From AI Beta to Profit Realization: How to Find the Next Profitable Moves in Q3 U.S. Stocks?

Let's be straightforward: the U.S. stock market still has support in Q3, but if you want to continue making money, your approach may need to change.

In the past Q2, the market's pricing of geopolitical shocks became temporarily desensitized. Combined with a recovery in AI infrastructure stocks and a rebound in risk appetite, U.S. stocks regained strength. Driven particularly by large-cap tech stocks and core AI assets, the market briefly returned to a familiar trading pattern: as long as capital expenditure continues to increase, as long as computing power demand remains strong, valuations can continue to rise.

But entering Q3, this logic is facing a higher verification threshold.

On one hand, inflation remains above the Fed's target, with long-end rates and the policy path continuing to limit the expansion space for high-valuation assets. On the other hand, AI-related company stock prices have already priced in quite optimistic growth expectations. What the market needs to see next is no longer just larger CapEx numbers, but orders, deliveries, gross margins, cash flow, and return on capital.

Therefore, MSX Research maintains a cautiously positive outlook on U.S. stocks for Q3.

The indices are not yet entering a systemic bear cycle, but the source of returns is shifting from "valuation expansion" to "earnings realization." AI remains the most important industry theme, but the trading focus will move from broad AI Beta further down to segments more easily verifiable by earnings reports—storage, networking & optical interconnects, power, cooling, data center delivery, and edge computing and Physical AI centered around real-world applications.

To summarize the Q3 market environment in one sentence: inflation caps the valuation ceiling, earnings determine the index floor, AI realization decides structural Alpha, and market breadth determines the quality of the rally.

1. After the Ebb of Valuation Expansion, Earnings Must Support the Indices

From Q2 to Q3, the dominant contradiction in the market has clearly shifted.

The trading chain in Q2 was relatively clear: geopolitical conflicts impacted oil prices and inflation expectations, leading to adjustments in the interest rate path. After risk appetite recovered, funds flowed back into AI and large-cap tech stocks. The core of the market's trading was valuation repair following the marginal easing of macro pressures.

By Q3, the contradiction has transmitted further down the chain. Specifically, inflation constrains valuations, the Fed reduces forward guidance, earnings need to support the indices, and AI must move from capital expenditure to tangible realization.

This does not mean the market is about to turn bearish. A more accurate description is that the barrier to returns is rising.

1. Inflation Remains the Ceiling for High-Valuation Assets

The first layer of constraint in Q3 still comes from inflation and the Federal Reserve.

U.S. inflation levels remain significantly above the 2% long-term policy target, meaning the foundation for "rapid rate cuts to support valuations" is not solid. Simultaneously, the communication style of the Fed under Warsh places greater emphasis on real-time data, price stability, and policy discipline, weakening the market's long-standing reliance on forward guidance.

This will have three main impacts:

• The familiar 'Fed put' is thinning: Investors can no longer simply assume that policy will quickly release accommodative signals whenever market volatility increases;
• The market will become much more sensitive to individual data points: CPI, PCE, wages, employment, oil prices, consumer data, and even corporate earnings can all trigger a repricing of the interest rate path and valuations;
• The margin for error for high-growth stocks will significantly decrease: The AI industry trend is still intact, but 'being in the right direction' is no longer enough to sustain stock price expansion. The market needs more proof from orders, revenue, profit margins, and cash flow to confirm that current valuations are not built purely on distant imagination;

Therefore, the macro backdrop for Q3 is not a typical recession trade, but rather a high-valuation market that still has growth support but is persistently constrained by interest rates.

In this environment, capital will favor two types of assets: one is companies with strong earnings certainty, high-quality cash flow, and robust balance sheets; the other is sectors with low duration, resource attributes, or inflation-hedging capabilities, including gold, resources, power, and certain high-cash-flow financial assets.

2. Indices Can Still Rise, But Not Solely on Higher P/E Ratios

The most important support for U.S. stocks in Q3 still comes from corporate earnings.

Several Wall Street institutions have continued to raise their year-end targets for U.S. stocks. The core basis for these upgrades is not that valuations can expand indefinitely, but that there is still further room for upward revisions to corporate EPS.

This distinction is crucial.

When market valuations are already at historically high levels, the key to whether indices can continue to rise is no longer whether investors are willing to pay higher multiples, but whether corporate earnings can continue to grow beyond expectations. Goldman Sachs has raised its S&P 500 year-end 2026 target to 8,000 points and its 2026 and 2027 EPS forecasts to $340 and $385, respectively.

At the same time, it expects the forward P/E ratio for U.S. stocks to remain roughly around 21x – a level already within the high range of the past 40 years.

In other words, the subsequent rise in indices depends more on EPS than on further valuation multiple expansion. If the earnings season drives sustained upward EPS revisions, U.S. stocks have a foundation for volatile upward movement. However, if earnings revisions begin to slow while inflation or long-end rates pick up again, the market could quickly switch from 'earnings-driven' to 'valuation compression.'

