Sorry about this one being a little rough, so I’ll keep it really short. It’s time to head to bed and I still wanted to push this one out while its ideas are fresh in my mind.
Just tell me if you want me to flesh it out more in the comments and I’ll spend more time on it. But the main ideas are here.
The entire AI bubble debate misses the point. Is it in a bubble? Is it not? Completely besides the point!
We’re not dealing with one bubble that either exists or doesn’t. We’re watching three separate bubbles operating by different rules, with different timelines, and different outcomes for anyone trying to navigate this mess.
The investment bubble is classic financial madness
This one’s straight out of the dot-com playbook. OpenAI goes from $150 billion to $340 billion valuation while burning $14 billion annually and projecting $44 billion in losses through 2028. NVIDIA trades at 29x sales, Palantir hits 69x—numbers that would make 1999 blush.
Where we are in the cycle right now is very similar to where we were between 1998 or 1999.
— Ray Dalio
The billionaire fund manager isn’t even being dramatic. The metrics are there. The Magnificent Seven now represent 60% of S&P 500 earnings growth, creating the kind of concentration that historically ends badly.
But unlike dot-com darlings that lived entirely on venture funding, today’s AI winners include cash-rich giants like Microsoft and Google. They can fund AI development from actual profits rather than burning through investor money.
This structural difference might moderate the crash, though it won’t prevent the valuation reset that’s coming.
The infrastructure bubble might actually help everyone
Data center AI chip spending hit $125 billion in 2024, projected to reach $370 billion globally by 2025. NVIDIA dominates with 92% market share and 78% gross margins—unprecedented for semiconductors. Sequoia Capital calls this the “$600B question”, highlighting the massive gap between infrastructure investment and actual returns.
Yet this bubble could benefit most organizations. When telecom companies overbuilt fiber networks in the late 1990s, the crash created abundant, cheap bandwidth that enabled the digital transformation of the 2000s. Today’s AI infrastructure buildout might follow the same pattern—overcapacity leads to falling prices and broader adoption.
GPU stockpiling and supply normalization signal we’re entering the late stages. Secondary markets show extreme premiums, suggesting speculative rather than productive behavior. But the infrastructure will outlast the speculation.
The hype bubble is the real problem
This one’s dangerous because it distorts business decisions. MIT research shows 95% of AI pilot projects fail to generate returns. Not because AI doesn’t work, but because organizations deploy it wrong—chasing technology rather than solving problems.
Hundreds of millions of people have tried ChatGPT, but most of them haven’t been back.
— Benedict Evans
Initial curiosity doesn’t translate to sustained value. Enterprise adoption tells the same story—78% of organizations experiment with AI, but only 1% achieve mature implementations.
Are we in a phase where investors as a whole are overexcited about AI? My opinion is yes.
— Sam Altman
When the OpenAI CEO acknowledges overexcitement, you know the hype has disconnected from reality.
Different bubbles, different timelines
The investment bubble faces immediate pressure from traditional valuation metrics and potential interest rate changes. Public market corrections could happen quickly, especially if earnings don’t justify current multiples.
The infrastructure bubble depends on utilization catching up to capacity. This could take 2-3 years as organizations slowly adopt AI applications that actually work. The timeline favors patience over panic.
The hype bubble requires the hardest correction—moving from promise to delivery. Enterprise AI needs to evolve from pilot projects to production systems that solve real problems. This transformation could take 3-5 years and separate genuine applications from marketing materials.
Why this framework matters
Understanding these three distinct dynamics will let you make your decisions better informed. The investment bubble mostly affects public market portfolios and VC funding—important for investors, irrelevant for most business decisions. The infrastructure bubble might create opportunities as prices fall. The hype bubble demands the most attention because it shapes how organizations approach AI implementation.
Smart companies ignore the investment circus, prepare for infrastructure price drops, and focus on problem-first AI deployment. While everyone else debates whether “the AI bubble” exists, they’re systematically implementing solutions that actually work.
The market correction of 2025-2026 will likely hit all three bubbles differently. Investment valuations will reset. Infrastructure utilization will slowly improve. Hype will give way to realistic expectations about what AI can and can’t do.
The winners won’t be those who called the bubble correctly. They’ll be the ones who understood which bubble mattered for their specific situation and acted accordingly.
Early morning thoughts... Recently portfolio managers have changed their recommendations on holdings increasing gold holdings up to 20% when previous recommendations were zero or maybe 2%. If this were to occur, a shift to gold, it would be a boom for gold. But it might crash AI and everything else when all that money, When asset allocation was shifted into gold requiring the sale of stocks and bonds. Imagine what might happen if this shift to gold was 20% and immediate. It would certainly hit the gold derivative's market which would then hit other derivatives and then you see a spiral unraveling... Just thoughts. Or am I mistaken.
and the bubble doesnt pop... As long as you can print and gift money for free...