World War AI

How's that whole golden age thing going for you so far? That golden age of human leisure and wealth awaiting us in a world optimized for the thinking machines.

Are you working a bit less today, enjoying the early fruits of all this 'AI productivity'? Or are you somehow working longer, more stressful hours than ever?

Is it your sense that life is getting a little bit easier for the poor or the middle class or anyone other than the very rich as the 'AI revolution' arrives? Is it your sense that young people are a bit more hopeful about the future now that it's an 'AI economy'? Is it your sense that 'AI friends' are beginning to enrich our social lives? Is it your sense that goods and services are becoming more plentiful and cheaper as 'AI deflation' kicks in? Is it your sense that news is more informative and shows are more entertaining as 'AI content' spreads? Is it your sense that job prospects are improving as we enter an 'AI employment boom'?

Yeah. Same.

Honestly, I don't see how the carrot was ever going to work. It's just too at-odds with our actual lived experience, even here in Fiat World where our reality is declared and announced to us. They're going to need the stick. They're going to need to tell us that national survival is at stake, that our enemies will triumph if we don't make the 'necessary sacrifices' to win this 'AI arms race'.

They're going to need a war.

Oh, maybe not an actual war, but the functional equivalent thereof, full of threats real and imagined and adversaries foreign and domestic. They're going to need World War AI.

We're already seeing it in our narrative tracking data on Perscient Pro, this growing drumbeat for World War AI. From our AI Pulse report on November 10:

The Techno-Oligarchs know what they're doing. They've got their Wall Street Renfields to dangle the bags of money and their Washington stooges to play the patriotic duty song. You will know them by their fruits:

Bah! Yes, China is our geopolitical adversary. No, this is not an AI arms race or an AI war.

First we see through the stories we are told. Then we take the fight to our true enemy.

Clear eyes, full hearts, can't lose.

The United States spent $296 billion over a roughly four-year period to fight World War II, which would translate to about $4 trillion in today's dollars.

At its peak (1943), the war effort accounted for 37% of US GDP, and no aspect of American life was untouched or unconstrained by the US government's reallocation of the three basic building blocks of economic activity -- labor, capital and energy (energy being my shorthand for all physical resources as well as the core input to mining, farming, manufacturing and transportation) -- and the enormous expansion of government's role in American society to carry out this reallocation. In particular, every aspect of consumer behavior was subordinated to the political will required to execute the war effort, a political will which created extreme shortages in the labor, capital and physical resources available to the consumer economy.

I think it's hard for Americans today to grasp both the level of consumer sacrifice that was required during World War II and the level of government propaganda 'nudge' involved in enforcing that consumer sacrifice.

I mean, I'm guessing that the mother and child in the poster above, dressed in their perfectly matching frocks and radiating Stepford Wives aura, maybe did not have enough food the winter before? And if you think that it's 'encouraging political violence' to call someone a Nazi today for supporting fascist policies ... in 1943 the government would call you a Nazi if you didn't carpool.

I find these posters and broadsides from World War II pretty funny, like they're from some cartoon world, and I bet you do, too. But when you read the memoirs and economic histories of the WWII homefront, there's nothing cartoonish about it. These were hard times! Shortages of food, energy and labor created extreme cost-push inflation, like our Covid-era supply chain inflation but on steroids, to which the government responded with draconian price controls on EVERYTHING. And when price controls didn't work, meaning that when even a suppressed market failed to distribute enough calories to enough people to prevent widespread hunger if not starvation, the government abandoned market mechanisms altogether and instituted outright rationing on food, energy and other necessities.

At the same time, every bit of available domestic investment capital and savings (which are the same thing) was absorbed by the federal government and unavailable for the consumer economy. That meant that in addition to the extreme inflationary pressures from widespread shortages, there was ZERO economic growth from small and medium businesses, which were an even larger portion of American GDP back then than they are today. The only thing that kept the American economy from collapsing into a stagflationary disaster was the $4 trillion that the US government spent on manufacturing war materiel and -- hold this thought! -- the enormous number of new jobs created from that.

The same amount of inflation-adjusted money we spent on World War II -- somewhere between $4 trillion and $5 trillion -- is scheduled to be spent on AI and datacenter buildouts in the United States over the next four years.

Yes, our economy is proportionally bigger today, so this is 'only' something like 15% of US GDP ($30 trillion in 2025), but an economic mobilization of this magnitude will require a similarly massive reallocation of our fundamental economic building blocks -- labor, capital and energy -- especially capital and energy.

