JPMorgan AI Trade Rebound Favors Chips After the Selloff

JPMorgan says the global AI selloff has created the conditions for renewed buying, despite growing doubts about technology spending and investment returns. The JPMorgan AI trade rebound thesis rests on cleaner positioning, lower valuations, strong earnings, and continued demand for semiconductor capacity.

The call does not predict a return to the broad technology rally investors previously enjoyed. Instead, JPMorgan sees a more selective recovery, with semiconductor companies positioned better than software businesses. That distinction turns a bullish headline into a sharper argument about where AI spending creates measurable value.

The debate now pits infrastructure suppliers against software companies exposed to intensifying competition and uncertain pricing. Investors must decide whether the recent decline corrected crowded positions or exposed a deeper problem with AI economics.

The JPMorgan AI Trade Rebound Starts With a Cleaner Market

JPMorgan argues that the selloff removed some market excess without breaking the fundamental case for AI infrastructure.

The bank delivered that assessment in a September 28 note led by equity strategist Mislav Matejka. According to a market summary, the analysts said investor positioning had improved and valuations had fallen across most AI-linked groups.

Positioning describes how heavily investors already own a stock, sector, or trading strategy. Crowded positioning can amplify a decline because many investors attempt to reduce similar holdings simultaneously.

A cleaner market therefore does not mean the underlying businesses suddenly became stronger. It means fewer investors remain trapped in an overcrowded trade, reducing one source of forced selling.

JPMorgan also expects capital spending to remain strong despite concerns about a slowdown. That spending includes processors, memory, networking equipment, data centers, and the electrical infrastructure required to operate AI systems.

The analysts tied their constructive view to continued earnings strength and growing evidence of AI monetization. Monetization means converting AI adoption into revenue, cost savings, or customer retention that can support the investment behind it.

Those conditions form the economic core of the JPMorgan AI trade rebound. Lower share prices improve prospective returns only if earnings remain resilient and customers continue paying for AI products.

The note also acknowledges an important limit. Matejka’s team said technology stocks might not regain their previous level of success, even though the fundamental outlook remains constructive.

That distinction matters because a rebound does not require every former winner to recover. Market leadership can narrow toward businesses with scarce products, stronger pricing, and visible orders.

The latest call follows several periods of sharp volatility during 2026. In July, J.P. Morgan Asset Management said some leading semiconductor stocks had fallen between 20% and 40% within one month.

Its AI boom assessment argued that demand was still compounding while financially strong hyperscalers funded the infrastructure buildout. Hyperscalers are large cloud operators that run computing platforms at enormous scale.

The September view extends that reasoning but places greater emphasis on market structure. Earlier declines reduced valuations and cleared speculative positions, while earnings provided a reason to reconsider selected stocks.

That is more restrained than treating every decline as an automatic buying opportunity. It also shifts attention from enthusiasm about AI capabilities toward evidence of durable demand.

JPMorgan’s case is therefore conditional. The correction created room for renewed interest, but earnings and commercialization must keep supporting the underlying expectations.

Why Semiconductors Sit at the Center of the Recovery

Semiconductors remain JPMorgan’s preferred expression of the AI theme because supply constraints and pricing give chipmakers clearer economic leverage.

The bank retained a bullish semiconductor outlook in its September note. Its analysts cited healthy fundamentals, pricing growth into 2027, and tight supply conditions expected to persist through 2028.

AI computing depends on more than graphics processors. Advanced systems also require high-bandwidth memory, conventional memory, networking chips, custom accelerators, packaging capacity, and manufacturing equipment.

Constraints at any part of that chain can strengthen supplier pricing. They can also delay deployments, making the availability of components almost as important as their performance.

This supply structure separates semiconductor companies from many software vendors. A limited number of businesses can manufacture advanced chips or provide the specialized equipment needed to produce them.

Software faces fewer physical constraints. New AI models can reduce development costs while allowing incumbents and startups to release competing features quickly.

JPMorgan’s preference therefore reflects more than recent price performance. The bank sees stronger visibility in the infrastructure orders needed to train and operate increasingly demanding systems.

