Is the “AI bubble” about to burst in late 2025 or 2026?

Market focus And pricing

A few big technology companies now dominate stock indexes.

• The largest technology platforms have an unusually high share of the S&P 500 and global indices.

• AI stories explain the majority of the stock market’s gains since late 2022.

• A minor shock, such as a surprise competitor or regulatory move, can move trillions in market value in a single day.

The DeepSeek incident in early 2025, where a cheaper model from China briefly wiped out massive amounts of market value, showed just how fragile sentiment can be. When the narrative changes, it can move very quickly.

Spending Which exceeds current returns

Capital spending on AI infrastructure has entered historic territory.

• Big tech companies collectively spend hundreds of billions of dollars annually on data centers, graphics processing units, and power.

• Some projections place AI-related capital expenditures at more than $500 billion annually for several years.

• In contrast, revenues from direct AI services are still much lower, and in some sectors are measured in the tens of billions rather than the hundreds.

Consulting reports and research line up around one awkward point: Most companies experimenting with generative AI are not yet seeing a significant impact on their profits and losses.

• Extensive studies have found that the majority of AI initiatives show little or no measurable ROI yet.

• Many projects improve individual productivity, but not gross margins or revenue growth.

• AI often remains stuck in experimental mode, not integrated into operations.

You can justify the huge early investment for a while. You can’t do this forever if the profit story remains vague.

circular And aggressiveness Finance

Some AI contracts and investments appear designed to keep the music playing.

• Vendors pre-purchase large blocks of cloud capacity from each other.

• AI labs are committed to spending huge sums on specific infrastructure providers.

• These liabilities then appear as future revenue growth on the provider’s side, even if the buyer does not yet have a direct way to recover that money.

This is not a scam, but it creates a feedback loop where the rosy assumptions on both sides reinforce each other. If one piece cracks, the ring can quickly come undone.

Physical limitations: Energy, cooling and earth

Artificial intelligence is no longer just software. These are concrete, copper and megawatts.

• Modern AI data centers can consume as much electricity as a large city.

• Local networks, water supplies and permitting processes began to collapse.

• Governments and regulators are questioning whether unlimited AI construction is compatible with local climate and infrastructure goals.

If power or cooling becomes a hard limit in key areas, some existing capital expenditure plans will have to be scaled back. This type of hard stop is a classic trigger for asset repricing.