On the capital side, it's difficult to communicate how much money this is over such a short period of time. As JPMorgan puts it in their magisterial research note on AI Capex financing, "The question is not which market will finance the AI-boom. Rather, the question is how will financings be structured to access every capital market.” Here's their chart for where they think the money will come from (slightly apples to oranges as this is global spend, not just US, but I figure 70-80% of this datacenter build is going to happen in the US, so it's essentially the same), and I'd call your attention in the $1.4 trillion attributed to "Need for Alternative Capital / Governments", which combines both our favorite financial topic du jour -- private credit -- with direct government subsidy/investment.

This is the necessary context for understanding OpenAI CFO Sarah Friar's recent comments at a Wall Street Journal conference that the company would 'welcome' a federal government 'backstop' on private debt financings of this datacenter buildout, as well as Sam Altman's unintentionally hilarious 5,000 word tweet to 'clarify' Friar's very clear and very correct and very intentional words.

Sarah Friar didn't 'misspeak' when she called for a federal backstop -- by which everyone means and intends a US Treasury guarantee -- on AI datacenter debt issuance, and she didn't need to 'phrase things more clearly'. She used exactly the right word to describe exactly the policy that OpenAI and Wall Street and every other participant in this $10 trillion ouroboros ecosystem desperately wants and frankly requires for this massive reallocation of capital to have a chance of succeeding.

I mean, a federal debt backstop is just the start. Within a couple of years -- and this is the point of the $1.4 trillion "Alternative Capital / Governments" item on the JPMorgan chart! -- the US government will need to allocate hundreds of billions of dollars directly to the AI buildout, maybe through defense appropriations, maybe through equity stakes, maybe through whatever. Otherwise, we're a good trillion dollars short in the funding required to make this work here in the US. All from additional borrowing and deficit spending, of course, just like in World War II when the federal debt skyrocketed to an amount that was 100% of GDP. What's different today, of course, is that the federal deficit is already at World War II debt-to-GDP levels before the additional borrowing for the AI buildout support. Bottom line: whatever you think the future path of US debt-to-GDP looks like, you're too low.

The economic term for the impact of capital reallocation at this enormous scale is 'crowding out'. The public and private capital that is invested in or lent to the AI hyperscalers and their counterparties over the next four years is that much less public and private capital available to be invested in or lent to the rest of the economy. And while I'm sure most large B2B enterprises will find a way to at least get a taste of what's being poured into the AI buildout, small and medium enterprises will be mostly shut out and consumer-facing enterprises are going to be completely shut out.

The inevitable impact of a massive reallocation of capital away from the consumer economy is that consumer credit becomes more expensive (if it's available at all), capital-intensive consumer services like health insurance and homeowners insurance become more expensive (if they're available at all), consumers stop spending (especially the bottom 50%), and consumer-facing businesses stop hiring (if they're not actively cutting back).

Sound familiar? That's because what I'm describing isn't some maybe-projection of some hypothetical future. This is all happening already. This is all happening NOW.

This is the latest data from the NY Fed consumer credit access survey, which they update every four months and goes back about a dozen years. Credit applications regardless of type (credit cards, auto loans, mortgages, mortgage refis) are being rejected at the highest rate since they started the survey. Mortgage refi applications (so these are not the poors, these are middle class and up consumers who own their home) are being rejected at an especially high rate, north of 45%. This is what a consumer credit freeze looks like -- already! -- here in the first inning of the AI capital reallocation.

As for consumer spending ...

These charts show the drivers of US GDP growth for the first half of the year. The chart on the left shows that spending on tech accounted for almost half of US GDP growth in the first half of the year. The chart on the right shows that growth in tech spending (which is about 5-6% of the US economy) accounted for more GDP growth than all consumer spending (which is about 60% of the US economy). And this is only through June 30. We'll get Q3 data at some point in the next month, and I'll eat my hat if these trends aren't accelerating, even here at this earliest stage of World War AI.

But wait there's more. What I've described so far is what happens with a war-footing reallocation of capital. The impact of a war-footing reallocation of energy is even harder on the consumer economy.

Once again, I'll turn to my friends at JPMorgan to set the stage. This is projected datacenter electricity consumption globally (of which the US is the largest portion, but less so than in projected capital spending). I think it's quite thoughtful and is a good balance of top-down power consumption estimates (which tend to be lower because from a top-down perspective of how many power plants can be built there's just no way) and bottom-up power consumption estimates (which tend to be much higher because the appetite for more compute is limitless). What you're seeing is annual terawatt-hours (TWh) of electricity consumed, which is the Godzilla-version of the same sort of kilowatt-hours information you'd get on your utility bill. I really like starting with the consumption projection -- how much electricity will all these data centers need -- before we start looking at where the electricity might come from.

JPMorgan is estimating that datacenters (globally) will consume about 1,100 TWh of electricity in 2028. My guess is that 70% of that electricity consumption will come from US datacenters, up from 60% in 2023 as the new construction will be predominantly in the US. Roughly speaking, let's say it's 800 TWh of projected electricity consumption by US datacenters in 2028. If that's growing by 25% per year (which would be a reduction from 2027 and 2028 growth levels of 32% and 29% respectively), then we're at US datacenter electricity consumption of 1,000 TWh in 2029 and 1,250 TWh in 2030.

