AI boom leads to massive computing crisis… Electricity bills skyrocket… But aren’t super-expanders supposed to pay?… Invest in ‘AI demand shocks’… as senior analyst Brian Hunt looks today
Everyone is talking about oil, but I think what the world is fundamentally lacking is tokens.
This line comes from Ben Bouladian, an engineer and technology investor the Wall Street Journal Sunday.
To make sure we’re all on the same page, a token is simply a unit of measurement to track how much computing power an AI task consumes.
Think of it as the underlying currency of the AI economy – every query, every document generated, every autonomous agent action derived from the view. Right now, that currency is running out.
This has major investment implications that we will get to shortly. But first, let’s look at the scale of the problem…
The computing crisis is already reshaping how AI companies operate
the Wall Street Journal This week we took a detailed look at what’s happening inside the AI infrastructure stack, and the numbers are staggering.
Over the past few months, demand for “agent” AI, the latest development in artificial intelligence, has increased dramatically. “Agents” not only answer questions, they perform tasks autonomously: writing code, scheduling appointments, and managing complex, multi-step workflows.
The shift from conversational AI to agential AI is causing a spike in computing consumption that current supply chains are not built to accommodate.
here Wall Street Journal With an example of an amazing request:
Token usage on the OpenAI API — a platform whose software most enterprise users access — rose from six billion per minute in October to 15 billion per minute in late March.
The supply side can’t match it.
According to the Ornn Compute Price Index, hourly rental rates for the most advanced Blackwell-generation GPUs from Nvidia company (NVDA) — those that power modern AI — have risen to $4.08 per hour, up 48% from just two months ago.
The pressure is already starting to manifest in ways that directly affect end users
Anthropic announced in late March that it would begin rationing access to computing during peak hours on weekdays. Enterprise customers are starting to turn to competing providers. OpenAI has canceled its video creation app Sora in part to redirect computing resources toward higher-priority products.
the Wall Street Journal I picked up on how intense the pressure had become. From the article quoting JJ Cardwell, CEO of cloud infrastructure company Vultr:
There is a massive capacity crisis unlike anything I have seen in more than five years of running this business.
The question is, why don’t we deploy more equipment?
Lead times are very long. Data center build times are long. Energy available until 2026 has already been talked about.
Re-read the last sentence…
Power has already been talked about.
So, why is this important?
Well, let’s take it a step further – what really is the token shortage?
Basically, it’s an electricity shortage wearing a technology hat.
Strength: the constraint behind the constraint
Pull back one layer of the computing crisis and you’ll find the energy problem is deeper and broader than most people realize.
on monday, Bloomberg published an in-depth look at the growing size of America’s electricity bills, and the surprising data behind them.
from Bloomberg:
North American Electric Reliability Corp., the nation’s grid security regulator, expects U.S. summer power demand to rise by 224 gigawatts over the next decade — roughly equivalent to adding 180 million homes.
One analyst said the last similar increase occurred during World War II.
Building AI is the main driver of this increase.
Let’s look at an example of driving this home…
The Pennsylvania-New Jersey-Maryland (PJM) interconnection is the largest power grid in the country, extending from the Midwest to the East Coast. Over the three years ending in May 2028, data centers are expected to add at least $23 billion to customer bills on the PJM network alone. This is an increase of more than 50%.
Meanwhile, in parts of eastern Pennsylvania, electricity prices have already risen 200% since 2020.
Now, to make up for it, President Donald Trump, in his State of the Union address on February 24, told America’s largest technology companies that they “will have a commitment to provide for their energy needs.”
Ten days later, Amazon.com Inc. (Amzn), Alphabet Company (Google), Meta Platforms Inc. (dead), Microsoft Corporation (MSFT), OpenAI, Oracle Corporation (ORCL)And Elon Musk xAI They gathered at the White House and signed what the administration calls the Taxpayer Protection Pledge — a commitment to “build, bring in, or buy” all the power and network infrastructure needed for their data centers, while passing none of those costs onto American households.
Trump has been frank about the policy that drives it. At the signing ceremony, he told the assembled technology executives:
They need some help with PR because people think that if a data center is built there, electricity prices will go up.
So, that should be settled, right?
This is the part that doesn’t issue press releases
The pledge is primarily forward-looking, but the price increases documented by Bloomberg have been building for years.
PJM grid capacity rates — what utilities must pay generators for electricity — rose from $28.92 per megawatt per day in the 2024-25 delivery year to $329.17 in the 2026-27 delivery year.
That’s an 11-fold increase that was already built into interest rate structures long before anyone in the White House signed anything. But that’s still not enough – PJM’s recent capacity auction was 6.6 gigawatts short of available supply.