So, the most critical question for Q3 is not whether the indices can still rise, but whether, under current valuations, earnings can continue to support them.

This also means that the allocation strategy should not remain passively chasing indices but should shift more towards sectors that can be validated by orders and earnings reports. This includes AI infrastructure, storage, power, data center infrastructure, industrials, finance, platform advertising, and consumer staples with stable cash flows.

3. Market Breadth Will Determine if the Rally is Healthy

Besides index levels, market breadth is another key factor to watch in Q3.

If U.S. stocks continue to rise, but the gains are heavily dependent on a few AI giants, market concentration will increase further. Any single earnings miss could trigger more significant volatility.

A healthier market structure would be: AI maintains its position as the main theme, while industrial, financial, platform advertising, and some consumer sectors begin to take the baton.

In other words, Q3 shouldn't just be about whether Nvidia, the semiconductor index, or the Nasdaq hits new highs. We also need to observe equal-weight indices, the number of advancing stocks, and whether the earnings expectations of non-AI sectors are improving simultaneously.

AI determines the market's height, but market breadth determines how far this rally can go.

2. AI CapEx 2.0: From Computing Scarcity to Delivery Realization

AI remains the most critical industry theme in Q3, but the trading logic has shifted from 'expectations' to 'verification.'

In Q2, the market primarily traded on computing power scarcity, upward capital expenditure revisions, and supply chain expansion. As long as tech giants continued to increase CapEx and GPU supply remained tight, the entire supply chain could be revalued based on higher demand.

But in Q3, the market will more directly question several things:

• Can financing truly be transformed into GPUs and data centers?
• Can GPUs be turned into computing power that is delivered on schedule?
• Can the computing power generate stable, long-term revenue?
• Can revenue cover depreciation, financing costs, and equity dilution?
• Can it ultimately generate positive free cash flow and reasonable ROIC?

This is the essence of so-called AI CapEx 2.0. It's no longer about betting on a single chip type or simply chasing a single optical module. Instead, it involves looking along the entire data center construction chain to find segments that can truly realize orders and profits. Examples include chips & platforms → networking & optical interconnects → storage → power & cooling → server & system delivery → computing operations → edge & real-world applications.

1. Chips Are Still the Entry Point, But No Longer the Only Answer

Among these, chips remain the most important entry point for the AI industry.

NVDA remains the pricing anchor for global AI assets. AVGO corresponds to custom ASICs and networking platforms. MRVL benefits from both custom chips and optical interconnects. TSM corresponds to advanced process nodes, advanced packaging, and the entire AI semiconductor manufacturing ecosystem.

However, the judgment on the semiconductor layer in Q3 will be more stringent than before.

The market won't just care about chip performance; it will continue to question whether orders can persistently beat expectations, whether advanced packaging and capacity bottlenecks can ease, whether the customer base is healthy enough, whether gross margins can remain high, and whether inference, AI PCs, enterprise AI, and Edge AI can form new growth curves.

INTC needs to be understood within a different framework. It is not a direct substitute for NVDA. It is more like a comprehensive option on U.S. semiconductor security, server CPUs, AI PCs, Edge AI, and wafer foundry services. Therefore, its thesis depends on whether low-valuation assets can benefit from a convergence of policy, industry, and fundamental catalyst.

2. The Larger the Cluster, the More Important Networking & Optical Interconnects

The larger the GPU cluster, the more critical interconnects become.

While the market fully priced in optical modules, switches, and high-speed interconnects in Q2, the focus in Q3 will shift from pure industry sentiment to a more detailed assessment of realization quality. This includes whether demand for 800G and 1.6T continues to be revised upward, whether order visibility is sufficiently high, whether customer concentration is manageable, whether capacity expansion and yields can keep pace with demand, and whether silicon photonics, upstream materials, and specialty processes become new bottlenecks.

This layer also represents one of the easiest directions for capital to flow from core AI leaders to secondary assets.

When order visibility improves, companies in optical communications, silicon photonics, and specialty materials often possess both earnings elasticity and valuation repair potential. Compared to companies relying solely on grand narratives, these firms undoubtedly find it easier to complete the verification process through orders, capacity utilization rates, and financial guidance.

ANET.M, CRDO.M, LITE.M, COHR.M, AAOI.M, FN.M, AXTI.M, and TSEM.M are important observation assets in this direction.

GLW.M is also worth including. It is not a pure-play optical module stock, but its fiber optics, glass, and data center materials business allows it to benefit from increased data center connection density and infrastructure spending.

3. Storage is Evolving from an AI Sideline to a Core Bottleneck

Storage remains an area where weightings should be increased in Q3.