To level-set a little bit, in 2023 the entire US economy consumed approximately 4,000 TWh, per the US Dept. of Energy. Of that total, datacenters consumed 175 TWh, or right at 4.4% of the entire economy.

Now in a good year, the United States grows its total electricity consumption by 2-3%, because electricity consumption is an effective proxy for GDP growth. So let's be generous and take non-datacenter electricity consumption and grow it by 2% per year, and see where we end up in terms of relative power consumption between datacenters and the rest of the US economy. We're not even thinking yet about whether or not it will be possible to build enough new power plants to satisfy both datacenter growth and non-datacenter economic growth. I'm just curious what the mix of electrical demand will be.

With 2% electricity consumption growth (i.e. economic growth) in the non-datacenter economy, datacenters will have a 15.9% electricity consumption share of 5,023 TWh in 2028, 18.8% share of 5,308 TWh in 2029, and 22.1% share of 5,644 TWh in 2030.

With 1% electricity consumption growth in the non-datacenter economy, datacenters will have a 16.6% electricity consumption share of 4,820 TWh in 2028, 19.8% share of 5,060 TWh in 2029, and 23.4% share of 5,350 TWh in 2030.

With 0% electricity consumption growth in the non-datacenter economy, datacenters will have a 17.3% electricity consumption share of 4,625 TWh in 2028, 20.7% share of 4,825 TWh in 2029, and 24.6% share of 5,075 TWh in 2030.

Up from 4.4% of 4,000 TWh two years ago.

That's a lot of numbers, but it's a simple conclusion:

The datacenter slice of the US electricity consumption pie is growing much larger and much faster than the pie itself can possibly grow.

Datacenters go from a rounding error in 2024 to consuming close to one-quarter of our electricity by 2030.

It is impossible to build enough new power generation supply to support both projected datacenter growth and growth in the rest of the US economy.

Ignore interconnects and transmission and all that stuff. Ignore the 3-year wait for new gas turbines from Cat or GE. Ignore all of the wind and solar projects that this administration has killed. Pretend -- and it's totally pretend -- that we can add 100 gigawatts or whatever the number is of new power generation to support the required electricity demands of the AI 'war effort'. It still doesn't work.

Even if you succeed in building all the power generation facilities required for the AI datacenters, there's a remaining deficit of hundreds of terawatt hours for the electricity consumption (i.e. economic growth) of the rest of the economy. That deficit cannot be jawboned away. That deficit must be rationed to the rest of the economy, first by price -- which is why your utility bill has gone up so much already -- and then by allocation, i.e. scheduled brownouts and price controls.

The kicker, of course, is that higher electricity prices filter into everything. This IS cost-push inflation, and the same story arc from World War II -- first rationing through price increases, then rationing through allocation and price controls -- will repeat itself in World War AI.

But here's what's different.

Unlike World War II, there will be no net new jobs created by World War AI.

Most of the readers of this note work for a non-tech company that is 'trying to figure out how to use AI'. Let me guess how that's going for you. At the vague direction of your board or CEO or EVP, your company has spent a not-small amount of money on AI 'projects' that started with some sort of chatbot running over internal documents or knowledge base and have now graduated into some sort of pseudo-agentic junior analyst or junior coder or copywriter or update provider. This has been going on for about two years now. There have been some modest process improvements in your products or services which, over time and over enough processes, may or may not add up to noticeable margin expansion. The jury's still out on that. AI has allowed you to delay or scale back some hiring plans, which is definitely a plus for your bottom-line, but you haven't seen much impact from these AI projects on your top-line revenue growth.

I'm sorry to say this, but that's all there is. AI can generate process improvements and margin expansion, principally by reducing headcount, but it's not going to open up new revenue opportunities or create new products and services. That's just not what inference does. AI doesn't discover new economic vistas into which human cogs may be inserted/hired as 'jobs'. AI takes existing economic activity and replaces human cogs with itself.

In fact, the best way to understand World War AI is that the 'war aims' are not directed at another country, no matter what you may hear about China, but are directed at human labor itself. The goal here -- not the unfortunate side effect but the intentional goal! -- is to eliminate much of human labor, both white collar / symbolic manipulation jobs today with AI virtual agents and blue collar / physical manipulation jobs tomorrow with AI-supported robots. Or in the words of Elon Musk, in the same speech where he accepted his $1 trillion pay package from Tesla:

The Techno-Oligarchs and their stooges intend to optimize our world for the thinking machines over humanity.

They're not confessing. They're bragging.