Listen to the audio version of this article (generated by artificial intelligence).
Why is everyone arguing about the AI bubble when they should be watching prices… What are two top tech CEOs pleading for… Watch this ‘escalator’…
Before we start today, a reminder about tomorrow morning Breakthrough 2026 The event is at 10 a.m. ET with legendary investor Louis Navellier and TradeSmith CEO Keith Kaplan.
They will cycle through TradeSmith’s seasonal tool. It scans more than 5,000 stocks across decades of price history, looking for one thing…
Time windows in which a stock has historically risen—or fallen—remarkably consistently.
Through an 18-year backtest, trading within these windows alone produced 857% of overall growth—more than double the S&P 500 over the same stretch, and the strategy still led even in 2007, the worst year in the backtest.
when Sign up to join Louis and KeithYou’ll get free access to the seasonality tool. Give it a spin in your own wallet.
Then, tomorrow morning, Keith will explain why he believes the period beginning around July 23 could represent an important shift in market leadership – and how to use the tool to capitalize on it. This relates to how Lewis used these timing signals with his stock rating system.
There will be plenty, including some free stock recommendations. Just click hereAnd see you tomorrow morning at 10 a.m. ET.
Everyone wants to discuss whether AI is a bubble
This is the wrong question.
The question of who wins and who loses – and what happens in your portfolio – is a much more mundane question…
How much does it cost to buy AI token?
To make sure we’re on the same page, a token is the basic unit by which AI models are billed – roughly a few characters of text. It’s a counter that runs every time a chatbot answers a question, an AI agent completes a task, or a program calls a form behind the scenes.
Think of nominal prices the same way airlines think about jet fuel or the way manufacturers think about steel. It is the cost of basic inputs to the AI economy.
Today, these costs are falling for a simple reason…
AI models are becoming dramatically more efficient, while competition among providers – including a wave of open source alternatives – is driving down prices. Every major AI lab is racing to deliver more intelligence for fewer dollars.
Whatever you think about valuations, this is the number that ultimately decides who wins and who loses within a trade.
And now, it’s falling apart.
The number has already been broken
In March 2023, running the best OpenAI model available cost roughly $30 to process 1 million tokens — roughly the amount consumed by a long AI conversation or thousands of simple prompts.
Today, performance of similar quality ranges from a few cents to a few dollars – a decline of 90% or more in just over three years, and continues to decline.
Now, despite this collapse, overall enterprise spending on AI has not declined. By most accounts, it is Tripled. Companies use it widely because It has finally become affordable enough to spread throughout the business.
Chatbots have given way to autonomous agents who iterate, recheck their work, and call on external tools dozens of times to finish a single task. Each ring consumes tokens. Therefore, even as unit prices decline, total usage grows faster than prices decline.
This is exactly why the AI infrastructure — most of which is demand for chips and computing — hasn’t cracked yet.
Prices fell, but overall spending rose. This contradiction is the whole game.
But no matter how much prices have collapsed so far, two powerful AI CEOs believe prices need to fall significantly further.
AI is still very expensive
last week, Palo Alto Networks (Bano) CEO Nikesh Arora continued CNBC Let’s say, in effect, that the current price of AI token is holding back enterprise adoption.
Here’s Aurora to clarify:
I think 54% is a good start… and I think we probably need another turn to achieve it.
This 54% was a reference to OpenAI’s claim that its latest model has 54% more code efficiency than the last model.
Aurora’s point of view: Nice start, not enough at all. Later in the same interview, he said it more clearly:
We need to see AI prices come down.
palantir (Belter) CEO Alex Karp went further the previous week, calling the token pricing model completely broken:
I’m not throwing shade at them, but something went completely wrong.
The basic view among businesses in this country is that I’ll just chill out and waste my time with codes.
In other words, customers don’t want to think about tokens. They just want AI that is cheap enough to be used everywhere.
It is assumed that two CEOs run two companies Winning During the AI boom, they publicly complain that AI costs too much to use on a large scale. This is not hype.
These are two big AI agents that tell you where the ceiling is today – and where things are headed tomorrow.
What our Eric Fry is watching
Our global expert, Eric Fry, editor Fry investment reportwas tracking a specific driver behind the price pressure: competition from open source models, including Chinese labs such as Z.aithey are deliberately tuned to run on older, cheaper chips rather than newer chips.
They obtained results that were almost indistinguishable from advanced Western models by several months.
Here’s Eric to explain:
Code costs have fallen by about 20% since the beginning of June, reducing what data centers can charge for computing power.
Data Center Company shares reflect this trend Core Wave Company (CRV) It’s down nearly 40% in the past two months.
