Will Manidis

A few weeks ago, the Secret Service locked Jensen Huang, Elon Musk, Sergey Brin, Satya Nadella, Lisa Su, and Michael Dell in a room in Washington, DC. While they waited for President Trump, Huang—who chooses not to own a jet—complained about a recent commercial plane ride. It had no Wi-Fi. “Best way to sandbox an AI,” Musk said. “Put it on a Delta flight.” All six were about to receive national medals from the President. The ceremony was the idea of Michael Kratsios, Trump's chief science advisor. He is 39, a Peter Thiel protégé, and has worked for Trump every day of both terms. For decades, the office Kratsios runs was known mostly for reports on things like algae blooms. Kratsios himself was barely known in Trump's first term. Now he sits between the President and tech royalty. He is one of a handful of people shaping America's AI policy, and the White House's point man on nuclear, space, and quantum. In our latest feature, @harrytlambert shows how the fight to regulate AI and reform American science runs through his office. colossus.com/article/michael…

The commanding heights - just read the whole thing. Took me about an hour to read, is the talk of the AI and Wall St crowd and essentially argues. - The US and China will be the two countries with AI, which creates a cold war condition. - The US durable advantage is compute and capital, with 70% of compute compared to 5% in China meaning they can stay at the frontier. - The US state will see this as a coming cold war, and look to own more of these companies, just as threatened states have historically in war conditions - China will not monetize its models because it won’t let its currency float, meaning even Chinese models will be better monetized by the dollar. - The most neutral dollar is a stablecoin, which while not always controlled by the US as tightly as the fiat dollar, will enable global debasement - AI labs will make such a massive portion of the K shaped economy, the government needs them to continue to grow - Therefore the capex must continue to go up and the US Government will backstop that, and potentially, put a limit on the yield from US Treasuries to ensure funding can continue into capex. - Therefore the US government look to own these companies either via a golden share or gifted equity. - AI will inextricably replace labor, causing developing economies to enter economic hardship, but the US will be partly insulated by government ownership enabling some level of redistribution - The value of AI will not accrue to labs which will become government subsidized utilities, or the existing AI supply chain bottlenecks, but the other industries that emerge because AI exists. Some thoughts. 🧠 This is dystopian as hell. And while the chain of logic is rational, it's also not confidence-weighted. There are several thousand words devoted to Ravi, who works in India, who’s basically seeing all of his friends out of work and is ultimately laid off via WhatsApp. The evidence is the gap between Tata Consulting Services Headcount and profitability. And *sure* there's a bit of this, but it doesn't have to end in riots on the street. Woof its dark. 🧠 They call the Wall St leveraged buy out firms of the 1980s “buyout artists.” This is clearly a play to attract capital to raise a sort of permanent capita / hedge fund vehicle for buyouts. It also speaks to the viewpoints of the writers. The piece regularly quotes scripture and is self-indulgent and elitist as hell, from the font to the tone, to the worship of Leopold. That creates its single biggest blindspot. Pragmatic realities. It's a well studied, well reasoned piece but distorted. 🧠 The link for stablecoins is perhaps the best I've seen. Person in Argentina wants to get paid and pay for AI, stablecoin linked cards are a great way to do that. China won't let its currency be debased and lacks the compute to serve even open-weight models. And yes, a person in Argentina is using stablecoin-linked cards today... But. China is building compute capacity. And stablecoins are so early - I think it's hard to credibly say this is an inevitability. 🧠 People are saying this is bullish for certain stocks, while hyperscaler share prices are cratering. As I write this Amazon stock is down 15% over the week, Google 16.7%. That could be the essay’s thesis, that value is moving away from those who own the rails. But its more about these companies entering negative free cash flow and AI impacting their ads business. I don't get how this is anything other than a play for investment dollars. If you're seeing something else here lmk, but convince me you read the whole thing first. 🧠 I don’t buy that existing incumbents are sitting ducks here. Regulation exists, network effects exist. Someone like an Amazon or JP Morgan can block AI agents, and will do so, and will bring people to the negotiating table.

