Global equities spent the week projecting confidence while displaying plenty of anxiety beneath the surface. The S&P 500 remained close to record territory, rising roughly 0.75% over the past month and more than 16% year-over-year, even as technology indices weakened and semiconductor stocks faced another round of profit-taking.

The calm at the index level concealed a more active rotation underneath. Capital moved out of some AI-exposed chipmakers, briefly found shelter in defensive sectors, then edged back towards large technology companies ahead of a crowded week of earnings and central-bank decisions.

The strongest part of the market was the Magnificent Seven, excluding Nvidia. Meta, Apple, Microsoft, Amazon, Alphabet, and Tesla proved relatively resilient while semiconductor shares declined. Investors appeared willing to distinguish between the companies building AI infrastructure and the platforms attempting to monetize it.

Meta was the clearest example. Its shares rose approximately 8% to 9% in a single session following reports that the company could turn excess AI computing capacity into a commercial cloud service. The market interpreted the proposal as a potential new revenue stream. It also raised a more uncomfortable question: if Meta has spare capacity to sell, could hyperscaler capital expenditure be approaching a peak?

Memory and semiconductor equipment stocks occupied the other end of the market. Micron, SanDisk, AMD, and Intel recorded high-single-digit or double-digit declines as profit-taking collided with concerns about supply-demand efficiency and elevated AI valuations.

Equipment and testing companies also came under pressure. Teradyne, Lam Research, KLA, and other fabrication-tool suppliers were among the weakest performers in the S&P 500 over recent weeks. Several recorded one-day declines of between 10% and 14%, leaving them near the bottom of the index leaderboard.

Meanwhile, less visible parts of the data-center supply chain continued to perform.

Astera Labs reported record quarterly revenue and 115% year-over-year growth in 2025. Its guidance pointed to further expansion as its connectivity products become increasingly important to rack-scale AI systems.

Comfort Systems USA also benefited from the infrastructure theme. The company provides heating, cooling, and mechanical systems for data centers and industrial facilities. UBS raised its price target while maintaining a Buy rating, reflecting the growing recognition that cooling high-density computing equipment is becoming a significant commercial market rather than a minor construction expense.

The market continued to separate AI accelerator designers and data-center beneficiaries from memory and semiconductor-equipment businesses.

Nvidia, TSMC, Broadcom, and the major cloud platforms generally remained resilient. Memory producers and equipment manufacturers faced concerns about oversupply, delayed capital expenditure, and an AI build-out concentrated among a small number of advanced fabrication facilities.

Recent analysis suggests the year-to-date performance gap between AI-focused chip designers and equipment manufacturers has reached approximately 35 percentage points.

Economic growth remained respectable, and inflation continued moving closer to target. That combination kept the Federal Reserve on course to hold its policy rate at 3.50% to 3.75% during the July 28 to 29 meeting.

The market is nevertheless paying increasing attention to the possibility of another rate increase later in the year. Every inflation release, employment report, and jobless-claims figure now carries additional significance because higher interest rates place greater pressure on elevated technology valuations.

AI companies do not need earnings growth alone. They also need the discount rate to remain contained.

Investors are no longer rewarding technology companies simply for mentioning artificial intelligence.

Second-quarter results from semiconductor and software businesses received a more cautious response than in previous reporting periods. The upcoming results from Microsoft, Meta, Apple, and Amazon will be judged primarily on three factors: AI capital expenditure, the impact on margins, and the pace of revenue monetization.

The market is effectively asking whether AI remains an accelerating profit opportunity or is becoming an increasingly expensive operating requirement.

Reports that Meta could sell excess AI computing capacity contributed to a sharp sell-off in memory-chip companies across Asia and the United States.

The market appeared to extrapolate a limited commercial proposal into a broader conclusion that hyperscaler demand had permanently peaked. That interpretation may have moved faster than the underlying evidence.

