The Genie and the Grid

There is a question about artificial intelligence that was worth asking five years ago and is not worth asking now: whether we should have built it at all. I have my own views on that question, and so do you, and they no longer matter in the slightest. The thing exists. The capital has been committed, the data centers are rising, the models are being trained on every continent that can afford the electricity. Whatever wisdom there might have been in restraint, the moment for it has passed. The genie is out of the bottle, and no one has ever put one back.

So let me set aside the entire debate that usually consumes these essays and grant the most ambitious version of the prevailing goal. Suppose you believe, with complete conviction, that the United States must win the race to build and deploy artificial intelligence - that this is the defining contest of the century, that losing it to China would be a civilizational error, that nothing should be allowed to stand in the way. I am not going to argue with any of that. I am going to assume it is exactly right and hold it fixed for the rest of this piece.

I want to be precise about what follows, because it is easy to misread. This is not an argument about values or which party should be in charge. It is an argument about means and ends. If you are serious about a goal, you must be serious about the conditions that make the goal achievable. And the policies currently being pursued in the name of American AI dominance are, taken as a whole, close to a systematic guarantee of failing at it. Not because anyone intends to fail. Because the program is at war with its own objective.

The clearest way to see this is to be honest about what “winning” means. It does not mean having the cleverest model on a leaderboard in some particular month. A leaderboard is a vanity metric. Winning means deploying the technology - broadly, across the whole economy, at a scale and a cost the country can actually sustain. It means adoption. And adoption stands on three legs. It needs cheap and abundant power. It needs cheap and patient capital. And it needs the trust of the people who are supposed to live with the result. Hold those three things in your mind - power, capital, trust - because the rest of this is simply the unhappy exercise of watching the current approach kick out each leg in turn, and then, in the matter of Iran, do something worse.

Begin with the least glamorous fact in the entire field, the one almost never spoken aloud in the speeches: the binding constraint on artificial intelligence is not genius. It is electricity. The frontier is not decided in a lab by a brilliant insight; it is decided at a substation, by whether there are enough electrons to run the machines. We are entering the first sustained surge in American electricity demand in a generation, and it is being driven by computation. The race, in other words, is a race for power in the most literal sense.

Now consider the energy policy we have chosen to run alongside it. The current posture treats wind, solar, and battery storage as objects of cultural contempt - as artifacts of an ideology to be defeated rather than as sources of generation to be counted. Support for them is being stripped out; the rhetoric treats their retreat as a victory.

I want to make the apolitical case here as airtight as I can, because this is exactly the kind of point that gets dismissed as partisan when it is nothing of the sort. Suppose you do not care about climate change at all. Suppose you find the entire subject tedious and suspect. It does not matter. The argument does not depend on it. Solar, wind, and storage are, right now, the fastest and cheapest new generation it is possible to bring online. A new gas turbine faces a multi-year queue for the turbine itself. A new nuclear reactor is a project of a decade or more. The renewables-plus-storage that the current posture is busy marginalizing are the megawatts that arrive soonest. To push them aside on cultural grounds is not an energy philosophy. It is a decision to have less power, later, and at a higher price - at the precise moment the country has declared that power is the thing it needs most.

And there is a bill, which arrives at an address worth noting. As enormous data-center loads are added to regulated grids, the cost of building out that capacity does not stay with the technology companies. It flows, through capacity charges and an expanding rate base, onto the electricity bills of ordinary people’s households. A policy sold to the public as strength ends up quietly taxing the public for the privilege of someone else’s compute. That is not a detail. It is the beginning of the trust problem we will come to shortly.

The second leg is money, and here the contradiction is even sharper, because it is so easily measured.

The cost of capital is the floor beneath every data center poured and every model trained. Raise that floor, and you raise the hurdle that every project in the country must clear. This is not an abstraction; it is arithmetic that compounds across hundreds of billions of dollars of planned investment.

So, what have we done to that floor? We have raised it, deliberately if not intentionally, through a fiscal and foreign policy that manufactures risk. Open-ended military commitments. A federal deficit that ran to roughly $1.2 trillion in just the first half of this fiscal year, against a national debt that has crossed $39 trillion. And a pattern of treating long-standing allies as adversaries to be shaken down - these are the same people who price American debt and have noticed. The result is visible on any screen. The 30-year Treasury yield has pushed to around 5.25 percent (as of this writing) - its highest since before the Great Financial Crisis. A 30-year mortgage now costs more than 6.6 percent, up from under 6 before the war in Iran. The market recognizes what it is bracing against: a second wave of inflation.

