Market Concentration and Pricing
A few large tech companies now dominate stock indexes.
-
The biggest tech platforms hold an unusually high share of the S&P 500 and global indexes.
-
AI stories explain the majority of stock market gains since late 2022.
-
A slight shock, such as a surprise competitor or regulatory move, can move trillions in market value in a day.
The DeepSeek episode in early 2025, where a cheaper model from China briefly erased vast amounts of market cap, showed just how fragile sentiment is. When the narrative changes, it can move very fast.
Spending that Outruns Current Returns
Capital spending on AI infrastructure has entered historic territory.
-
Big tech companies collectively spend hundreds of billions of dollars per year on data centers, GPUs, and power.
-
Some projections have AI-related capex exceeding $ 500 billion annually for several years.
-
In contrast, direct AI service revenue is still much smaller, and in some segments it is measured in tens of billions rather than hundreds.
Consulting and research reports line up on one awkward point: most enterprises experimenting with generative AI are not yet seeing a significant impact on their P&L.
-
Extensive studies find that the majority of AI initiatives show little or no measurable ROI so far.
-
Many projects improve individual productivity, but not overall margins or revenue growth.
-
AI is often still stuck in pilot mode, not embedded deep in operations.
You can justify heavy early investment for a while. You cannot do it forever if the profit story stays vague.
Circular and Aggressive Financing
Some AI contracts and investments seem designed to keep the music playing.
-
Vendors pre-buy large blocks of cloud capacity from one another.
-
AI labs commit to spending giant sums on specific infrastructure providers.
-
Those commitments then appear as future revenue growth on the provider side, even if the buyer does not yet have a straightforward way to recoup that money.
This is not fraud, but it does create a feedback loop in which rosy assumptions on both sides reinforce each other. If one piece cracks, the loop can unwind quickly.
Physical Constraints: Energy, Cooling, and Land
AI is no longer just software. It is concrete, copper, and megawatts.
-
Modern AI data centers can consume as much electricity as a large town.
-
Local grids, water supplies, and permitting processes are starting to creak.
-
Governments and regulators are asking whether unlimited AI buildout is compatible with climate targets and local infrastructure.
If power or cooling becomes a hard limit in key regions, some of the current capex plans will need to be scaled back. That kind of hard stop is a classic trigger for asset repricing.