I'm still processing commanding-heights.com/ by @WillManidis and Joe Roberts, but think two of its arguments deserve deeper discussion: First: the contention that Chinese labs intentionally accept extremely thin margin and are competing away US inference margins. I don’t believe that is borne out by the data, or by my direct experience running open source models on our own infra at Tolan. Instead, I believe low Chinese inference margins are largely a function of access to lower quality hardware and a more limited inference buildout on the Chinese mainland. Some back of the envelope math: at prevailing spot compute prices, US providers can run open source models at roughly 45-60% gross margins (B200 at ~$5.70/hr spot market estimates, observed GLM-5 token throughput at 80% utilization). For context, Zhipu reported a 24.6% gross margin for its open-platform and API business in H1 2026, up from −0.4% a year earlier. Today, Z.ai and Together both list GLM-5.3 at $1.40 per million uncached input tokens and $4.40 per million output tokens, though those prices postdate the reporting period and don’t establish equivalent realized margins. In short: at prevailing open source model pricing, US inference providers seem to be able to earn healthy margins relative to their Chinese counterparts.

There are a few things that Manidis and Roberts get exactly right. Small firms will achieve outcomes completely disproportionate to their headcount. These “raiding party” sized groups will outperform at a shocking scale. The countries that rose through the outsourcing of backoffices will fall as those jobs evaporate into the algo. Call center countries reduced to scam hovels. Scarce, permanent assets will be the holy grail of the AI economy. Assets that cannot be vibe-coded, that will grow under AI dominance rather than shrink. These will often be "rails", physical sites, or “nation-state” assets like a power grid, a copper mine, or a strategic landmass. “The fourth settlement is one sympathetic to great men and leaves little room for labor as you know it.” This is one of the best lines in the piece and a nice key to the next few decades. If you told people two hundred years ago that labor would be abolished, they would be stoked. Today they whine, “don’t take my toil, let me listen to a pod while I type.” Inevitably, some of the predictions in the piece will miss. Where this is the case, I think it will be a result of the authors' sheer proximity to the center. Too near to the fire, lobotomized by the Flow. NYC and SF crowds are certain that their world is the only one. A perennial solipsism dominates both bubbles, a gravitational pull that Manidis and Roberts (and the friendly ghost of Leopold Aschenbrenner) are not immune to. The authors argue that all capital will be made subservient to the Line. The Line must go up. AI must continue its march, even irrationally, to prevent massive capital destruction of all that has come before. But outside of the epicenters, the people have no allegiance to the Line. My brother in Idaho has no allegiance to the Line. Your friend who bartends in Orlando has no allegiance to the Line. For the Leopolds in SF and the FO boys on wall street, it may be their only reality, their existence and work and net worth but the everyman has not corrupted his mind with expectations of winning the final gluttonous end times tech lottery. He has no allegiance to the Line. Manidis and Roberts argue that the populace will be tied via pensions, equities, and trump accounts to the success of AI, but I don’t see that that coupling has truly taken place yet. Downstream of this, I still don't see the US nationalizing the AI labs. The labs tried hard but the critical moment has come and gone. The administration has already decided that it is intolerable to be equity partners with Altman or Dario. Some will live and some will die. As long as boomers have a blood in their veins, they'll cling to their free-markets while the groypers and DSA kids would rather see recession than the tech oligarchs and bankers enriched in victory. Where does this take us? Coming years will be dominated by powerful firms in the private sector and great men, value will accrue to assets that are scarce and durable, companies will act more like countries. Will and Joe are going to make some killer, generational investments during this era. Thanks to Will and Joe for writing and giving people something think about and to shoot at. I saw some folks complain that the piece is too grim or apocalyptic. I blame the popularization of end times biblical language by the one paypal co-founder. The future that the bible actually predicts is much more interesting. One of the inevitable victory of christendom–––the past and future conquest of Christ over all. Shaping the victorious timeline is our ultimate duty and opportunity.