State the contradiction plainly, because it is the heart of the matter. You cannot declare a national priority that depends on the cheapest possible capital and simultaneously run a fiscal and geopolitical policy that produces the most expensive capital you have had in two decades. These are not separate stories filed under different headings. They are the same story. The risk premium you create on Monday is the hurdle rate your AI champions confront on Tuesday.

And it lands unevenly, in a way that should worry anyone thinking about financial stability rather than headlines. The young companies financing GPU clusters through private credit pay this tax first and most brutally; their cost of capital moves the instant the term premium does. But the giants are no longer exempt either. The hyperscalers, having spent the early phase funding the build-out from their own cash, have begun to borrow in earnest - something on the order of $182 billion in data-center debt was issued last year alone, a great deal of it routed off the balance sheet into special-purpose vehicles and the bonds of others, so that the obligation sits with a pension fund or an insurer rather than with the iconic company. The risk does not vanish when you do that. It simply changes whose retirement is exposed to it. We are building the most capital-intensive project in modern industrial history on increasingly expensive and increasingly diffuse credit and calling it ‘market confidence’.

The third leg is the one most easily mistaken for a soft concern, which is why I want to insist on its hardness. There is, at present, essentially no coherent federal framework governing artificial intelligence. This vacuum is defended as freedom - as the wisdom of not strangling a young technology in its crib.

I understand the appeal of that argument, and I will return to its legitimate core. But consider how the vacuum actually reads to the American people. It does not read as freedom. It reads as: no one is minding the store. No one has decided who is liable when the system fails, who pays when the grid strains, what happens to the person whose livelihood the technology plausibly dissolves. And here is the mechanism that the cheerleaders consistently miss: people who do not trust a technology do not adopt it. They slow-walk it. They organize against the data center proposed for their county. They punish, at the ballot box and in the marketplace, anyone associated with having imposed it on them.

This is the part that should be stated without any euphemism. Trust is not a tax on adoption. Trust is the precondition for adoption. A population that feels unprotected and unconsulted becomes, in aggregate, the single largest brake on the very deployment we claim to want. The absence of rules does not accelerate the build-out. Past a certain point, it stalls it, because it forfeits the consent that large-scale deployment requires. This is not sentiment. It is mechanics.

The three failures above are expensive. They are also, in principle, recoverable - a posture can be reversed, a budget repaired, a framework written. I have left the Iran escalation for last because it belongs to a different and graver category. Its errors are not recoverable. They are the kind that reset the board for a decade, and they happen to bear directly on the same three inputs, only abroad and in China’s favor.

Start with energy. The disruption of the Strait of Hormuz has taught every oil-importing nation on earth the same lesson at the same time: dependence on a single chokepoint, mediated by an increasingly unpredictable security guarantor, is intolerable, and the way out is to electrify - to build solar, wind, and storage as fast as possible. That is now the global project. And the supply chain for that project, the manufacturing of nearly all that hardware, is dominated by China. We have, through a war, manufactured the largest clean-energy boom in history and arranged for our chief rival to be its principal supplier. The instrument we are sidelining at home is the instrument the world will now buy from Beijing.

Then there is the question of the compelled bedfellows. I want to be careful here, because the careless version of this argument is wrong and the careful version is damning. It is true that the war and the sanctions architecture around it press Russia and China into closer alignment. But the much-touted Power of Siberia 2 pipeline and a deal for it amongst Russia and China, is not the proof of it that some claim: their May summit produced an understanding on the route, not a finished commercial deal, and China is still hedging and driving a hard bargain on price, as it always does. The structural fact is not the pipeline. The structural fact is the alignment of intent - and the slow, deliberate draining of the dollar’s privilege at the margin, as more of this energy trade is settled in yuan. That privilege of the US – as the default world currency - took eighty years to build. It is the cheapest strategic asset the United States possesses, and it does not come back once spent.

