The AI growth is main to an enormous computing crunch… electrical energy payments are skyrocketing… however aren’t the hyperscalers imagined to pay?… put money into the “AI demand shocks”… the place senior analyst Brian Hunt is wanting right this moment
Everybody’s speaking about oil, however I feel what the world is especially wanting is tokens.
That line comes from Ben Pouladian, an engineer and tech investor quoted in The Wall Road Journal on Sunday.
To ensure we’re all on the identical web page, a token is just a unit of measurement monitoring how a lot computing energy an AI process consumes.
Consider it as the essential foreign money of the AI financial system – each question, each generated doc, each autonomous agent motion attracts from the provision. And proper now, that foreign money is working quick.
This has vital funding implications that we’ll get to shortly. However first, let’s have a look at the scope of the issue…
The computing crunch is already reshaping how AI firms function
The WSJ ran an in depth look this week at what’s taking place contained in the AI infrastructure stack, and the numbers are placing.
Over the previous a number of months, demand has exploded for “agentic” AI, the newest evolution of AI. “Brokers” don’t simply reply questions however autonomously carry out duties: writing code, scheduling appointments, managing advanced multistep workflows.
The shift from conversational AI to agentic AI is inflicting a dramatic spike in computing consumption that current provide chains weren’t constructed to soak up.
Right here’s the WSJ with an instance of the astonishing demand:
Token use in OpenAI’s API—a platform the place principally enterprise customers entry its software program—rose from six billion a minute in October to fifteen billion a minute in late March.
The provision facet can’t match it.
In response to the Ornn Compute Value Index, hourly rental costs for probably the most superior Blackwell-generation GPUs from Nvidia Corp. (NVDA) – those that energy fashionable AI – have risen to $4.08 per hour, up 48% from simply two months in the past.
The pressure is already exhibiting in methods which might be hitting finish customers straight
Anthropic introduced in late March that it will start rationing computing entry throughout peak weekday hours. Enterprise purchasers have began switching to competing suppliers. OpenAI scrapped its Sora video-generation app partly to redirect computing assets towards higher-priority merchandise.
The WSJ captured simply how acute the stress has grow to be. From the article, quoting J.J. Kardwell, CEO of cloud infrastructure firm Vultr:
There’s an enormous capability crunch that’s in contrast to something I’ve seen in additional than 5 years working this enterprise.
The query is, why don’t we simply deploy extra gear?
The lead instances are too lengthy. Knowledge heart construct instances are lengthy. The facility obtainable via 2026 is already all spoken for.
Reread that final sentence….
The facility is already spoken for.
So, what’s the importance of that?
Properly, let’s take it one step additional – what actually is a token scarcity?
Mainly, it’s an electrical energy scarcity sporting a tech hat.
Energy: the constraint behind the constraint
Pull again one layer from the computing crunch and you discover an power drawback that runs deeper and wider than most individuals understand.
On Monday, Bloomberg printed an in-depth have a look at the rising dimension of America’s electrical energy payments, and the information behind it raises an eyebrow.
From Bloomberg:
The North American Electrical Reliability Corp., the nation’s grid safety regulator, forecasts that US energy demand in summer time will rise 224 gigawatts over the following decade — roughly the equal of including 180 million houses.
One analyst stated the final comparable surge got here throughout World Conflict II.
The AI buildout is a central driver of this improve.
Let’s have a look at an instance to drive this dwelling…
The Pennsylvania-New Jersey-Maryland (PJM) Interconnection is the nation’s largest energy grid, stretching from the Midwest to the East Coast. Over simply the three years ending in Might 2028, information facilities are projected so as to add a minimum of $23 billion to buyer payments on the PJM grid alone. That’s a rise of greater than 50%.
In the meantime, in components of jap Pennsylvania, electrical energy costs have already risen 200% since 2020.
Now, to offset this, in his February 24 State of the Union deal with, President Donald Trump informed America’s largest know-how firms they’d “have the duty to supply for their very own energy wants.”
Ten days later, Amazon.com Inc. (AMZN), Alphabet Inc. (GOOG), Meta Platforms Inc. (META), Microsoft Corp. (MSFT), OpenAI, Oracle Corp. (ORCL), and Elon Musk’s xAI gathered on the White Home and signed what the administration calls the Ratepayer Safety Pledge – committing to “construct, deliver, or purchase” all the ability and grid infrastructure required for his or her information facilities, with none of these prices handed to American households.
Trump was candid concerning the politics driving it. On the signing ceremony, he informed the assembled tech executives:
They want some PR assist as a result of folks suppose that if an information heart goes in there, electrical energy costs are going to go up.
So, that ought to settle it, proper?
Right here’s the half that doesn’t make the press releases
The pledge is principally forward-looking, however the value will increase Bloomberg documented have been constructing for years.
The PJM grid’s capability costs – what utilities should pay mills for electrical energy – exploded from $28.92 per megawatt-day within the 2024-’25 supply yr to $329.17 within the 2026-’27 supply yr.
That’s an 11X improve already baked into utility price buildings lengthy earlier than anybody signed something on the White Home. Nevertheless it’s nonetheless not sufficient – the latest PJM capability public sale fell 6.6 gigawatts wanting obtainable provide.
