economy — territory he argues the finance world has been slow to engage with seriously.

The sovereign frontier

The book's central claim is that frontier AI models will become inaccessible to the public. As training costs compound along the Epoch AI scaling curve, Manidis argues only one or two states will be able to afford to keep training world-class models. The best American models end up inside classified facilities; the Chinese maintain their own equivalent. Everyone else sits within the diffusion net of whatever open-source models those powers allow out.

For commercial workloads, Manidis expects Chinese open-source models to power most US businesses — a trend he says is already visible in OpenRouter data. Frontier lab revenues, meanwhile, shift increasingly toward the US government as the only customer that can actually finance the training runs.

“Labs will become something much closer to the defense industrial complex, that they will have models that are available to nation states for national security purposes. You will not be using them. If these models are transacting in the dollar, if all of agents are transacting in the dollar, then the rest of the world is going to dollarize quite quickly. We introduced the concept of the neo dollar.”

The neo-dollar

The second major idea is the neo-dollar. If AI agents transact in dollars, if the sovereign frontier is dollar-denominated, the rest of the world dollarizes to participate. Manidis sees two competing financial systems — the dollar and the RMB — each racing to generate enough capital to maintain their sovereign tier.

The paradox he flags: the US will print enormous amounts of dollars to fund the AI buildout, inflating the currency toward zero at the same moment it spreads dollar dominance globally. His portfolio implication is straightforward — own equities, because long equity is short the dollar.

Commanding heights businesses

The book's investment thesis centers on what Manidis calls "commanding heights" businesses — companies that get better as intelligence scales. Meta is his clearest example: payment rails, WhatsApp's reach across the global south, and frontier capability through Meta Superintelligence Labs combine into, by his read, the single best-positioned business to oversee both the dollarization of the world and the diffusion of intelligence across it.

On the ADT acquisition by Thrive, Manidis calls ADT's sensor and camera network "probably the greatest on Earth" — a dataset he thinks has to be among the most valuable in the world even if the core business generates no AI lift.

The forced settlement

On firm structure, Manidis argues labor will leave the firm the way it offshored in the 1990s and 2000s — this time into models and data centers. Management teams shrink and get compensated like capital. He uses Alex Wang at Meta as a marker of that compression. The result is leaner firms with stronger balance sheets, more single-proprietor businesses at the bottom end, and heavy consolidation in regulated industries — banking and healthcare specifically — driven by cybersecurity pressure. The argument is that cyber threats are now sophisticated enough that only the largest organizations can mount a meaningful defense, making fragmentation in community banking and community hospitals untenable.

Interest rates

If technology progress continues at the pace Manidis expects, he argues rates will need to go significantly higher. He adds a non-linear wrinkle: when the risk-free rate doubles, the required return on risk assets rises by a four-to-five-times multiple, not a proportional one. The data center buildout alone, in his telling, could saturate the monetary supply.

---

Five minutes after publishing, Manidis received a six-page AI-generated rebuttal — which he takes as confirmation that critical reasoning is already within the capability cone of current models. The book's darker projections, particularly around Southeast Asia, have drawn the most resistance. A thousand physical copies shipped on the day of the conversation.