Analysts in Asia and the US argued that the reaction was excessive. The wider AI investment cycle remains active, while falling computing costs could stimulate new use cases and increase overall demand. The market’s initial response nevertheless treated spare capacity as evidence of structural oversupply.

A Google research paper suggesting that future AI data centers could operate with less memory than previously expected contributed to the loss of approximately $100 billion in market value among US memory companies.

Micron accounted for an estimated $70 billion of the decline, while SanDisk, Western Digital, and Seagate also recorded substantial losses.

The research reportedly concluded that the long-term effect on memory demand could be neutral. Investors focused instead on the prospect of improved efficiency, which immediately translated into lower profitability.

Markets have a recurring tendency to interpret optimization as obsolescence. The two are not the same.

The July Federal Reserve meeting is widely expected to result in no change to interest rates. Even so, a Reuters poll showing that many economists see a meaningful probability of a later rate increase revived concerns that rates will remain high for longer.

The response was notable because equities have already advanced for months under restrictive monetary conditions. Yet each new inflation figure is treated as though the market has only just discovered that the Federal Reserve remains uncomfortable with inflation above target.

The policy rate may stay unchanged. The language surrounding that decision could matter more.

The semiconductor industry can currently be divided into two broad groups: companies supplying the central processing power behind AI, and companies supplying the memory, testing equipment, and fabrication tools needed to support it.

The performance gap is becoming difficult to ignore.

Key takeaway: The market is assigning a premium to scarce AI computing capacity while discounting businesses perceived as cyclical, replaceable, or exposed to excess supply.

This is what a narrow investment boom looks like. Profits accumulate around a small number of architectures, advanced manufacturing nodes, and dominant platforms. Much of the wider supply chain is left waiting for demand to broaden.

Last week’s movements revealed more about positioning and expectations than about a sudden change in the long-term AI investment cycle.

The structural themes remain visible. Artificial intelligence is becoming part of corporate infrastructure. Hyperscalers are industrializing their data centers. Connectivity, power, and cooling are becoming increasingly valuable components of the computing stack.

What changes from week to week is the price investors are willing to pay for each part of that story.

Exposure to the AI theme now extends beyond semiconductor designers. It includes foundries, networking technology, cloud platforms, electrical systems, and mechanical infrastructure. This wider view reduces dependence on a single company or product cycle.

Frequent portfolio checking does not reduce risk. It mostly increases exposure to market noise.

The Financial Times examines how a single Google research paper disrupted the assumption of a permanent shortage in memory chips. The resulting sell-off removed tens of billions of dollars from the sector’s market value within days.

Read it on the Financial Times

Reuters outlines why the Federal Reserve is expected to leave its target rate unchanged at the July 28-29 meeting, despite inflation remaining above target.

The article also explains why economists continue to assign a meaningful probability to another rate increase later in the year. That possibility is particularly relevant for high-growth companies whose share prices remain sensitive to changes in discount rates.

Read it on Reuters

This calendar brings together the week’s major US economic events, including the Federal Reserve decision, second-quarter GDP, and the Personal Consumption Expenditures inflation report.

All three arrive within a concentrated 72-hour period, creating an unusually dense cluster of economic and market information.

Read it on ThriveInMarkets

The coming week combines four major market events: the Federal Reserve meeting, US GDP data, PCE inflation data, and large-cap technology earnings.

Index volatility has remained relatively contained given the scale of the calendar. Positioning beneath the surface looks less stable. Investors remain heavily exposed to AI beneficiaries, with relatively lighter allocations to traditional cyclicals and defensive sectors.

That concentration leaves the market sensitive to any disappointment from the major cloud platforms.

The central question is straightforward: can AI-related capital expenditure and revenue growth remain strong enough to support current valuations while the Federal Reserve maintains restrictive monetary policy?

The answer will affect more than the hyperscalers. It will influence semiconductor designers, memory producers, networking companies, equipment suppliers, and the wider data-center ecosystem.

