The Shrugflation Triangle: How AI, Oil, and Treasuries are Impacting the Economy

The Shrugflation Triangle: How AI, Oil, and Treasuries are Impacting the Economy

why everything is getting more expensive and why some simply don’t care

kyla scanlon

Oct 14, 2026

595

71

110

Share

Good morning from Tulsa, Oklahoma!

This piece was written in several cities - Durango, Colorado, Chicago, St. Louis, Las Vegas, New York City, Cleveland, and Tulsa, Oklahoma. I love traveling because I get to talk to so many different people about how they experience the economy, and I’ve told some of those stories here. The goal of the piece is to connect a lot of dots that tend to float by one another, and to hopefully provide a comprehensive explanation of the three core factors in the economy right now – AI, Treasuries, and oil – and talk about how they feed into one another.

Here is the audio version:

0:00

-56:31

Audio playback is not supported on your browser. Please upgrade.

ALSO NOTE:

This piece is really long. So it’s likely cut off in your inbox. Right now, my plan for this Substack is to publish one very in-depth essay about once a month. I find that it allows me to spend more time within the ideas and make sure that I am publishing something that feels worthy of your time and attention. So to those who have been so supportive (and so patient between pieces), I am very appreciative - more to come!

OTHER THINGS:

The paperback

version

of In This Economy? came out!! Yay! It has a new afterword about tariffs and AI. I partner with Skylight Books if you want to support

local

as well!

I am also working on a new

website

called Ask Kyla where you can submit any questions that you have about the economy and markets.

This is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

PART 1: LET’S PLAY SLOTS

In Vegas, very early in the morning (although there really isn’t such a thing in Vegas) I was getting a coffee. The barista asked if she could make my coffee without the machine (a touch screen machine was humming behind her, where you can pick exactly what drink you want to make from a menu of pictures.) She said that she liked to add a “human touch” to the coffee. I said yes, of course.

I couldn’t sleep. Neither could the people peppered around the casino playing slots, all of us floating in a world where time didn’t matter.

Bleary eyed, drinking my human-made coffee, I watched them install four new slot machines, absolutely enormous, reaching into the ceilings, like Easter Island heads. The slot machines have screens, so for a while, as the workers were setting them up, lines of raw code danced across the uncalibrated screens, a brief otherworldly glimpse into what would soon be hypnosis. Then the display flipped to three cherries in a row, and played a bright, celebratory chime. A jackpot. But there wasn’t a player in that chair. Everyone else was crammed around the other machines down the row, still feeding in their tokens. Still hoping. Still waiting.

It’s a decent model for the American economy right now: a room full of people playing the slots, trying to get rich, and the jackpot hits for an empty seat.

Most people are trying to do everything the economy asks of them. They’re working and spending and paying more for gas and to borrow money. But the economic winnings keep showing up somewhere else. So where are they going?

THE TRIANGLE

There is a lot of debate, huge debates, over what is going on and why. What’s causing the rise in Treasury yields (the thing that prices our economy!)? What’s causing inflationary pressures? What’s causing the economy to do anything, really? Huge debates, with people saying

no it’s AI spending, it’s the sheer amount of debt, it’s the war, it’s Uncle Phil in Chicago, he actually controls the economy from an old Nintendo 64.

It’s everything, all at once. It’s oil, AI, and Treasuries, because they are feeding into each other. It’s a Triangle (a Bermuda Triangle as Coby Weiss puts it) - three points on a map.

Here is how the three connect:

Oil

pushes up rates: the war with Iran has throttled oil shipping, diesel has spiked to record prices, and everything that rides on a truck or train or ship or plane costs more. That’s a big part of why the

Fed has had to hike interest rates

The government happens to owe the size of the entire economy right now in debt, so when rates go up, so does their interest bill, the fastest-growing thing in the budget. The government covers that by selling more Treasuries, the thing that prices our economy at the same time the

AI industry is borrowing unprecedented sums

from the same pool of money.

Two giant borrowers, one pool of money, the sheer demand can push up long-term rates even more. And they shrug at any change in rates! The government borrows to pay bills. AI borrows to build, and that requires a ton of power.

AI needs energy

, in a world where energy is really expensive. Data centers need a huge amount of power, and when the grid can’t deliver it, they burn it. Oracle is

trucking natural gas to data centers

. AI doesn’t really move the price of oil but it can’t turn on without it, which brings us back to Point 1.

So that’s the triangle. And who has to deal with it? People. People have to deal with the higher borrowing costs and the higher energy costs and the uncertainty about what AI will do to them. And no one wants to slow down because AI investment is a huge share of both the economy and the stock market, bolstering household wealth, so if the party stops, it reverberates across the entire economy.

It’s a form of

shrugflation

(can’t help a portmanteau) – usually, money gets expensive and people borrow less, right? Gas gets expensive, you drive less. That’s the whole thing. But right now, a war has made energy expensive and the Fed can’t do anything about that, because you can’t go to a Strait and say listen buddy I’m hiking rates over here. All you can do is make borrowing cost more.

And the two biggest borrowers in the economy hear that and just sort of … shrug? The government borrows at whatever rates are. The AI companies feel like they have to keep borrowing in order to survive, what Tracy Alloway

calls

the “existential imperative to borrow.”

What does this look like across the economy? Let’s start with oil.

PART 2: OIL AND THE STRAIT JACKET

At a finance conference a couple of months ago,

the FT

quoted a banker who said “Does anyone really care if the Strait of Hormuz is open?” perhaps a defining quote of this era. And to this bankers credit, no. The government seems to be only complicating the situation and AI doesn’t seem worried, at all. The only people who care are

people

.

I had the honor of moderating a

panel

on consumer sentiment at the Chicago Federal Reserve. The panel was extraordinarily brilliant and the room was full. One of the panelists talked about how consumers are getting “double squeezed” because they have to absorb both (1) high diesel prices through a higher cost of goods and (2) high gas prices to even go get those goods. The room nodded along.

And they nodded because people

care

. On October 9th, the University of Michigan’s

consumer sentiment index

came in at 46.3, the second-lowest reading in its history (the lowest was in May of this year). Its measure of how people feel about their current conditions is the lowest ever recorded

1

. People expect prices to rise 4.7% over the next year (not great for the Fed).

I’ve been inadvertently watching where all that diesel goes. A big portion of my current road trip has been driving. I drove from Albuquerque to Durango (and back), and from Oklahoma City to Tulsa. All three of these adventures happened between 10pm and 2am. So during those hours, it’s me and a bunch of truck drivers.

You can see the hidden machinery of the country so clearly. Trucks hauling other trucks, hauling cars, hauling horses, hauling food, hauling anything and everything. There are so many of them - on my 90 minute drive from OKC to Tulsa, which started at 12am, I saw about 100 trucks cruising along the interstate. All of them carrying the economy behind them. All of them are incredibly reliant on diesel.

