Amoroso Sees the AI Trade Moving Past Chip Stocks

Strategist Anastasia Amoroso said the AI trade is shifting away from semiconductor stocks toward hyperscalers, power and software, where she sees the strongest remaining opportunities for investors.

The most crowded trade in the market is being asked to prove it can survive a change of leadership. Anastasia Amoroso argues it can — just not through the same stocks. In remarks to TheStreet, Amoroso said the artificial intelligence trade is shifting away from chip stocks, and pointed instead to three areas she considers overlooked: hyperscalers, power, and software.

That is a compact thesis with a lot inside it. It says the money already made in semiconductors reflected the buildout phase — the hardware bill for training large models — and that the next phase of returns comes from the layers above and beneath that hardware: the companies renting out the compute, the utilities and generators supplying the electricity that makes it run, and the applications that eventually have to justify the whole thing with revenue.

Why the chip leg gets harder from here

Semiconductors were the cleanest way to own AI because the spending was visible, immediate and concentrated. Order books filled, and the market paid for them. The problem with a trade that obvious is that expectations catch up. Once a sector is priced for continued acceleration, incremental good news stops moving the stock and any wobble in the capital expenditure cycle moves it a great deal.

Amoroso's framing does not require chip demand to fall. It only requires that the marginal dollar of investor return move elsewhere — that the surprise, in other words, comes from parts of the chain that have not already been repriced. That is a rotation call, not a bearish one, and it is the sort of call that tends to be made when index-level gains flatten out and performance has to be earned through selection rather than exposure.

The tape is not doing the work anymore

The market backdrop as of the last trade on Wednesday, 26 August 2026 at 19:55 GMT was almost perfectly inert. The S&P 500, tracked by the SPDR S&P 500 ETF Trust (NYSEARCA: SPY), was at $766.25, up 0.04% on the day from a prior close of $765.91, inside a day range of $763.93 to $767.35. The Invesco QQQ Trust (NASDAQ: QQQ), which tracks the Nasdaq 100 and is the closest thing to a listed proxy for the AI complex, sat at $711.19, up 0.07%, having traded between $707.97 and $713.02. The SPDR Dow Jones Industrial Average ETF Trust (NYSEARCA: DIA) was the laggard at $534.41, down 0.16%.

The gap between the Nasdaq 100 proxy and the Dow proxy on the day came to about 0.23 percentage points — a rounding error, and that is exactly the point. When the tech-heavy index and the broad index are separated by fractions of a percent, the beta trade is not paying. Sessions like this one are where rotation arguments get made, because there is no index-level drift to hide inside.

Hyperscalers, power and software: what each leg actually depends on

The three areas Amoroso identified are not interchangeable, and each carries a different risk.

• Hyperscalers — the large cloud platforms that buy the chips and rent out the capacity. The bull case is that they capture the recurring revenue stream from AI workloads rather than a one-time hardware sale. The bear case is that they are the ones absorbing enormous capital spending, which pressures free cash flow even when revenue is growing quickly.
• Power — data centres consume electricity on an industrial scale, and that demand lands on utilities, independent generators, grid equipment makers and anyone with contracted supply near the right substations. This is the least AI-native of the three legs, which is part of the appeal: the businesses are regulated or contracted, and the demand growth is physical rather than speculative. The constraint is that utilities move slowly and rate cases take years.
• Software — the layer that has to convert model capability into money customers will actually pay. Software has lagged in the AI narrative partly because investors have worried that AI disrupts incumbent applications rather than enriching them. Amoroso's inclusion of it implies the opposite read: that the monetisation phase favours the companies with distribution and customer data already in place.

What would confirm the rotation, and what would break it

Rotation theses are testable. If Amoroso is right, the evidence shows up in relative performance over weeks and months, not in a single session: hyperscaler, utility and enterprise software shares outperforming semiconductor shares on the same headlines. Capital expenditure guidance from the cloud platforms becomes a bullish signal for the power complex even when the market treats it as a cost for the spender.

What would break it is straightforward. If AI capital spending decelerates outright, the chips fall first but nothing in the chain is spared — hyperscalers cut orders, power demand forecasts get trimmed, and software budgets tighten alongside everything else. A rotation only works if the total pie keeps growing and the slices change hands. Investors should also be alert to the possibility that the power leg becomes crowded in its own right; a trade described as overlooked stops being overlooked once enough people act on the description.

For readers positioning around this, the practical question is not whether to own AI but which part of the value chain they are being paid to own at current prices. That answer changes faster than the narrative does, and on a day when the two headline benchmarks moved less than a tenth of a percent apiece, the market is offering no help in deciding.

Frequently asked questions

What did Anastasia Amoroso actually say about the AI trade?

Amoroso said the AI trade is shifting away from chip stocks. She identified hyperscalers, power and software as overlooked areas where she sees the strongest new opportunities for investors. She did not call for AI spending to fall — the argument is that the best returns now come from different parts of the value chain than semiconductors.

Why would semiconductors stop leading the AI trade?

Chips were the most direct way to own the AI buildout, so they repriced first and fastest. Once a sector trades on expectations of continued acceleration, good news stops moving the shares while any disappointment moves them sharply. Rotation arguments assume the growth continues but the surprise, and therefore the return, shifts elsewhere.

What are hyperscalers and why do they matter here?

Hyperscalers are the very large cloud computing platforms that buy AI hardware at scale and rent computing capacity to customers. The bull case is recurring revenue from AI workloads rather than one-off hardware sales. The offsetting risk is that they carry the capital spending burden, which pressures free cash flow even as revenue grows.

How does electricity fit into an AI investment thesis?

AI data centres consume electricity on an industrial scale, which pushes demand onto utilities, independent power producers, grid equipment suppliers and anyone with contracted generation near suitable sites. That demand is physical and contracted rather than speculative, though utilities move slowly and regulatory rate proceedings can take years to conclude.

What did markets do on the day of these comments?

As of the last trade at 19:55 GMT on 26 August 2026, the S&P 500 proxy SPY was $766.25, up 0.04%. The Nasdaq 100 proxy QQQ was $711.19, up 0.07%. The Dow proxy DIA was $534.41, down 0.16%. All three were essentially flat, offering little directional signal.

What would prove the rotation thesis wrong?

An outright deceleration in AI capital spending. A rotation requires the total spending pie to keep growing while the slices change hands. If cloud platforms cut orders, chip makers fall first but power demand forecasts get trimmed and software budgets tighten too — every leg of the chain suffers together rather than rotating.

Sources

• The AI trade is changing — here’s where to look for opportunity next — TheStreet

Photo: Robert So · Pexels Licence — source