The performance gap has already been substantial. Reuters reported that the MSCI World Semiconductors and Semiconductor Equipment Index had risen about 48% during 2026 by September 28.

The MSCI World Software and Services Index had gained only 1.3% over the same period. Those figures show how decisively investors had separated hardware beneficiaries from software companies.

They also explain why the semiconductor trade became vulnerable to a correction. Strong gains attracted momentum investors, increased concentration, and raised expectations for future growth.

A July analysis attributed to JPMorgan described a more dramatic portion of that adjustment. Korea’s KOSPI had fallen 25% from a recent peak, while the Philadelphia Semiconductor Index had dropped 20%.

Samsung, Micron, and other memory-related companies had declined between 20% and 50% from their highs. Those moves reflected concerns about capital spending, memory demand, and the sustainability of pricing.

Yet the selloff did not immediately create new semiconductor capacity. Building a fabrication plant takes years, while advanced packaging and memory production require specialized equipment and experienced operators.

That lag supports the JPMorgan semiconductor outlook. If AI demand remains strong, supply cannot respond as quickly as a software company can release another application.

Investors still need to distinguish scarcity from permanent competitive advantage. High prices encourage capacity expansion, product substitution, and more efficient model designs.

Efficiency presents a particular complication. AI developers are improving inference, which is the process of running a trained model to produce an answer.

More efficient inference can reduce the computing required for each task. However, cheaper usage can also increase total demand by making AI practical across more products and workflows.

The net effect depends on adoption. Falling computing requirements per request would hurt suppliers if overall usage stopped growing, but not if cheaper access produced far more requests.

J.P. Morgan Asset Management described this demand challenge in a February AI evaluation. It said major cloud companies averaged 35% year-over-year growth across key AI-related segments during the fourth quarter of 2025.

The firm presented that growth as evidence that AI was generating revenue. It also cautioned that acceptable investment returns require sustained adoption.

That balance is crucial. Semiconductor demand can remain strong while individual stocks disappoint because their valuations already assume years of exceptional growth.

JPMorgan’s preference is therefore relative, not unconditional. Chips offer clearer demand and supply signals than software, but they remain exposed to spending cuts and inflated expectations.

The Real Contest Is Semiconductor Scarcity Versus Software Competition

The central divide is not AI winners against AI losers, but scarce infrastructure against software whose competitive defenses are becoming harder to measure.

JPMorgan remains cautious on software because AI growth clouds the sector’s long-term outlook. The bank sees rising competition and uncertainty around which vendors will retain pricing power.

Its analysts stopped short of recommending an outright bearish position against the entire sector. They noted that software valuations had already fallen dramatically, making indiscriminate short positions risky.

Instead, the bank recommended revisiting a relative trade favoring semiconductors over software. A pair trade combines a preferred position with an opposing position to isolate the expected performance difference.

That recommendation clarifies the main opponent in this story. JPMorgan is not simply betting on a broad technology recovery. It is betting that hardware economics will remain more attractive than software economics.

The distinction begins with capital intensity. Training large models and serving millions of users require chips, memory, networking, facilities, and electricity.

Those expenses benefit infrastructure providers before an AI application proves it has durable demand. Software companies must then recover their costs through subscriptions, usage charges, advertising, or operational savings.

This creates an uneven timeline. Semiconductor suppliers can earn revenue during the buildout, while application vendors may need longer to demonstrate attractive margins.

Software companies also face AI cannibalization. A new assistant or automated workflow can reduce demand for an older product without producing equivalent revenue for its provider.

Established vendors can bundle AI features into existing contracts, but bundling complicates measurement. Higher usage does not necessarily mean customers are paying enough to cover computing and development costs.

Competition adds another layer. Foundation models increasingly offer overlapping capabilities, while open models allow businesses to deploy alternatives under their own control.

Application developers can switch model providers, combine several models, or use smaller systems for routine tasks. That flexibility can pressure margins throughout the software stack.

Hardware is not immune to competition. Cloud companies are developing custom accelerators, model designers are optimizing workloads, and manufacturers continue improving production capacity.