I will point out that the CoreWeave chip has more than one story behind it – reports on dead (dead) Building its computing-for-rent business has received most of the mainstream press attention.
But Eric believes that falling token prices are the primary force behind this story – an alternative explanation.
Running against the escalator
Now, what is the impact for investors?
Well, let’s first understand the scene through an analogy – trying to climb a descending escalator.
Usage growth is like walking up this escalator. Declining token prices are like an escalator moving down under your feet.
Right now, you’re going up faster than you’re going down the escalator, so you’re still making progress toward the top — and total AI spending continues to rise. This is a win for AI infrastructure companies, and somewhat of a win for companies that want cheap AI.
But if the elevator speeds up (prices fall faster) or your legs tire (usage growth matures), the elevator wins, and you’re carried down instead of up. That’s not a win for AI infrastructure companies, but it’s a big win for companies that want cheap AI.
Right now, usage dominates, it is growing faster than prices are falling, which is why demand for infrastructure remains strong even as the economics of each unit erode beneath it.
Our growth investing expert Louis Navellier, editor Growth investorprovided evidence of how real the “use wins” phase still is…
yesterday, ibm (IBM) It issued an unexpected preliminary second-quarter earnings report and profit warning, leading to the worst single-day drop in the company’s shares in its history.
But this was not a problem for Lewis. Here he explains why:
Earnings season is off to a very good start. I know International Business Machines missed it, but they missed it because they are losing market share to data centers.
So, good for us, guess what we have?
Lots of data center related stocks.
This is Lewis – one of the best analysts in our industry – who knows exactly where the money is flowing today and successfully manages the escalator.
But as Lewis knows, and will eventually take into account his recommendations, this same elevator will win in the end.
In other words, at some point in the future, token prices will fall enough, or usage growth will mature enough, that the balance will tip — and when that happens, AI trading will reach a major inflection point.
To join Lewis Growth investor So you can navigate with him through this transformation, Click here to learn more.
Here’s the big version of the people on each side when that happens
Exposed to: Companies that have built specialized and expensive infrastructure rent computers by the unit. We’re talking about chipmakers, operators of new cloud data centers, and any hyper-fast company that sells raw processing power because its pricing power depends on scarcity.
When computing stops being scarce, that pricing power goes with it.
Help: What Eric calls “AI implementers” – companies that embrace AI as a tool within their existing business rather than selling computing as a product, expanding their margins every time their AI bill shrinks.
And the largest pool of value lies in a furthest layer: mainstream organizations in finance, healthcare, retail, and industry, whose AI costs go from a budget headache to a rounding error, opening the door to productivity growth that shows up as real profit growth rather than a larger technology bill.
Eric has already recommended specific AI implementers in Fry investment report– Companies are in a position to catch this margin expansion before the rest of the market catches up. To find out what they are, click here to learn about joining.
There’s one category I won’t even pretend to have a clear answer for: traditional SaaS companies. These are the companies that suffered a “SaaSmageddon” earlier this year.
In theory, cheap AI should allow legacy software platforms to gain powerful features without inflating their profit margins. But cheap AI also lowers the barrier that prevents customers – or smart competitors – from building capabilities that software vendors used to charge dearly for.
Whether SaaS incumbents will end up net winners or victims of the price collapse itself is not really resolved, and will likely depend on the specific company, not the sector.
We are monitoring this closely and will keep you updated.
How to monitor all this
No one knows exactly when this escalating drama will reach its tipping point, and we won’t pretend we do.
Instead, watch if major AI labs start reporting increasing margins even as prices continue to fall — for example, OpenAI announced that the cost of a query service fell faster than the price it charges for a single query. This means that the efficiency gains outweigh the price cuts, which is a bad sign for hardware-heavy names like nvidia (NVDA).
And also watch to see if the software companies that benefit from all this actually start showing cheaper AI in their reported profit margins, not just their marketing.
Bottom line
The token collapse is not a one-time event, and it is not over yet.
This has already happened, is still happening, and by Arora’s own calculations, will need to happen even more before AI adoption becomes wide open.
What has yet to be resolved is not the trend, but how long usage will outpace the price collapse, and when it eventually reverses, what will happen if infrastructure trade remains priced such that the demand side wins forever.
This is the issue that decides who wins and who loses here. It’s not a question of whether “AI is just a bubble.” It’s not just some abstract multiple of some chip stock that everyone is already arguing about on TV.
This issue requires more analysis and action, which is exactly why it is the issue that most people watching from the sidelines will not grapple with…
But that’s you He should Grappling with if you have money in artificial intelligence.
We will keep you updated.
I wish you a good evening,
Jeff Remsburg