And then there is the sharpest error of all, the one that would be implausible in a novel. We are confronting China while standing inside its supply closet. The advanced chips this entire program depends on, and a great deal of the US weaponry the Iran (and other) war itself requires, are built with rare-earth elements that China overwhelmingly controls - not merely mines, but refines and turns into magnets, a capability the rest of the world cannot replicate for years. The harsher of Beijing’s export controls are not gone; they are suspended, on a clock that runs out this November. And the lever already works: American imports of yttrium oxide, used in the coatings of jet-engine turbine blades, collapsed by something like 83 percent in a single month this past spring. This is not strategy. There is no framework of statecraft under which you provoke the one supplier you cannot replace, in the middle of the build-out that needs them most. It is the behavior of someone who has confused the feeling of strength for the fact of it.

Pull the four threads together and notice what does not connect them. There is no need to impute bad faith to anyone, no need for a villain. The indictment requires only that you add the policies up and look at the sum. Taken together, they raise the cost and shrink the supply of power, capital, and trust - the three things the enterprise runs on - and then hand our adversary the inputs and the alignment besides. That is not a disagreement about values. It is an engineering failure with a flag planted on top of it.

It would be unfair, and worse, unhelpful, to catalog all this without saying what the competent version looks like. The encouraging and slightly humbling truth is that the fix is not exotic. For the most part it consists of declining to sabotage oneself. If you genuinely mean to win, here is what that means, in the same order.

On power, adopt one principle and apply it without favor: an electron is an electron. Build everything that deploys quickly - solar, storage, wind, gas, nuclear, and above all the transmission lines that connect them - and reform permitting evenhandedly rather than weaponizing it against the technologies you find culturally distasteful. Take energy out of the culture war entirely. This is the single highest-return pro-AI policy available, and it costs nothing but the willingness to stop scoring tribal points against the cheapest megawatts on the board.

On capital, understand that the cheapest possible subsidy to the build-out is a lower risk premium, and that you do not purchase it with appropriations. You purchase it with credibility - with a fiscal trajectory that bond markets can believe and a foreign policy they can predict. Stop manufacturing the premium you then complain about paying. And treat allies as what they actually are, which is the cheapest force-multiplier in the history of statecraft, rather than as marks at a table.

On trust, write the minimally, credible framework that a serious country would write: clarity on liability, basic transparency, and an honest answer to the question of who pays for the power. Here I will concede the legitimate fear directly, because it is legitimate. Europe has regulated itself into a torpor from which it may not recover, and that is a real warning, not a strawman. But the cure for too much regulation is the right regulation, not none. A framework is the on-ramp to adoption, not the brake on it. Trust is infrastructure. You build it on purpose.

And on the matter of strategy, the discipline is simply to sequence rather than to stack. Secure the inputs before the confrontation. Build and stockpile allied rare-earth processing first; keep allies inside the tent; and internalize the asymmetry that the Iran episode has already demonstrated - that every act which pushes an adversary toward self-sufficiency is a permanent loss, paid in a currency no election returns. You do not open a second front while still dependent on the first.

Return, at the end, to where we began. The genie is out of the bottle. We are going to build this technology; that is settled. The only question still open - the only one that was ever really open - is whether we will build it competently? And the honest answer, given the policies currently in force, is that we are not presently trying to.

I have taken some care to keep this out of the usual channels of right and left, because the usual channels are precisely how an argument like this gets neutralized. This is a matter of right and wrong - it was never about which tribe you belong to. Are we a country that matches its means to its ends, or one that has mistaken posture for strategy and volume for resolve? We have told ourselves that confidence is the same thing as competence. It is not, and the difference is not academic.

The most important race of the century will not be won by the loudest competitor. It will be won by the most coherent one. Presently, we are neither - and the bill for confusing the two does not arrive on the convenient schedule of an election cycle. It arrives, eventually and without appeal, as a smaller country.

Disclaimer: This writing reflects my own personal views, not those of any entities or universities that I am affiliated with. My bias? I am committed to making high risk-adjusted after-tax returns in markets, and have significant market exposure to AI in my public and private markets investments (if you have a diversified portfolio, you share this concentration). As mentioned, it’s not relevant whether you like it or believe in the margin producing benefits of AI for US business - we are where we are and have made the bets we have, given the alternatives. But we must recognize the folly of believing the policies of America, as currently articulated and executed, are best suited to deliver on my (our) investment objectives as shaped by AI.