Additionally, as I just mentioned, pledge addresses future Data center constructions, not existing ones.
So, today’s energy bills reflect:
- Power infrastructure has already been established to serve the existing expansion facilities
- Rate increases already approved by utility commissions to pay for related grid upgrades
These costs are very real, very large, and implicit, and a pledge does not solve them.
In addition, even for new projects, signing a pledge to build your own power plant does not trigger this. Permitting a new generation facility takes two to four years. Construction takes more years moreover. By the time the concrete is poured, the demand it was designed to meet has already doubled.
Which brings us to the investment implications…
When demand outstrips supply, the winners aren’t always who you expect
Every major technological breakthrough in history has eventually encountered massive demand for the critical components involved in building that technology.
Brian Hunteditor Money and mega trendshas built an entire investment framework around this dynamic. He calls it the “AI demand shock.”
Here’s Brian explaining the basic concept:
Over the past three years, the best way to make money fast in stocks has been to identify an industry where an AI “demand shock” is about to hit… and then invest there before the shock arrives.
It’s not a supply shock, mind you, where a war or pandemic suddenly cuts off supplies of a resource like oil.
Instead, I’m talking about a “demand shock,” where demand for a particular resource or manufactured product suddenly rises… causing its price to rise by hundreds of percent.
The historical examples Brian cites confirm his point.
In 2023, a shock to demand for advanced semiconductor AI sent Nvidia’s sales soaring 525% in less than two years. Around the same time, there was a sudden need for data center cooling systems Comfort Systems USA (It works) Up to 1000%. At the same time, demand has risen for advanced optical systems – components that allow data to be transferred quickly between AI servers Lumentum Holding Company (Light) An increase of 1,164% in two years.
None of these companies are AI companies in the main sense like OpenAI or Anthropic. Instead, they are the primary suppliers of building AI — the companies that sit at the forefront of the technology, making the physical things without which the technology could not exist.
Brian explains why these moves tend to be so big and so fast:
Artificial intelligence is advancing at such a rapid pace that AI-driven demand shocks now occur every year… and create the fastest—and most profitable—stock market moves we have ever seen.
The typical manufacturing industry takes five to ten years to build operations capable of meeting growing demand. The same applies to mining industries that provide important raw materials.
But our new, super-fast technological cycles now move much faster…
We are now seeing crazy mismatches in interconnected and interconnected parts of the economy.
It’s as if we have a rocket motor connected to the propulsion system of a Toyota Corolla.
The mismatch between the Corolla’s rocket engine and propulsion system is where the investment opportunity lies.
Brian’s current focus: Artificial Intelligence Chemistry
While opportunities exist everywhere, Brian Tuesday issue of Money and mega trends Shed light on a sector that may surprise you…
Chemicals.
Here’s Brian:
The chemicals sector is usually viewed as an “old economy” industry that produces products such as plastics, paints, solvents, detergents and pesticides.
However, some chemical companies are participating in “brand new economy” activities related to AI.
The chemicals industry is at the forefront of almost every physical component in the AI stack: from the specialty gases and materials used in semiconductor manufacturing to the high-purity solvents, coatings, refrigerants, flame retardants and advanced polymers that make modern data centers possible.
Each AI server relies on a long line of ultra-specific, ultra-pure chemicals. As the use of artificial intelligence increases, the need for more, more complex and higher purity chemicals increases.
Brian Recommendation March 6 Adding chemicals to your investment portfolio really pays off…
Chemours Companycopy) It’s up 37% since that note and just hit a new one-year high. and Elements Solutions Company (Easy) It added 15% and also reached a new one-year high.
For readers who want Brian’s full analysis—including the specific names he monitors and associated with all the demand shocks he tracks beyond chemicals— His research is available for free at Money and mega trends
Back to the token shortage
Ben Boldian’s note that opened our door digest – The world is fundamentally lacking in tokens – that’s true. The shortage is real, and it’s crippling AI companies’ ability to serve their users right now.
But step back far enough and you’ll see the broader sequence…
- The token shortage is downstream of the GPU shortage…
- Which is downstream from the data center shortage…
- Which is downstream of the energy shortage…
- This is the process of building infrastructure that the physical world did not have enough time to complete.
This is what true technological transformation looks like from the inside. Demand arrives faster than the supply chain can respond.
And the companies in the middle of these captive points – the ones that make chemicals, cool servers, and supply energy infrastructure – are where the big money is made today… and where we want to be.
We will continue to track this with you here at digest.
I wish you a good evening,
Jeff Remsburg
(Disclaimer: I own LITE)