Plus, as I simply famous, the pledge addresses future information heart buildouts, not current ones.
So, right this moment’s energy payments replicate:
- The facility infrastructure already constructed to serve present hyperscaler services
- The speed will increase already accredited by utility commissions to pay for the associated grid upgrades
These prices are very actual, very huge, and baked in – and the pledge doesn’t unwind them.
Plus, even for brand spanking new initiatives, signing a pledge to construct your personal energy plant doesn’t make one seem. Allowing a brand new era facility takes two to 4 years. Building takes extra years on prime of that. And by the point the concrete is poured, the demand it was designed to fulfill has typically already doubled.
Which brings us to the funding implications…
When demand outruns provide, the winners aren’t all the time who you anticipate
Each main tech growth in historical past has finally run into huge demand for the vital elements associated to that know-how’s buildout.
Brian Hunt, editor of Money & Megatrends, has constructed a whole funding framework round this dynamic. He calls it the “AI demand shock.”
Right here’s Brian explaining the core idea:
For the previous three years, one of the best ways to make cash shortly in shares has been to find an business the place an AI “demand shock” is about to strike… after which make investments there earlier than the shock arrives.
Not a provide shock, thoughts you, the place a warfare or a pandemic abruptly cuts off the provision of a useful resource like oil.
As a substitute, I’m speaking a couple of “demand shock,” the place demand for a particular useful resource or manufactured product abruptly skyrockets… and sends its value a whole bunch of p.c increased.
The historic examples Brian cites drives dwelling his level.
In 2023, the AI demand shock for superior semiconductors despatched Nvidia up 525% in below two years. Across the similar time, the sudden want for information heart cooling methods drove Consolation Programs USA Inc. (FIX) up 1,000%. In the meantime, demand for superior optical methods – the elements that enable quick information switch between AI servers – drove Lumentum Holdings Inc. (LITE) up 1,164% in two years.
None of those firms are AI firms within the headline sense like OpenAI or Anthropic. As a substitute, they’re the picks-and-shovels suppliers to the AI buildout – the companies sitting upstream of the know-how, making the bodily issues the know-how couldn’t exist with out.
Brian explains why these strikes are typically so giant and so quick:
AI is advancing at such a fast tempo that AI-driven demand shocks at the moment are taking place yearly… and creating the quickest – and most profitable – inventory market strikes we’ve ever seen.
The standard manufacturing business wants five-to-10 years to construct operations able to assembly rising demand. Similar with mining industries that provide vital uncooked supplies.
However our new, lightning-fast technological cycles now transfer means, means quicker…
We now have loopy mismatches within the financial system’s interlocking and interdependent components.
It’s like we’ve a rocket engine connected to the drivetrain of a Toyota Corolla.
This mismatch between the rocket’s engine and the Corolla’s drivetrain is the place the funding alternative lives.
Brian’s present focus: the chemistry of AI
Whereas alternatives are in all places, Brian’s Tuesday issue of Money & Megatrends highlighted a sector that may shock you…
Chemical compounds.
Right here’s Brian:
The chemical compounds sector is often regarded as an “previous financial system” business that produces merchandise corresponding to plastics, paints, solvents, cleaners, and pesticides.
Nevertheless, some chemical companies are concerned in “model new financial system” actions associated to AI.
The chemical compounds business sits upstream of virtually each bodily element within the AI stack: from the specialty gases and supplies used to fabricate semiconductors to the excessive‑purity solvents, coatings, coolants, flame retardants, and superior polymers that make fashionable information facilities doable.
Each AI server depends on an extended chain of ultra-specific and ultra-pure chemical compounds. As AI utilization explodes, the necessity for extra chemical compounds and extra subtle, higher-purity ones will increase.
Brian’s March 6 recommendation so as to add chemical compounds to your portfolio is already paying off…
Chemours Co. (CC) is up 37% since that be aware and simply hit a brand new one-year excessive. And Component Options Inc. (ESI) has added 15% and in addition simply hit a brand new one-year excessive.
For readers who need Brian’s full evaluation – together with the particular names he’s anticipating linked to all of the demand shocks he’s monitoring past chemical compounds – his research is available for free in Money & Megatrends
Circling again to the token scarcity
Ben Pouladian’s commentary that opened our Digest – the world is especially wanting tokens – is true. The scarcity is actual, and it’s disrupting AI firms’ capability to serve their customers proper now.
However pull again far sufficient and also you’ll see the broader sequence…
- The token scarcity is downstream of a GPU scarcity…
- Which is downstream of an information heart scarcity…
- Which is downstream of an influence scarcity…
- Which is downstream of an infrastructure buildout that the bodily world merely hasn’t had time to finish.
That is what a real know-how transition seems like from the within. The demand arrives quicker than the provision chain can reply.
The businesses positioned in the midst of these constraint factors – those making the chemical compounds, cooling the servers, supplying the ability infrastructure – are the place the large cash is being made right this moment…and the place we need to be.
We’ll preserve monitoring this with you right here within the Digest.
Have a great night,
Jeff Remsburg
(Disclaimer: I personal LITE)