WHY IS GAS SO EXPENSIVE?

High energy costs trace back to the war with Iran. Since the war started, shipping through the Strait of Hormuz (what a fifth of the world’s oil usually goes through) has been throttled.

Saudi Arabia had a

side chute

that lessened the blow, with a pipeline across the country into the Red Sea, but drones took that down too. For about two weeks, the main road

and

a detour were both closed.

The pipeline is running again, below capacity, and the Strait is

running

at 80% capacity, but only unrefined crude oil is getting through (and Iran is

stepping up attacks

on tankers).

Renting a supertanker to carry oil from the Persian Gulf to Asia now costs about

$1.4 million a day

according to Bloomberg. Pre-war, it was about $30,000.

A logistics guy summarized it beautifully for the

WSJ

– “chaos increases prices.” Our entire economy is truly an oil guzzling machine, as both gasoline and diesel come from oil, with things going all places all the time, across trucks and trains and tractors and ships.

If it gets more expensive to get things places, things are going to get more expensive!

Diesel hit a record $6.53 a gallon in September, up 77% in a year, according to

the Wall Street Journal.

So governments are trying to free up fuel reserves. The G7

agreed to release 100 million barrels of diesel and crude

from their reserves over four months. That’s only about a day of what the world uses, according to the

New York Times

. We use SO MUCH FUEL. And the thing about reserves is that they… tend to run out.

Expensive fuel makes everything more expensive, which creates inflationary pressures, which is the Federal Reserve’s problem. To combat this, the Fed

raised rates

on September 16th, for the first time since 2023. But the Fed can’t go march down to the Strait of Hormuz and broker a peace deal. They can’t direct ships through. All they really can do is make borrowing more expensive for everybody, so that way, people slow down, spend a little less, the economy therefore slows down, and hopefully, inflation slows down.

HOW SHOULD PEOPLE FEEL ABOUT HIGH GAS PRICES?

And we were recently reminded the importance of the Fed’s role in setting expectations like this! The

Nobel memorial prize in economics

went to Robert Hall and Michael Woodford for work showing that what households and businesses

expect

is a big part of what holds an economy steady. Shaping those expectations is a big part of setting policy.

Fed Chair Warsh has

put aside forward guidance

, one of the tools his predecessors used to steer those expectations. So we have an energy shock, people expecting prices to rise almost 5% over the next year, and a Fed Chair that is trying to pull back on some element of signaling where it’s going. Expectations matter. How people feel matters!

The expectations are already in action. Cleveland Fed President Beth Hammack spoke at

NABE

, a conference I was attending, and she mentioned that a retailer in her district told her that they’re planning to raise prices by more than their costs went up, because they figure more inflation is coming from somewhere – and that’s how inflation becomes self-fulfilling. So it’s the beginning of the loop in a way — what people expect to happen can be what happens, which is why it’s important to discuss people as a part of this triangle.

Oil prices and the 10-year Treasury yield are also

moving more closely

than at any time in 35 years, according to CBOE, which means that yields are driven heavily by whatever is going on with oil. Which is a problem for the one industry that needs to both (1) borrow a bunch of money and (2) use a bunch of energy.

PART 3: AI AND THE CAPEXISTENTIAL CRISIS

SHOULD EVERY AI COMPANY AUTOMATICALLY GET $2 BILLION?

AI is a ghost of sorts, haunting most rooms, with people telling stories about it that sound like the scary ones that we tell over a campfire. It has made the future look bright for some, but certainly, not for all, at least not yet (if ever?)

But the price of the companies making an AI product keep getting bid up. Instinct, a personal AI assistant (these assistants are an agent, AI that has access to the computer) has an interesting product but they make $0. They have about 100,000 users or so. They, naturally, raised $1 billion on a $10 billion

valuation

(most of which they plan to spend on compute, the thing that powers the AI, which is in increasingly shrinking supply.)

And of course this product that makes $0 raised one billion dollars! The venture capitalists that wrote the checks were probably like wow, I can text it and it can text my friend and we can all get dinner Tuesday at 7:30pm and also I have a meeting right now what is it about oh it’s this other AI company wow I love this assistant give them one billion no actually… make it two billion.

This is the shrug in its purest form and the race is on. Carl Benedikt Frey, an economist at Oxford, told the

NYT

that, “these companies are essentially in an AI arms race. If they don’t invest, they are acknowledging defeat.”

IS THIS THE SCAMINGULARITY?

Crypto’s original theory was

“what if we had peer-to-peer transactions and there wasn’t a financial middleman”

and within a little over a decade, the technology had evolved into a way to talk about the speculative value of a picture of a monkey –

“ape holders can use

multiple slurp

juices on a single ape.”

And in AI, one company set out to redefine Hollywood filmmaking and now they help people make, in

Ryan Broderick’s words

, “hyperrealistic One Piece character that can promote your crypto scam on Instagram.”

One might say it’s the singularity of SCAMMING! The scamingularity, if you will!

It’s not totally clear to me why everything becomes sloppy and scammy. Survival of the scammiest? Market forces, an economy that rewards growth at any cost, at any price, and preys on people’s gaps in understanding those forces? How it has always been and how it will always be?

Bloomberg has one answer

here

. When something gets cheap, the people with the lowest standards tend to be first movers on the product. So a cybersecurity firm found that phishing pages built on free AI app builders are up more 4x in two years because now everyone can will a scam into existence by typing “

Chat make me a bank website make it look really real and perfect.”

Reports found a training manual in an abandoned scam compound (scampound?) in Cambodia, where it’s advised that stories can be made up, but have to “show genuine feeling.” The human touch. Scammed.

Utopia is always the pitch, but slop is usually the final product.

And AI’s original theory is just as lovely as crypto’s was. “We will all live in a utopia and write poetry all day because AI runs our lives for us and everything will be free :) and that will be awesome” is a beautiful idea and I hope it comes true? Like genuinely, I hope that the visions come to fruition (perhaps not in the form of one billion AI assistants, but you know.) We should want a better world for those that come after us, always.

WHO GETS THE GAINS?

And we have always been dreaming of this world. John Maynard Keynes wrote

Economic Possibilities for Our Grandchildren

in 1930, a vision of a universe where he thought that his grandchildren would live in an economy so productive and so rich that they would only work 15 hours a week. Keynes said that the extra time that comes from having solved “the economic problem” would be used to solve a person’s “real, his permanent problem” which is “how to live wisely and agreeably and well.”

That’s not really happening. Keynes grandchildren are, according to BlackRock’s

read

of the Fed’s Beige Book “trading down” while living through “the divide between the capital-spending economy and the consumer economy is becoming more pronounced.”