However, changing the supplier of an advanced chip involves technical, manufacturing, and software compatibility constraints. Those obstacles can preserve pricing longer than a feature advantage in software.

The contrast helps explain the 48% semiconductor index gain against the 1.3% software increase. Investors rewarded businesses serving the physical bottlenecks and questioned companies facing uncertain product substitution.

That gap can narrow in several ways. Semiconductor shares can fall, software earnings can improve, or both developments can happen together.

A successful software rebound would require clearer evidence that AI features produce incremental revenue or durable cost savings. Customer experimentation alone will not resolve the concern.

Infrastructure companies face a different test. They must show that large orders reflect end demand rather than customers building excess capacity.

The two sides are connected because hyperscalers sit between them. Companies such as Microsoft, Amazon, Alphabet, and Meta buy infrastructure while also selling cloud services and AI applications.

Strong capital spending helps chip suppliers immediately but raises the financial hurdle for cloud platforms. The more these companies invest, the more revenue and cash flow their AI services must eventually generate.

This is why the JPMorgan AI trade rebound does not restore the old technology narrative. Previous rallies often lifted infrastructure, platforms, applications, and speculative software together.

The new argument demands selection. Businesses with scarce products and confirmed orders receive preference, while companies with uncertain pricing face greater scrutiny.

Investors have already seen how quickly sentiment can change. An AI stock rebound in July lifted the S&P 500 within 1% of its record, even as concerns about spending returns persisted.

Such moves show that doubts and optimism can coexist. A rally can reflect bargain hunting without settling the long-term question about who captures AI profits.

JPMorgan’s relative preference offers a direct answer for now. The bank expects infrastructure scarcity to provide more support than software competition allows.

What the Bull Case Still Has to Prove

A cheaper market is not automatically an attractive one when the earnings required to justify AI spending remain uncertain.

The skeptical case starts with capital expenditure. Technology companies are committing substantial resources before the final demand for many AI services is fully visible.

Strong balance sheets make that investment possible. They do not guarantee adequate returns.

Data centers also require power, cooling, networking, land, and long construction schedules. Supply constraints can protect equipment makers while increasing costs for the companies financing deployment.

The central question is whether AI revenue grows quickly enough to absorb those costs. JPMorgan says evidence of monetization is increasing, but the evidence remains uneven across industries and products.

Cloud demand provides one positive signal. Enterprises are renting more computing capacity and experimenting with AI-assisted development, customer support, research, and document analysis.

However, experimental spending differs from permanent production demand. Companies can reduce projects that fail security reviews, produce unreliable results, or cost more than expected.

Usage metrics also need context. A service can process more tokens, users, or queries while generating limited profit because inference remains expensive.

Price reductions further complicate the picture. Lower model costs encourage adoption, but they can reduce revenue per task and intensify competition among providers.

That tension places hyperscaler spending at the center of the risk. If cloud companies reduce expansion plans, semiconductor orders could weaken before new application revenue fills the gap.

The July market decline illustrated how sensitive the trade is to this concern. Investors reacted sharply to questions about overbuilding and the timing of additional chip demand.

Earlier gains also created demanding comparisons. A semiconductor company can report growing revenue and still disappoint if investors expected faster growth or stronger guidance.

The broad market backdrop matters as well. Interest rates affect the present value investors assign to future earnings, especially for companies priced around distant growth.

Macroeconomic volatility can therefore overwhelm company-level progress. JPMorgan expects reduced volatility to support better trading, but that condition remains outside any chipmaker’s control.

Supply tightness carries its own reversal risk. High margins motivate manufacturers to add capacity, while customers seek alternative suppliers and more efficient designs.

If new capacity arrives as demand growth slows, pricing can weaken quickly. Semiconductor cycles have repeatedly moved from shortage to excess because production decisions occur long before final demand becomes clear.

Geopolitical exposure adds another uncertainty. Advanced chip production, memory, packaging, and manufacturing equipment span several regions with different trade and security policies.

Export controls can restrict markets or alter product designs. Trade disputes can also influence where companies build facilities and which suppliers customers select.