In the past, when the market talked about AI, it first thought of GPUs and networks. But as model parameters, inference calls, and data volumes continue to grow, AI's consumption of HBM, DRAM, NAND, enterprise SSDs, and HDDs is also steadily increasing.

Storage is no longer a sideline to the AI industry; it is becoming an increasingly unavoidable core component in data center construction.

Micron's recent earnings and guidance have reinforced the judgment that "AI storage is entering a realization phase." The company's Q3 FY2026 revenue reached $41.456 billion, with Non-GAAP gross margin rising to 84.9% and adjusted free cash flow around $18.3 billion. For the fourth fiscal quarter, it provided revenue guidance of approximately $50 billion, plus or minus $1 billion, with gross margin guidance around 86%.

These numbers show that AI's pull on storage is no longer just about order expectations; it is beginning to manifest as the simultaneous realization of revenue, profitability, and cash flow.

However, storage trading in Q3 should no longer be simply understood as a "single-stock MU trade." A more reasonable structure is to break down storage into three tiers:

• Tier 1 is the NAND, SSD & HDD diffusion, including WDC.M, STX.M, and SNDK.M. They benefit from AI data growth, improved enterprise storage demand, and the recovery of the traditional storage cycle, with relatively limited direct competition with HBM leaders;
• Tier 2 is MU.M. Micron remains one of the most core storage assets in the U.S. stock market, benefiting from improvements in HBM, DRAM, and NAND sentiment. However, with the advancement of the SK hynix ADR plan, MU's premium as a "scarce HBM proxy in U.S. stocks" may be partially diluted (for further reading: Watching Hynix by Day, Trading U.S. Stocks by Night: Is a New "Asia Session Bellwether" for Global AI Markets Emerging?);
• Tier 3 includes SIMO.M and other controller and second-order elasticity assets. They benefit from enterprise SSD, AI PC, and Edge AI storage proliferation, but their certainty and priority are currently lower than those of memory manufacturers and the HDD/NAND main theme;

For the entire storage sector, the SK hynix ADR is a classic double-edged sword.

On the positive side, it will enhance the public market pricing of the global HBM leader and increase investor attention on the entire storage industry. On the negative side, once U.S. stock investors have a more direct investment channel for an HBM leader, MU's original premium for being a scarce proxy could be partially undermined.

Therefore, the storage logic for Q3 will gradually shift from "single-point scarcity" to "full supply chain diffusion."

4. Data Center Infrastructure Must Be Grouped Separately

The bottleneck for AI is expanding from "having GPUs or not" to include "having power or not, having a facility or not, having cooling or not, and being able to connect to the grid or not."

This layer should no longer be simply categorized under industrials or utilities. As AI capital expenditure progressively enters a phase of real construction, power, thermal management, electrical equipment, construction delivery, and high-reliability components have become part of the AI CapEx trade.

Data center infrastructure can be broken down into at least five layers:

• Power and thermal management: VRT.M;
• Electrical equipment and power distribution: ETN.M;
• Grid engineering and interconnection construction: PWR.M;
• Power generation and grid equipment: GEV.M;
• System delivery, PCB, connectors, and materials: DELL.M, SMCI.M, TTMI.M, APH.M, GLW.M;

The biggest advantage of this direction is that the more AI capital expenditure pushes towards real construction, the harder it is to bypass data center infrastructure.

After all, compared to assets reliant solely on valuations and narratives, data center infrastructure companies often have clearer backlogs, order cycles, and delivery schedules, making it easier for them to validate industry trends through revenue and cash flow.

5. From Standalone Hardware to AI Factories

As the market moves from purchasing a single type of hardware to building complete AI systems, the importance of AI factories, server delivery, high-end PCBs, and enterprise AI infrastructure will also rise.

Judgment criteria for this layer include whether orders are sustainable, whether products can be delivered on time, whether gross margins are stable, whether the customer base is diversifying from single large clients to more enterprises, and whether enterprise AI deployments can generate scalable revenue.

DELL.M and SMCI.M both fall under the system delivery direction, but their natures are not identical. Comparatively, DELL's business structure leans more towards enterprise AI, servers, and complete system delivery, providing a relatively clearer revenue verification path. SMCI offers higher earnings elasticity but also carries more pronounced risks related to volatility, governance, and expectation gaps.

Other directions worth watching include PENG.M and HPE.M.

6. Computing Operators Have the Highest Elasticity, But Also the Highest Verification Hurdle

Computing operators represent the layer with the highest elasticity within the AI main theme, but also the highest risk.

These companies have the most intuitive growth story: secure financing, procure GPUs, build data centers, and then generate revenue through long-term computing contracts.

However, the capital market ultimately needs to verify whether this business model works. This includes whether GPUs actually arrive, whether power and facilities are delivered on schedule, whether the quality of long-term customer contracts is sufficiently high, whether computing utilization can continually improve, whether depreciation, debt, and financing costs will eat