An economy with extraordinary wealth, increasingly concentrated among

older

folks and already rich folks. Median net worth for families under 35 is down 23% since 2022. For families 75 and older, it’s up nearly 40%. But the gap really isn’t young versus old — it’s ownership versus not. If you rent, your median net worth is about $10,000. If you own, it’s over $450,000.

A utopia for some, but not for all. And AI is only complicating things more, mostly due to the wealth it’s creating for these two specific groups who “tend to be more invested in the stock market,” according to reporting from the

Wall Street Journal.

Most of the big ideas that have made society richer and more convenient over the last couple of decades, the ones nudging us toward perhaps some strip mall version of utopia, have made some people radically wealthy and have made most people’s lives a bit easier. But is the wealth evenly distributed? And can AI complicate that even more?

Owen Zidar and Eric Zwick went on

Odd Lots

to talk about their excellent new book the

Everywhere Millionaire

and they note that in 2001, the value add per worker to a company was about $34k, which went up to $52k in 2021. Owners got $15k of that $18k increase. The worker got $3k of the gains that the worker created. $3k!

That is the economy now, as I wrote about last

time

. This is not an economy for workers. This is an economy for capital owners. And AI, at least the way it’s working now, and is enriching a few people very quickly, while threatening those who have the least cushion to fall. It allows workers to automate tedious tasks (or themselves) and has polluted the Internet, our digital commons, our shared space that is owned by the public, with a lot of junk. Convenience while a select few reap the gains, across any industry, is not a utopia.

IS AI TAKING EVERYONE’S JOB?

To be fair, jobs themselves haven’t gone anywhere. The people who study the impact that AI has on the labor force say that the scary forecasts are usually mostly wrong. As Ernie Tedeschi wrote in a recent

Stripe Economics

piece, when factories got electricity in the 1920s, people predicted mass unemployment — but factory jobs grew for another 50 years. Postings for software engineers are up over the last two years, even though AI can now write code.

So the fight so far isn’t really about if there

is

work but rather, who gets

paid

for that work.

And that’s the whole thing. People don’t really want AI on these wildly unequal terms. 80% of voters think it’s very or somewhat likely that AI will cause widespread job loss, almost 2/3rds think AI could be a threat to humanity itself, and most folks want the government to do the job that the taxpayer pays them for and to regulate it, according to David Shor at

Blue Rose Research.

And it’s all moving very fast. As Vernor Vinge wrote back in 1993,

“The Coming Technological Singularity: How to Survive in the Post-Human Era”

and admitted that “sometimes I think I’d be more comfortable if I were regarding these transcendental events from one thousand years remove… instead of twenty.” Thirty years later, living in some form of the said-transcendental events, he does appear to be correct about many things

2

, including the philosophical battle that comes with it. Technological progress is never easy, and this one is no different.

COULD THIS THING BE MORE EXPENSIVE?

But the AI buildout needs people to really like it! Somebody has to buy what it makes! According to

Stijn Van Nieuwerburgh’s paper

on the AI buildout, the AI industry needs to bring in about 9% of GDP in AI revenue by 2032 for the investment to pay off. Greg Ip

asks

:

Is it plausible that Americans will spend as much of their income on this one technology as they do on food? Roughly twice what the nation pays for all forms of energy or all computers and software? Seven times what consumers spend on phone, streaming, and internet services combined?

Jared Bernstein

pointed out

that both Amazon and Google would have to build two brand new AI-specific Amazon and Googles, on top of the existing Amazon and Google that already rake in a bunch of cash, in the next 10 years, in order to justify some of these valuations. They would have to build a second, AI-only version of themselves within the next decade, just as profitable and successful as the companies that took 30 years to build the first time around.

WHO WANTS IT?

And then there’s who is supposed to be buying. As

Torsten Slok at Apollo

has written about, tech has a really rosy vision for its future, but the other silos in the economy, the ones that would pay for this tech, do not see such a rosy future. As he writes, this “raises the question of who exactly will be writing all those checks to buy AI services.”

That’s the other thing that a lot of people talk about but no one really seems to know how to answer? If you displace the entire labor force, is it going to be like 10 guys buying stuff via Meta Muse from each other forever? And that’s the AI industry?

Maybe - I was in St. Louis at

FREDCon

recently, where we talked about agentic AI and how we might have to write research for the AI agents to read because no one is going to want to read anymore because their agent is going to summarize it for them and then they will put that into a report that other people’s agents will summarize and that’s the future. I mentioned that I thought about it when I wrote, that maybe I have started writing for the machine.

But… hardly anyone is really paying for AI, much less using agents read various Substacks like this one.

About half of Americans have tried a chatbot, according to

Pew

.

Two banks checked their customers’ statements and found that only

2 or 3 in every 100 households

actually pay for one.

And even with that, the companies are already making more money than anyone thought they would be at this point. So you can see how much room they really do have to grow and make a ton of money. As Ip

notes

, OpenAI and Anthropic are growing at double-digits each quarter. Exactly how fast depends on who is counting.

HOW DO AI COMPANIES MAKE MONEY?

These companies don’t report sales the way a normal business does. They report a “run rate” meaning they take revenue from a recent month and multiply it by 12. A few weeks ago, the run rate going around for OpenAI was $70 billion a year. Recently, the

Financial Times

reported that the number OpenAI gave investors was closer to $50 billion (the stock market HATED this).

How do you lose $20 billion? A lot of AI gets sold through the big cloud companies, so Anthropic, which sells through Amazon and Google, counts the full price of those sales as its own revenue. OpenAI only counts the share it keeps. The $70 billion was OpenAI’s number redone via Anthropic’s way (that they didn’t deny when it first was floated?) and if you’re confused, so is everyone else.

OpenAI expects to burn almost $280 billion by

2030

. They are holding off on going public, but Anthropic did file their

IPO prospectus

, which was leaked. They made $4.6 billion in revenue and spent about $12.65 billion (half of this is compute and infrastructure costs) - and they have over $500 billion in cloud, computing, and infrastructure obligations!

They are expected to go public at a $2T valuation, which would make them in the top 10 most valuable public companies in the United States, despite their revenue being extremely tiny.

Maybe this will all work out. Maybe these companies will find a moat and the data centers will get paid back. Maybe the neolabs, the new AI labs that “lack products, markets or revenue” and who have in the past six months, have raised “$24 billion, nearly 5x as much as OpenAI and Anthropic did in the years up to the launch of ChatGPT” according to the

FT

, will turn into real businesses. Maybe it won’t become a commodity, something like oil and natural gas. Maybe.

IS AI A COMMODITY?

Chris Wood

at Jefferies thinks that’s exactly where this is going. A commodity is a product that isn’t really differentiated, so you can only compete on price. He wrote this week that AI is headed there, because Chinese companies are giving away models that anyone can download and run for free.