These factors do not invalidate the JPMorgan semiconductor outlook. They show why the thesis depends on several moving parts rather than one permanent shortage.

The strongest evidence would combine high utilization, firm pricing, growing orders, and expanding customer revenue. Any single indicator can mislead when viewed alone.

Investors should also separate AI adoption from stock performance. A technology can spread rapidly while shareholders receive poor returns because competition transfers much of the value to customers.

The internet created enormous economic value, but many infrastructure and application providers still failed. AI can follow a similarly uneven path without matching every historical detail.

JPMorgan’s note recognizes this selectivity. Its warning that technology may not recover its previous success prevents the thesis from becoming a simple return to speculative enthusiasm.

The cleaner positioning argument is also temporary. Once investors rebuild large positions, crowding can return and make the sector vulnerable again.

The JPMorgan AI trade rebound must therefore earn support through results. Lower valuations create an opening, while earnings determine whether that opening becomes a durable recovery.

Three Signals Will Decide What Comes Next

Earnings, capital spending, and the semiconductor-software performance gap will determine whether JPMorgan identified a reset or only a temporary pause.

The first signal is the next round of semiconductor earnings and guidance. Revenue alone will not provide enough evidence because investors already expect substantial AI demand.

Pricing, order visibility, inventory, and production utilization will matter more. Continued pricing strength into 2027 would reinforce JPMorgan’s view that supply conditions remain tight.

Management commentary about 2028 capacity will also deserve attention. Accelerated expansion could confirm demand, but it could also increase the risk of later oversupply.

Investors should compare guidance across processors, memory, networking, equipment, and advanced packaging. Weakness isolated to one category carries a different meaning than a broad order slowdown.

The second signal is capital expenditure from the largest cloud companies. Microsoft, Amazon, Alphabet, and Meta collectively influence demand across much of the AI infrastructure chain.

Rising expenditure would support semiconductor suppliers, but investors must examine why spending increased. Capacity tied to committed customers is stronger evidence than construction based mainly on projected demand.

Cloud revenue growth must also keep pace with investment. A widening gap between spending and AI-related revenue would weaken the broader monetization argument.

Investors should listen for changes in depreciation, utilization, and the expected life of AI equipment. Those accounting and operational details affect the return generated by each infrastructure dollar.

A spending reduction would not automatically end the AI buildout. It would, however, challenge expectations based on continuously accelerating demand.

The third signal is whether software begins closing its performance gap with semiconductors. That shift would test JPMorgan’s preferred relative trade more directly than a broad market rally.

A credible software recovery requires more than lower valuations. Vendors need to show that AI features improve retention, increase contract values, or reduce delivery costs.

The quality of revenue matters. Temporary consulting projects and promotional trials offer less support than repeatable subscriptions or usage-based demand from production systems.

Improving software economics would broaden the AI trade and reduce its dependence on physical infrastructure. It could also suggest that customers are moving from experimentation to commercial deployment.

Continued software weakness would support JPMorgan’s caution. It would indicate that infrastructure spending still captures more value than the applications meant to justify it.

These three signals should be read together. Strong chip results without improving cloud monetization can extend the rally while increasing the eventual financial hurdle.

Higher cloud spending without firm semiconductor pricing might show that supply is becoming easier to obtain. Better software earnings could validate adoption even as hardware growth normalizes.

The most constructive outcome would combine durable chip demand, disciplined infrastructure investment, and measurable application revenue. That combination would distribute economic value across the AI chain.

The most concerning outcome would pair slowing cloud growth with rising capacity and weak software pricing. It would suggest that expenditure ran ahead of profitable demand.

Between those outcomes lies a wide range of selective opportunities and disappointments. That is exactly the market JPMorgan’s note describes.

Readers tracking the JPMorgan AI trade rebound should resist treating daily price gains as confirmation. The stronger test comes from earnings, customer commitments, and pricing across the supply chain.

Watch the next semiconductor reports first, then compare hyperscaler spending with AI revenue. Finally, test whether software companies can turn adoption into durable margins.

If all three signals improve, the rebound will look increasingly fundamental. If only share prices recover, the market may simply be rebuilding the crowded positions it recently cleared.