Michael Burry of The Big Short fame is predicting artificial intelligence (AI) stocks are set to crash.
I agree with Burry that there’s an AI collapse coming.
But not the one he’s expecting.
After surging throughout the 1990s, tech stocks plunged 80% starting in 2000.
I don’t think that’s what’s about to happen with AI stocks. I explain why in our “AI Endgame” event—you can catch the replay right here.
The more important dot-com “crash” for investors to study was the collapse in costs.
In the early '90s, launching a tech company was a millionaire’s game.
You needed your own servers, data centers, and expensive software licenses. A single website could cost hundreds of thousands just to keep online.
Back in 1998, sending data across the web was expensive. It cost roughly $1,200 per megabit per month. Today, it’s pennies. The cost of moving data through fiber has fallen by roughly 10,000X:
Source: ChatGPT
Telcos laid 80 million miles of fiber, most of it “dark” and unused. This glut caused bandwidth prices to collapse 90%.
That price crash unlocked a wave of creativity the world had never seen. All of a sudden, you could launch a billion-dollar business out of a college dorm room.
Think about the all-time-greatest internet companies. The biggest moneymaking stocks of our generation. All were born in that era.
Netflix (NFLX)—online video streaming.
Google (GOOGL)—online advertising.
|
Meta Platforms (META)—online socializing.
The real opportunity of the internet wasn’t in the cables and modems. It was in what you could build on top of them.
And none of it would’ve happened without costs plunging first.
The cost of generating AI tokens is in freefall.
If you’re hearing about AI tokens for the first time, they’re the most important metric you’ve never heard of.
When you talk to ChatGPT, it doesn’t read your words like a human. It breaks them into tiny chunks called tokens. Each token is like a puzzle piece the AI studies to understand your question.
You can think of tokens as the fuel for the AI’s brain. The more complicated or longer your request, the more tokens it burns through to give you an answer.
Asking the AI to write a short recipe for pancakes might use a few hundred tokens. But asking it to code an entire website or write a detailed research paper could burn through millions.
The more tokens flying through the system, the more chips, compute, and electricity are being burned to make it all happen.
The cost to use AI is collapsing by about 10X every year. A task that cost $60 per million tokens in 2021 costs about six cents today. Running an AI model like ChatGPT is now 1,000X cheaper than it was just three years ago!
Source: ChatGPT
In the 1990s, engineers figured out how to push more bits through the same fiber using tricks like new lasers and smarter routers.
Today, AI engineers are learning to push more intelligence through the same hardware with better algorithms and more efficient chips.
We know when broadband got cheap enough. It unlocked new “apps” like YouTube, Netflix, and Spotify (SPOT).
Now, we’re creating intelligence “too cheap to meter.” Popular AI companion app Character.AI serves roughly 20,000 queries/second and has been able to cut costs by 33X since 2022.
Cheap cognition will mint the Googles and Amazons (AMZN) of AI, companies that turn all this infrastructure into pure profit.
When costs collapse, innovation takes off.
It’s a simple law that drives every major technology boom.
Bandwidth costs crash, and the result was trillions of dollars in new internet companies.
The same happened in space. Elon Musk’s SpaceX slashed the cost of launching rockets into orbit by 96% over the past decade. The number of space launches has surged 10X since 2014.
Source: Our World in Data
The first phase of the AI boom was all about the infrastructure: chips, server racks, cooling, and everything else that goes into high-end data centers.
Congratulations to Disruption Investor and Disruption_X members who made good money from AI infrastructure stocks.
But Phase 2 belongs to the AI-native companies taking advantage of collapsing AI costs to disrupt whole industries.
Imagine lawyers using AI to close contracts in seconds. Biotech firms designing new drugs in weeks instead of years. Factories using AI systems that optimize production without human input.
It’s already happening.
The easy money in AI infrastructure has been made. It’s time to shift our attention to the next group of winners.
If you want to learn more about this opportunity, watch the replay of my “AI Endgame” event.
I also discuss one of my favorite AI-native stocks in a special report.
You can access it all by clicking here.
Stephen McBride
Chief Analyst, RiskHedge