If customers can get nearly the same thing for nearly nothing, the American companies can’t charge enough to pay back what they borrowed to build it. He calls the likeliest ending “massive capital destruction” which is a polite way of saying financial implosion, which sure. The likeliest case for any sort of implosion is a slow unwind, sort of scraping the AI gum off the bottom of the economy’s shoe. As Ernie Tedeschi

wrote

AI capex being more & more financed by interest-rate sensitive means like corp bonds means you should at once be 1) less worried about an out of control “bubble” as there are more breaks that can prevent overheating, 2) more concerned about investment crowd out, eg less housing.

Not everyone thinks that China is the reason for this either –

Ara Kharazian at Ramp

says that the American companies are doing it to each other, cutting prices to win customers. Either way, what they can charge seems to be falling while what they owe is… not.

There’s also a whole world where everyone just develops their own models, which can disrupt the whole party.

Latham & Watkins

, a gigantic law firm, bought their own Nvidia GPU servers and started customizing AI models in-house, which keeps their clients data inside their own systems and gives them options if OpenAI and Anthropic raise their prices.

DEBTACENTERS (OR, HOW DOES ONE GET TRILLIONS OF DOLLARS?)

So how is all this getting paid for? With borrowed money.

AI is bigger than any investment, ever, and the companies building it are spending far more than they are making. So far the big boys have been able to pay for it because of who they are, not what they’re building. And that’s kind of the general feeling?

As Carlyle wrote in their

Rules of Relevance

, “more than 80% of anticipated AI capex through 2032 is expected to require external financing” noting that “lenders are effectively underwriting a handful of investment-grade counterparties rather than the economics of the hardware itself.”

And as Greg Ip

pointed out -

It was luck that the companies best suited to building out AI happened to be immensely profitable quasimonopolists. They could finance the first leg of the buildout internally without having to meet external ROIC tests. Though they are now issuing debt, that debt is backed by the hyperscalers’ non-AI cash flows, not the AI itself. In other words, the AI capex boom has been possible because the technology has never had to prove its profitability. That wasn’t the case with the TMT or other prior booms, which went bust when the capex ran too far ahead of the prospective profits.

What that means is lenders see Amazon and Google and say, “okay sure, they can make big money loans because they are Amazon and Google!” But even these companies are spending a lot of money. Capex, which is shorthand for what the companies are spending on long-lasting stuff like data centers, is starting to outrun how much money they are making.

The big 5 are expected to spend about $800 billion this year, roughly double last year. For the first time, that is about as much cash as their businesses generate.

Goldman Sachs has them spending

more

than they bring in this year and next.

So the companies are buying a ton of chips, on borrowed money. GPUs, the chips that stack to make servers, which stack to make data centers, depreciate quickly, in about 2 to 3 years. And each one is being swamped by new chips — Epoch AI estimates that there are about 20 million AI chips in data centers worldwide right now, a number that has been doubling roughly every nine months, which would put it near 200 million by the end of 2028, according to

the NYT

. That’s so many chips! They depreciate so quickly!

The other big investment booms were physical infrastructure like fiber from the Internet buildout and railroads from the railroad buildout. As Hanno Lustig

put

it, “rails last a lot longer than GPUs.”

If a chip is worthless in three years (unlike railroads) the money borrowed ends up outliving the thing that it paid for. The first big wave of these chips turns five and six years old between 2027 and 2029, and we will certainly find out more then.

PART 4: THE BOND MARKET

THE GOVERNMENT NEEDS YOUR MONEY RIGHT NOW

Fed Chair Warsh was asked at his September press conference why long-term rates keep rising. He said well, (1) the economy is strong and (2) there is

competition for capital

because “the so-called hyperscalers are out in the market raising funding” and (3) geopolitics. That’s the entire triangle.

Fed Governor Christopher Waller

said

that the bond market is doing “some of the work” of rate hikes for the Fed, by slowing down the parts of the economy that are sensitive to rates, which are like houses and cars, not data centers. He thinks the AI spending pressure could last two to three years, and the productivity payoff is a few years out. So. You know.

The government’s fiscal year ended September 30th, and it spent $7.4 trillion and took in $5.4 trillion, which is a $2 trillion gap, over $200 billion larger than last year, and the third largest EVER in dollar terms behind the two pandemic (pandemic!!) years. The debt is now about the size of the entire

economy

.

And people don’t want as much US debt as they used to, at least not at the old prices

3

, especially foreign buyers, who would much rather buy the shiny and beautiful AI stocks. As Brad Setser said, “it feels like the world is very high on US equity.” According to the

FT

, “overseas investors bought a net $188 billion of US debt securities in the second quarter, down from $314 billion in the first.” Some central banks are pulling back in other ways too. The Netherlands moved their gold closer to them, citing “increasing geopolitical unrest.”

And that’s the thing, it’s not just the US that is wrestling with all these problems.

The IMF put out its big

twice-a-year report this week

, and it said that wealthy countries are on track to borrow 5.2% of their economies this year, the most since 2021, when we were in a PANDEMIC. Everyone’s shrugging! And the IMF said that “growth is being reallocated across countries” depending on how exposed they are to the war and how exposed they are to AI. In April, 54 countries produced 95% of the world’s growth. Now it’s 32.

Global debt is sitting at about 310% of global GDP.

Japan especially is a point of pain, with the 10-year yield there going north of 3%

, which is extremely unusual. They held rates near zero for a very long time, which supported the yen-carry trade that some people like Ed Yardeni think is causing little explosions across the entire debt market. But Debt is no longer cheap. Money is no longer free. Real rising interest rates are something that not many people have too much experience with.

Rates have not been this high in a long time. The 10-year Treasury is at its highest since 2002. For my dad, rates were this high when he bought a house in the 1980s, but they’ve dropped meaningfully since then, which allowed the baby boomers to benefit from easy money and a rising stock market. That could all be changing.

When the 10-year climbs, it hurts people. As Heather Long pointed

out

, mortgage rates go up, auto loans go up, personal loan rates go up, refinancing gets harder, banks have to hedge, credit cards get more expensive. It’s a ripple effect, at the same time that workers quite literally have to work harder than ever to afford ground beef and coffee (the work-time-to-caffeine ratio is at the highest it’s been in modern times, according to

UBS

.)

For a long time, the economy didn’t seem to care too much about interest rates. But now, it might, because increasingly, not even AI can

really

hide from the bond market.

AI NEEDS YOUR MONEY RIGHT NOW

The bond market is already jittery. Bonds are a beacon of truth. That matters for the AI buildout and the

$10.3 trillion (!)

in expected financing it needs by 2032, but it also matters for the Treasury market. And every other market. Private non-residential fixed investment and residential investment have both shrunk in comparison to AI, which has exploded upwards.

