Insight & Outlook: Fidelity Market Signals Weekly

Midterms and the AI trade: Speed bumps, not roadblocks

The trigger of the recent sell-off was political, not fundamental. And while it did reprice the pace of the AI buildout, it did not diminish the demand behind it. For investors, the key now is to consider which parts of the AI trade the market is mispricing.

What did the markets actually reprice?

The move after midterm elections was violent and indiscriminate. On Nov. 5, the Philadelphia Semiconductor Index fell roughly 10%, its worst session since March 2020, and the Nasdaq Composite dropped about 5.5%, confirming a correction from its Oct. 6 high. Hyperscalers, chipmakers, equipment suppliers, and neoclouds were sold in unison.

The trigger was the election results, not fundamentals. Democrats won control of both chambers of Congress, and governor races in Texas, Ohio, and Georgia went to candidates who campaigned on restricting data-center construction. Voters sided against data centers in 40 of 41 local ballot measures. Texas alone accounts for roughly a quarter of planned U.S. data-center capacity, and its governor-elect has pledged a moratorium on new projects. Projects already authorized are expected to continue.

The bear case has quietly changed shape. Through the summer, the worry was overbuilding of data centers. Now it is whether demand arrives fast enough to justify the spend at a slower, costlier pace. Note the irony: Constraint on new supply is the natural remedy for a potential overbuild.

Is this a verdict on AI demand, or on the pace of the buildout?

We don’t believe the election changed the demand outlook. Three points argue that the data-center story is bent but not broken.

• 2027 is largely built. Most of the data-center capacity scheduled to come online next year is already under construction and sits largely outside the scope of proposed moratoriums. The real exposure is 2028 and beyond, a later and narrower risk than current prices imply.
• Capex is still rising. Alphabet raised its 2026 capex guidance twice this year, first to $195 billion, and then to $205 billion, and told investors last month that 2027 spending will significantly increase. Google Cloud revenue more than doubled year-over-year in the third quarter. Over the past quarter, 2027 capex estimates moved from roughly $247 billion to $286 billion for Amazon and from about $250 billion to $318 billion for Alphabet. The companies writing the checks have not blinked.
• Scarcity cuts both ways. When it becomes harder for new data centers to acquire permits and local approvals, capacity that is already operating becomes more valuable. That helps explain why hyperscalers and owners of existing capacity held up relatively well, even as the market painted the complex with a wide brush.

Do semiconductor fundamentals justify the drawdown?

Price and fundamentals have rarely been this far apart. Semiconductor industry revenue has compounded at roughly 33% annually over the past three years, a pace in the 94th percentile of all three-year periods since 1970. Over the same stretch, the sector’s multiple has compressed to below that of the broader market. That combination has historically been constructive.

When elevated sales growth has been paired with a below-market multiple, semiconductors have outperformed over the following 12 months about 91% of the time since 1970, by an average of 32% (Exhibit 1). We don’t read this as a reason to buy the index wholesale. Base rates are not guarantees, and the historical sample does not include a politically driven supply shock of this kind. But it does suggest the market is pricing a deeper earnings disappointment than the revenue trend supports.

Why does a high-correlation sell-off favor active management?

Last week tested the idea of a multilayered AI stack, and the notion that the AI trade was broadening. High correlations on the way down are typically when dispersion increases on the way back up.

The pressure points are specific. The names most exposed to a slower, costlier buildout are debt-funded neoclouds, a Texas-heavy population of data-center developers, power, cooling, and electrical equipment suppliers, semiconductors and equipment, and the application software and IT services. Holding up relatively well are hyperscalers, infrastructure software and cybersecurity, health care and biotech, and non-AI sectors such as energy, staples, and utilities.

Similar to the recent path of consumer spending, AI market performance may be K-shaped. The indiscriminate rally that defined the first half of 2026 appears to be giving way to a market that is increasingly rewarding fundamentals. That differentiation is arguably healthy. Selectivity matters across both stocks and bonds. We favor companies that can fund growth without relying on favorable financing conditions or tax incentives, a filter that applies as much to data-center credit exposure as it does to equities.

Is this red versus blue, or a global race?

Data-center skepticism is bipartisan; so is the AI race. Texas’s Republican governor paused new large-load grid approvals months before the election, and local opposition has crossed party lines nationwide. At the same time, policymakers in both parties share an interest in keeping the U.S. competitive in an increasingly global race for AI leadership. That tension tends to resolve through negotiation rather than prohibition: community-benefit agreements, local investment commitments, and rules requiring data centers to cover their own grid costs, which over time could even ease pressure on household electric bills. We believe the debate is about the terms of the buildout, not whether it happens.

What would change our mind?

Capex guidance is the cleanest signal. An explicit pullback from a major hyperscaler would be the clearest evidence that demand, not politics, is the binding constraint. So far, guidance is still moving up.

Fiscal policy is the sleeper risk. A potential debt-ceiling negotiation early next year could put 100% bonus depreciation for data centers on the table, a plan favored by the Trump administration that would allow more immediate, more favorable tax treatment. Recent Senate proposals would deny full expensing to certain new AI data centers, which would raise the after-tax cost of the buildout and weigh most on marginal, debt-funded builders.

The bottom line

The midterm elections introduced speed bumps, not roadblocks. Data centers are still moving forward, at a slower and possibly healthier pace, and the fundamentals underneath the AI complex are stronger than the recent price action suggests. The opportunity lies in separating short-term sentiment from long-term fundamentals, which is where active management can add value. We view the policy debate as a question of global competitiveness rather than red versus blue, acknowledge the genuine uncertainty around 2028 capacity, and resist the conclusion that the AI trade is over. It is simply no longer a trade where everything wins together.