The stock market has two stories at once. AI momentum still carries the loud trade, but broader market leadership is starting to matter. That is a better setup than a one stock circus, but it is also less forgiving.

Momentum is now the risk

The cleanest warning sign is not a crash. It is an extreme reading. The MSCI Momentum ETF has been outperforming the S&P 500 by more than four standard deviations, which is the sort of number that makes a statistician put down the coffee and check the data twice.

That does not mean the AI trade is finished. It means the trade has become crowded. When a factor moves that far ahead of the broad index, future returns depend less on the original thesis and more on who still has cash left to buy.

The SPDR S&P 500 ETF Trust, ticker SPY, remains a useful shorthand for the broad US market. But SPY is no longer just a passive mirror of the economy. It is also a container for a very large AI and data center bet. That matters when investors start talking about mean reversion in the same breath as long run growth.

Breadth is doing useful work

The better news is that market leadership has been broadening. Small caps and the equal weight S&P 500 have been outperforming, which means investors are not only chasing the largest technology names. A rally with more legs is usually less fragile than one carried by a handful of giants.

There is a macro reason this can happen. Falling inflation and steady job growth give the Fed room to be patient on rates. A patient Fed is not the same thing as an easy Fed. It simply removes one source of panic, which is often enough for investors to take another look at smaller companies and cyclical stocks.

This is the part of the bull case that deserves respect. If earnings outside mega cap technology keep improving, then a 2027 extension is not a fairy tale. It is a plausible path. Markets do not need perfection. They need enough cash flow and not too many nasty surprises.

AI costs are moving into the spreadsheet

The AI story still has a physical bill attached to it. Clean power purchase agreement costs in the US are expected to rise by 40 to 120 percent as demand from companies such as Google and Meta meets reduced federal clean energy support. That is not a small rounding error in a data center budget.

Power is where technology becomes industrial. The model may live in the cloud, but the cloud lives near transformers, power lines, cooling systems, and people arguing with local permitting offices. Very glamorous, if your idea of glamour is a substation.

Grid stress is also visible outside corporate contracts. Storms and heat near 40C left more than 800,000 households without power in the eastern US. Electricity prices rose as utility systems strained. For AI investors, that is not just weather. It is an input cost signal.

Chips are no longer a simple proxy

Memory chips have been one of the cleaner ways to express optimism about AI demand. That makes recent weakness more interesting. AI memory and chip stocks slumped as global peers sold off, with SK Hynix and Samsung weighing on the KOSPI.

Micron also came under pressure as short interest talk returned to the name. The point is not that one chip stock defines the cycle. The point is that the AI supply chain is no longer trading as a single line going up and to the right.

That is healthy in the long run, but awkward in the short run. Investors are starting to separate compute demand, memory pricing, power capacity, capital spending, and end demand. Good. Markets should eventually learn arithmetic. They usually do, after charging tuition.

What to watch

First, watch whether breadth lasts. If small caps and the equal weight S&P 500 keep leading while SPY holds up, the bull market has a sturdier base. If breadth fades and mega cap AI names regain all the burden, concentration risk is back.

Second, follow the Fed through inflation and jobs, not through wishful rate cut calendars. Steady hiring and cooling prices give policymakers patience. A shock in either direction would change the equity story quickly.

Third, watch the boring infrastructure numbers. Power contracts, grid reliability, data center capital spending, and memory prices will say more about AI margins than another round of polished product demos. The market can still climb into 2027, but the next stage will be more selective. That is usually what happens when a great story meets an actual bill.