So is AI borrowing literally crowding out everybody else, the Treasury included? FT Alphaville ran the numbers and said

probably not

, pointing to economic strength and geopolitics instead. Tracy Alloway

cites

Matt King in her Bloomberg piece, who said “the issue is less the volume of current borrowing than the desire.” She summarizes it as:

One broad way of looking at the current bond environment is as a place where issuers who feel they have no choice but to borrow are running headfirst into buyers who feel they have a rapidly-expanding array of investment options.

Basically, the borrowers feel like they have to borrow no matter what, and the buyers get to be picky. The two biggest issuers appear to be willing to sell bonds at any price, which is the shrug in action.

The AI companies are borrowing so much

money

(and the most money ever) at the same time that the government is borrowing even more, to the point that BlackRock

wrote

“heavy sovereign borrowing is increasingly competing with extraordinary private financing needs, particularly from AI, pushing prospective capital demand toward levels rarely seen outside crises.”

Outside a CRISIS? And maybe this is a CRISIS! The economy is reliant on a technology that the creators of said technology seemingly can’t

control

(OpenAI keeps

hacking into the government

) with no real

moat

and depreciating physical infrastructure and unproven demand at the prices needed to pay it back, while again, spending more money than any money has ever been spent ever,

counting

for 25% of US GDP growth and 8% of total GDP and more than ⅓ of the stock market, at the same time that we face a war that seems endless, pushing up the cost of oil, the foundation for any economy (and the AI buildout) and a government that is borrowing also a lot of money for bigger and weirder projects.

So the fight over money is a big fight, but the fight over real stuff has already mostly happened. You can see it in what’s getting built. In August, spending on data center construction was up 73% from a year earlier. Factory construction was down about 20%. Home building was down about 5%. So - the data center buildout

was up about

$9 billion year to date through July of this year, and private construction (so homes and apartments and shopping centers) fell $46 billion over the same time period.

President Hammack made this point in her speech too. Oil isn’t the whole inflation story and the AI buildout likely has something to do with it. Data centers are bidding up the price of computer and electrical parts, and those same parts go into aviation, electricity production, and construction – kind of everything, you know.

This kind of pressure relies on the Fed to start hiking rates – to slow it all down, right? But who does higher rates slow down? Not the big boys. Everyone else around the big boys. So the AI buildout helps lead to inflationary pressures, and it’s the borrower most likely to just… shrug.

Torsten Slok at Apollo calls this

“Dutch disease”

— a rate-insensitive boom draining capital, power, and labor from the sectors that are rate sensitive, like housing and autos. He says the only solution is expanding the supply of chips and power, which the industry is trying to do, but the shrug isn’t as easy as it used to be. Everyone is starting to raise an eyebrow. Shrugs can’t last forever.

CAN YOU SHRUG FOREVER? BOND MARKET SAYS NO!

And the bond market is getting increasingly wary about all this - credit default spreads, which you can think of as like car insurance for bonds, are getting more and more expensive because the AI companies are becoming riskier and riskier drivers.

Oracle is a potential canary in the coal mine here

according to Bloomberg

(and it’s VERY important to note that some of this volatility is happening because yields in general are going up). Oracle’s credit default swaps, so the cost to insure Oracle’s debt, hit a record high. Oracle is one notch above junk, and Larry Ellison just… keeps spending money. The Ellison family put $17 billion into Paramount’s purchase of Warner Bros. (and that combined entity

has

$80 billion in debt)! As Campe Goodman says, “you have to consider your total Larry Ellison risk.”

But if Oracle does get downgraded to junk OR OpenAI, one of its biggest customers (the whole circular financing thing), goes sideways, people are going to have to sell Oracle bonds, which would be not so good, creating ripple effects throughout the market - quite

literally

, as this map of revenue dependency across the AI ecosystem from the FT shows.

The big cloud companies report something called “remaining performance obligations” which is money customers have signed contracts for but haven’t paid yet. For Microsoft, Amazon, Alphabet, and Oracle - that backlog went from about $0.5 trillion in early 2024 to over $2 trillion now, per Jefferies. That looks like guaranteed money, at least on paper. But a lot of it is owed by the AI labs.

Anthropic’s $500+ billion in infrastructure obligations is somebody else’s backlog. So the lenders are counting on the cloud companies, the cloud companies are counting on the labs, and the labs are counting on… people? To eventually pay for all of this? Who, per Anthropic’s filing, aren’t locked into long-term contracts and could cut spending. Nearly a quarter of its revenue last year came from two customers!

Jefferies says that it can “no longer be assumed” the models will ever be profitable, and sees a “2008 real estate credit angle” to this

boom

. A Barclays

trader

says he hasn’t seen a “credible framework as to where this money is going to come from.”

And the risk is getting spread around. The FT

reported

that SpaceX is trying to raise $40 billion in debt to buy Nvidia chips. Investors who were pitched on it earlier said that they got “a short two-page deal memo with pictures of outer space and an arrow pointing out that the company was going to build data centres “somewhere in the universe”. As one of them put it “How are we supposed to take that to the IC?” (the investment committee, the people who have to say yes to an investment.) But because SpaceX’s debt is rated investment grade, insurance companies and pension funds are allowed to buy it, despite the whole “we are going to space :p moon data centers” thing.

Insurance companies?? Pension funds??

Nvidia is working on the other end of these loans. The FT

reported

that Nvidia has talked to insurers about covering the losses if one of its smaller customers defaults and the chips can’t be resold for enough to pay back the loan. Jensen Huang wants chips treated like an “investable asset class” like airplanes. So the company selling the chips is helping to arrange the loans to buy the chips, and now the insurance on the loans to buy the chips.

And Blue Owl, which leases 140 data centers to the big tech companies, said this week it wants more

insurance

money to do it. According to their co-CEO, Doug Ostrover:

We are not managing enough capital. We’re realizing that if we had a bigger pool of capital, meaning insurance capital, we could provide really unique solutions.

Spreading risk doesn’t make it smaller! It moves it to whoever is on the other side of the trade, and here, it’s insurers and pension funds, which is to say, people’s annuities and retirements. And the insurers are already getting full. Zach Ablon at Goldman Sachs said on the

firm’s podcast

that earlier this year about 15 insurers were putting in big orders for 30-year AI bonds, but by June, that had been cut in half. They are, he said, “clearly losing their appetite for this risk.”

Ostrover also pointed out something really bleak, as the FT summarized:

Ostrover said that even if the infrastructure was worth “zero” at the end of the leases, its investors would make double-digit returns. “You can make 10 to 12 per cent [without relying on the terminal value]. We think that’s a really compelling investment.”

AND WHAT ABOUT THE INFRASTRUCTURE?

So the investors will be okay, but what about the communities stuck with these huge, expensive sheds? As we know - they are really expensive.

Bloomberg

got a cost breakdown for a single one, a one-gigawatt data center in Texas, from the real estate firm Cushman & Wakefield.

About $30 billion at the high end. Half of that is the chips and servers, but the developer has to set aside nearly $3 billion just in case things cost more than planned (which they will).

Bloomberg also did the math on what 4 of the big companies are spending this year (Meta, Amazon, Microsoft, and Alphabet) - enough to cut a $5,500 check to every household in the United States.

A lot of communities are saying no. Heatmap

counted at least $260 billion of data center projects canceled

so far this year after local opposition, half of it in the last three months. And the ones that do get built end up hitting a totally different wall.

Even if the money keeps coming and the investors start taking baths in gold or whatever, there is a physical limit, which is how the triangle really connects. Energy (and trust) is the foundation of an economy. There is not enough energy for all the power that these big boys want.

The CEO of Eaton

, which makes electrical equipment, said that 342 gigawatts of data centers have been announced, and only 50 gigawatts are operating today. As the CEO said, that’s 6 to 7 times what exists! As Goldman Sachs wrote, “the physical grid could emerge as the ultimate constraint.

4

”

But some of those gigawatts are like ghostwatts?

The Wall Street Journal

reported that developers shopping for data center sites ask several utilities for power, so the same data center gets counted a couple of times. Only about a quarter (!) of the requests seem “high confidence” (meaning, a real request) and so now, the utilities are like “you’re going to have to enter a binding commitment with us before we build ANYTHING.” As Exelon’s Jeanne Jones put it, “if we build out the grid and then that data center walks away, someone’s got to pay for it.”

And of course, someone is everyone else, the people that can’t just shrug all these problems away.

Oracle is running head first into these energy problems in New Mexico. It gave

force majeure notice

(meaning they were like if this thing is late due to things outside our control, we get to delay paying rent) on its New Mexico data center. The project is waiting on power (waiting on a gas pipeline, which got delayed more than half a year after regulators kept denying permits) and is facing local opposition – 65% of New Mexicans don’t want big data centers at all.

Bloomberg reported that Oracle is starting to truck (diesel) natural gas to data centers in Utah and Texas and may do the same in New Mexico.

Trucked gas costs

about 4x as much as pipeline gas, but these data centers HAVE to turn on to be useful

5

.

Diesel hauling natural gas to a data center. Oil, AI, and borrowed money, all while the people who live there ask for it to not happen. The whole triangle in one picture.

AND WHAT PLANE OF REALITY IS THE STOCK MARKET ON?

Almost half of the earnings growth in the stock market comes from AI investment, across “chipmakers, tech hardware suppliers, industrial firms, and utilities”

which

has fed into a booming stock market and into household wealth.

Corporate profits jumped almost 60% this spring, but more than half of American companies made less money than the year before, according to

Aswath Damodaran

.

BCA Research charted how much profit Wall Street expects from big companies over the next year. It’s up 31% since January, most of that is tech, and everything outside of tech, communications, and energy adds up to about zero.

The market is at all time highs and 75% of those stocks are trading

below

their 50-day moving average, meaning, they aren’t doing great. But those other 25% are doing

fantastic

. Since June, more than half of stocks have fallen at least 20% according to Morgan Stanley. Not too unlike “computer terminals printing out market prices while the world blows up,” as Annie Dillard

wrote

.

Part of this is because one company’s AI spending is another company’s sales. When Microsoft buys chips, that money shows up as Nvidia’s profit – the whole circular financing thing, right? So the money goes around in a small circle of companies. And that teeny circle is the entire economy, kind of.

That’s the worry that everyone keeps harping on, this feeling that everywhere you turn, there is AI and it has to be there, or else everything explodes into little pieces. If the AI spend stops and the debt has to be paid back –on top of a stock market that is overexposed to AI, with most other companies in it struggling and increasingly so, with SpaceX’s IPO and Alphabet’s public stock sale and Anthropic’s planned IPO, there will be a glut of share supply.

And when there is too much supply of something, the price of it has to go down, which, if you follow the loop means, household wealth could fall too.

Add it all up, and you get what Jim Chanos, summarizing Paul Kedrosky in a recent

interview

, called a boom that checks all four boxes, pointing out that “major booms are either credit-related, government-related, technology-related, or real-estate-related” and AI is the first in a long time that checks all four boxes. And that’s where the risk comes from.

There comes a point in an economy where you have to ask – how much is too much? Usually this is answered by a bubble bursting, too much money going into places where it isn’t needed. Right now, because so much of this AI money is borrowed (from all pockets of markets, across all banks and insurers and private credit), it’s extremely risky.

PART 5: THE SHRUG HEARD AROUND THE WORLD

WHY IS NO ONE REGULATING AI?

So if this many people are paying for it and everyone keeps writing these Substacks about what’s happening, why doesn’t someone slow it down? The Trump administration has every incentive to boost AI as much as possible and to regulate as little as possible because it’s the thing holding up the stock market. The Atlantic

reported

that Trump “has expressed worry to those close to him that a halt to AI expansion could send the economy into a recession.”

“Whoever wins AI, WINS!” is the current

messaging

announced aloud on the casino floor (while having the companies sign a document where they pinky promise to self-regulate (and where the United States is spelled wrong, an omen for a not-too distant future)). The president recently posted that the White House considers anyone who says “Artificial Intelligence” instead of “Super Intelligence” to be

“THE ENEMY!”

So that’s probably all we are going to get from the government for now, a semantic war.

Meanwhile, Axios reported that the AI companies are planning for “

the day after”

if something goes badly wrong. They expect some sort of cyberattack on banks, the Internet, or power and water systems. And

Sam Altman

, who runs OpenAI, told Politico that “the world should accept some bad things happening for the benefits of this technology.”

WHAT! You can’t SAY that!

It feels like everyone is just pointing fingers when it does come to regulation. Treasury Secretary Scott Bessent

told Axios

“I believe that we have to be prepared for every occasion, but on the other side, this alarmism without solutions by some of the AI community, that’s not leadership.”

SOMEONE HAS TO DO SOMETHING!! The Vanderbilt Policy Accelerator (I am a Senior Fellow there) launched their version of something, a good something, called the

AI Governance Project

that provides blueprints for how AI gets governed, who gets the benefits, and who pays for all the stuff happening.

WHAT IS CHINA DOING?

So there are solutions, but the solutions would require rules, and one of the reasons no one wants rules is The Race. If we slow down, China won’t, seems to be the thinking. But China is thinking about things a bit differently. As Kyle Chan

writes

:

Most new public-facing A.I. models must be

registered with the government

and pass a battery of tests before they can be released. China has created extensive rules requiring labeling of A.I.-generated content and limiting how

humanlike A.I. chatbots can operate

.

Of course, the answer China lands on for all of that is the Party controlling everything. A state-backed paper, the Global Times, called the CEO of Anthropic Dario Amodei’s call to slow down AI a “Cold War playbook” targeting China specifically in order to make them fall behind. So that’s all happening at the state level, power versus power, as it’s been since we started organizing ourselves into little villages and factions thousands of years ago (not to diminish this in the slightest, but it’s history repeating itself.) As Chang Che wrote in the excellent

There Is No A.I. “Race”

For years, the powers in Silicon Valley and D.C. have conjured an image of China barrelling toward superintelligence. The irony of this is twofold. First, Chinese A.I. policy remains focussed primarily on diffusion: how to drive A.I. into public and commercial life. (It’s rare to hear a Chinese leader mention the risks of runaway, self-improving A.I.) Second, as Matt Sheehan, a researcher at Carnegie focussed on global technology issues, recently observed on Ezra Klein’s podcast, in the years in which Chinese A.I. companies caught up to the American frontier, they also had to accommodate their own government, home to “the world’s strictest, most comprehensive, most burdensome A.I. regulations.”

So nobody at the top is stopping it, it seems. And the cost of that lands on people.

PART 6: THE LUDDITE AT THE END OF THE TUNNEL

An economy that shrugs at prices ultimately shrugs at people. The worry here is that technology costs more than promised, makes life worse for everyone who doesn’t have it , and makes the people who do have it less than they were.

And that’s the other side of all this. What often gets lost inside the triangle is the value of people in an economy that keep telling them that they suck. AI could have been something as a complement to humans, something that could challenge us and help us, but instead, as is the Law of Slopnomics, it’s turned into a way to make hot girls and more work and more noise and more more more more more but all of this, theoretically, is for US. So why isn’t it centered on us? Why is it marketed as the Job Destroyer and Human Killer 1000 Bot when it could be so different? (we all know why, but you know, why?) And for the stuff it’s touching, does it actually work?

WHAT HAPPENS IF IT COSTS MORE THAN PROMISED?

As Greg Ip writes, two doctoral students at Harvard found “AI’s boost to projects was small and statistically insignificant.” The reason was “users were spending much more time reviewing, commenting on, and changing pull requests.”

Take healthcare. AI is also making healthcare more expensive, at least according to insurers. The

Blue Cross Blue Shield Association

says hospitals’ use of AI was responsible for about $1 billion in additional costs for its companies over the last two years. Using AI tools,

hospitals

collect more information about patients, which enables them to submit bigger claims. By adding a secondary diagnosis for things like anemia, the hospitals were able to collect $12,000 more per case (the insurers are one side of the fight here, so take it with a grain of salt).

Benefit costs could go up by more than 11% for employers next year. When asked if the technology will save money, one economist said that “a huge share of cost” will be freed up by, as the NYT

writes

, “replacing the people that handle claims processing.” Another person described where this goes: “it’s bots fighting bots, agents fighting agents, a horrible dystopic future nobody wants to live in.” And another man summarized it well - “I don’t think there is anything intrinsically different in A.I. that will escape the market forces.”

WHAT HAPPENS IF IT MAKES EVERYONE ELSE’S LIFE WORSE?

OpenAI has

dots

now, that are a team, a bunch of dots (Dots are what OpenAI calls its new assistants that work around the clock, and soon you’ll be able to have a bunch of dots on demand). The AI assistant land is going to get increasingly crowded as the companies try to spur up more and more demand for their products. They will read your mind.

The magic of dots is when they bring you work done the way you would do it, sometimes before you even think to ask. With dots alongside you, ambitious ideas and projects you haven’t had the time or support to pursue start to feel achievable.

And this seems to be the future? Little guys doing little tasks for everyone? I think this is plausible but it could crater the parts of our economy that rely on an enormous amount of friction. Torsten Slok at

Apollo

pointed out that the average checking account pays about a tenth of a percent, and you can clearly get a higher rate elsewhere. The business model of a bank is essentially predicated on most people never bothering to move their money to somewhere they can make a little more money. If everyone’s AI assistant moved their money out, he wrote, that “would be a problem for the entire financial system.”

There are good parts to this. Patrick Collison, who runs

Stripe

looks at it and sees good news. Companies have always made money off people not paying attention, he says. Today, “people who forget to cancel subscriptions subsidize those who don’t.” If your AI assistant never forgets to cancel anything, ever, the companies will have to start competing with a better product. This all “hinges on the assumption that personal agents will be on the consumer’s side.” So the same situation, from two different angles.

Jamie Dimon

said

that JPMorgan is “not going to allow just any outside agent to come in.” So the banks see this too.

But these AI assistants pollute the commons for individual gains. It creates negative externalities for the humans that don’t have a bot to manage the pounding requests of an AI agent. One

person

posted that her restaurant team scrambled to fill a client’s last minute requests, and she realized the requests “were likely made by an AI agent and that the human client might not have personally cared. This felt bad!” One man’s AI assistant asked Resy for a table about 200 times an hour, until Resy s

hut down his account.

WHAT HAPPENS IF IT MAKES PEOPLE LESS THAN WHAT THEY ARE?

Going back to the worker, the human ecosystem, the people underlying the technology who often do not get credited or paid (especially in the case of the models getting trained on stolen work), there is a big worry that AI is going to make us dumb. One of the worst case scenarios here is that everyone uses AI to get stuff done but forgets how to do stuff. It’s not good enough to replace people entirely, but it makes people “meat proxies” by any other name, a vessel for technology rather than humanity.

We see this happening in a microcosm at

MIT

, where study groups are disappearing and students are using AI on problem sets and are losing confidence. Students worry about overreliance and losing themselves in the machines.

In response, MIT is starting to do more oral exams and semester portfolios and in-classroom work in order to rethink education during this time of enormous change. They encourage people to think beyond dollars and cents, that people are sometimes more important than output, and that “however sociable or responsive AI systems become, they cannot substitute for the relationships and practices necessary to grow and mature as a human being.” Their posture is not anti-AI, but rather towards figuring out how to “provide students with appropriate AI learning experiences from the time they arrive at MIT through graduation” in the service of the “betterment of humankind.”

There is a name for that spark! Tyler Cowen recently

wrote

about Polanyi knowledge, which is “knowledge of time and place, inarticulable knowledge, custom and habit, and many other particularities.” Contextual knowledge, human knowledge, or as Michael Polanyi describes it, “ we know more than we can tell.” Cowen says there is no way to “bring an AI into your office and have it figure out how that office works.”

HOW HAVE WE HANDLED THIS BEFORE?

And what are these market forces pointing us to? And how have we responded to them in the past?

Thomas Pynchon wrote an essay in 1984 (of course) titled

Is It O.K. To Be A Luddite?

In it, he talks about the push and pull between literature and science, that we have enormous flows of data encircling us at any time (reminding us that time itself is a loop), that anything and everything is at the tip of our fingers.

He walks through the history of those that fought back against technology, known as the Luddites, often confused as those who hated progress but really, were fighting back against “the machines coming more and more to be the property of men who did not work, only owned and hired” and the “concentration of capital each machine represented” alongside “the ability of each machine to put a certain number of humans out of work - to be “worth” that many human souls.” The Luddites did not hate technology. They hated what it was doing to their lives. They did not hate the product, they hated the application, and the inequality that festered as a result.

Towards the end of the piece, Pynchon wonders about the future. That perhaps the computer will do good. That it will “cure cancer, save ourselves from nuclear extinction, grow food for everybody, detoxify the results of industrial greed gone berserk - realize all the wistful pipe dreams of our days.” He mentions the challenge of the convergence of “artificial intelligence, molecular biology, and robotics” by name. He ends with Byron’s lines for King Ludd, an acknowledgement of the problems present and the problems to come.

As the Liberty lads o’er the sea

Bought their freedom, and cheaply, with blood,

So we, boys, we

Will die fighting or live free

And down with all kings but King Ludd!

And today, we are having the same conversation that he had over 40 years ago. We will likely have it in another 40 years. There is always going to be a tense relationship between humans and machines, because the makers of the machines want to make enormous profits.

Enormous profits require paying no people for anything, just selling the products, seemingly to other makers of machines, who also seek enormous profits. But who funds the makers, at the end of the day? Who is truly the bed rock of the economy? The human souls, as Pynchon posits: the ones who put in the work to make each component of the machines possible. Those human souls do not hate technology, they hate what technology is doing to them.

But growth at any cost, whoever wins, does not allow for the foresight to examine “a human soul.” As Pynchon wrote, “to insist on the miraculous is to deny to the machine at least some of its claims on us.”

As playwright James Graham

wrote

in a meditation on Keynes:

What kind of growth, and to what end? An AI data centre in a hollowed-out post-industrial community might provide 30-odd jobs and create numerical growth on a Treasury spreadsheet but does it materially address the lack of meaning, purpose or agency many feel in their lives?

Pope Leo has also pointed

this out

, how at risk we are of losing humanity within this technology (in a tweet!):

In this era of artificial intelligence, it is becoming urgent to distinguish human art from what machines produce. There is an ontological difference, even before an aesthetic one, between art and what a machine can generate through statistical calculation based on millions of images created by others. Algorithms lack the spark of humanity. For this reason, the Church wishes to renew an alliance with artists and cultural institutions to safeguard our humanity.

That’s the thing that gets lost inside the triangle. The things that can’t be priced, at least yet now, perhaps not ever. People will try their best to commoditize that which cannot be sold, that which cannot be automated into bytes and bits, and perhaps, they will be successful, or perhaps, we will stop and ask what we are doing this all for.

PART 7: FINAL THOUGHTS

The slot machine from Vegas haunts me. It’s a metaphor for where we are. Most people in the economy are putting in the work – paying higher prices, taking on higher interest rates – but the jackpot isn’t for them. Wealth instead sits inside a self-referential loop, where hyperscalers borrow billions to buy chips from chipmakers, inflating stocks and GDP, while consumers report that they are near rock bottom.

And who should be winning? In the old, imperfect math of the American dream, the answer was supposed to be

people

. James Truslow Adams,

the original American Dreamer

, described it not as “a dream of motor cars and high wages merely” but more as people being able to become the fullest version of themselves, where each person can reach “the fullest stature of which they are innately capable, and be recognized by others for what they are.”

The promise wasn’t that everyone would have ten yachts, but it was a promise that if you worked hard, you’d get an honest shake at it all. A home maybe, some savings, dignity, space to help your kids, our next generation, grow up to be better than we are today.

I kept running into that all week. The point of the Chicago Fed panel was to talk about why it matters how people

feel

about the economy. We often write off these feelings saying “well they’re still spending so can’t be feeling that bad!” We conflate money and happiness so often, spending with contentment. We throw pennies into the fountain to try and beget some luck, a donation to some unknown figure, somewhere, money in exchange for happiness.

I’ve been rereading Annie Dillard’s work, and Wendell Berry’s, mourning them and the prose we often sacrifice for efficiency. Writing is similar to the way a person is constructed. You have the bones of the piece and you have the connective tissue between paragraphs, and somewhere in there, you should have a heart. The thing we try to protect most in writing is a heart, the thing we try to protect most in each other should be our heart. Annie Dillard, writing about a total eclipse

said the same thing in a different way:

Significant as it was, it did not matter a whit. For what is significance? It is significance for people. No people, no significance. This is all I have to tell you.

And I think that is the question that we don’t really know how to answer with AI. Is the end goal here just… agents talking to agents and 8 billion people on some form of UBI? Or will this article sound foreign and far away to those perhaps 5 years or 100 years into the future to people, living in a renovated data center, that have a brain chip that holds a file titled “2020s: What Various Word Engineers Of Yesteryear Got Wrong About the Today” that contains these here words? Are we ushering in a utopia? Or will it be a slow and arduous future, where not much changes but where things get better and better, but slowly, and often not before they get worse?

A cab driver in Vegas started asking me what I thought about AI, and as we were driving, a Zoox scooted by, lights twinkling. My driver was retired, driving because he was bored and wanted some cash, and he scoffed at the Zoox. “I can’t stand those things,” he said. I asked him if it’s impacted his driving. “Of course,” he said. How did he handle it? “I mostly handle it by being a better person. That’s my hope, I guess. That by being a good person, people will keep wanting to be around people.”

This is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

Thanks for reading.

1

Joanne Hsu, who runs the Michigan survey, summarized it all so well – sentiment is one person, one vote. The data is dollar-weighted, and dollars are clearly doing fine. But the people behind some of them are not doing so fine.

2

I don’t think I share his same optimism , but I do share the concerns

3

To be fair, the buyers are still buying. When the government borrows money, it holds an auction, and investors bid for the chance to lend to it. A few weeks ago, the government sold $39 billion of 10-year debt, and had nearly 3x as many bids as it needed. They’ll buy it, especially if they get over 5%!

4

China has the opposite problem. The Financial Times went to Ulanqab, a city in Inner Mongolia, with 89 data centers built or planned, one of them going into an abandoned high school (a bit on the nose). Electricity in that area costs about half of what it does in Beijing, and China can put a data center up in 12 to 18 months, against the 18 to 24 we have in the US. What it can’t get is enough chips to fill them, because the best ones are American and Washington restricts how much can be sold to China. So America has the chips and is short on power. China has the power and is short on chips.

5

As Bloomberg noted, “If Oracle powered just 100 megawatts — about 4% of its ultimate 2.45-gigawatt footprint — with CNG deliveries, each large trailer would only provide about 40 minutes of electricity”

595

71

110

Share

Previous

Next