The US will publish June Durable Goods Orders, Core Durable Goods Orders, and Non-Defence Capital Goods Orders excluding aircraft.

These releases provide insight into whether US companies are continuing to invest in equipment and productive capacity or beginning to restrict spending.

The Federal Reserve’s two-day meeting concludes with the interest-rate decision and press conference on July 29.

The consensus expectation is for the target range to remain at 3.50% to 3.75%. The more important information will come from the accompanying language on inflation, economic growth, and the possibility of future rate increases.

The advance estimate of second-quarter US GDP will be released alongside the GDP price index, core PCE, headline PCE, and employment-cost data.

The concentration of releases makes this the most likely point for a new macroeconomic narrative to emerge.

Stronger-than-expected inflation could increase pressure on growth-stock valuations. Weaker growth could raise questions about corporate demand and the durability of capital expenditure.

Microsoft, Meta Platforms, Apple, and Amazon are all expected to report during the week.

Together, these companies represent a significant share of the Nasdaq and much of the market’s AI investment thesis. Their results will also affect sentiment toward Alphabet, Nvidia, Broadcom, TSMC, and the broader semiconductor industry.

For the AI trade, these earnings are not simply company updates. They are a test of whether current spending is producing measurable commercial returns.

Strong results from Microsoft, Meta, Apple, and Amazon could reinforce the argument that AI is becoming a core platform rather than a peripheral experiment.

For Microsoft, attention will center on Azure AI demand and Copilot monetization.

For Meta, the key issues will be AI-driven engagement, advertising efficiency, and the scale of infrastructure spending.

Apple will face questions about whether its AI features can stimulate hardware upgrades or increase services revenue.

Amazon’s results will be assessed based on AWS growth, adoption of AI services, and the cost of expanding its data-center footprint.

Positive results would extend the period over which the market is prepared to tolerate elevated valuations.

Further revenue growth from Astera Labs and constructive data-center commentary from Comfort Systems could strengthen the case for owning the enabling infrastructure behind AI.

Connectivity, networking, cooling, and power systems generate revenue regardless of which consumer-facing AI application gains the most users.

Astera Labs, Comfort Systems, and parts of Broadcom’s business sit directly within this market layer.

A slowdown in expected AI capital expenditure from any major hyperscaler could affect the entire ecosystem.

The consequences would extend from cloud platforms to semiconductor designers, connectivity suppliers, equipment manufacturers, and memory companies.

Micron, SanDisk, Lam Research, and Teradyne are already trading under the pressure of weaker sentiment. Negative hyperscaler guidance could deepen those concerns.

Stronger-than-expected inflation could revive concerns about higher interest rates. Weaker-than-expected growth could undermine confidence in corporate spending.

Both outcomes would be difficult for highly valued software and semiconductor companies. These businesses depend on strong demand and a discount rate that does not rise materially.

Weak growth would also raise questions about whether current AI capital-expenditure plans can continue at the same pace.

Asian markets have already shown how quickly concerns about semiconductor demand can spread.

South Korea’s KOSPI reportedly fell almost 8% in a single session amid chip-led selling and foreign capital outflows. US semiconductor stocks remain exposed to the same narrative, particularly as the market focuses on memory supply, capital expenditure, and unofficial distribution channels for restricted AI chips.

A company-specific concern can quickly become a sector-wide event when positioning is crowded.

Microsoft, Amazon, Meta, Alphabet, and Apple remain at the center of the market.

The coming earnings reports will help determine whether they retain their status as AI growth platforms or begin to be viewed as mature technology utilities carrying unusually high investment costs.

The difference matters because utilities are normally valued for dependable cash flow, not indefinite multiple expansion.

AI accelerator designers and advanced foundry companies remain in the high-growth, high-expectation segment of the market.

Nvidia, TSMC, and Broadcom continue to benefit from demand for advanced computing and networking. AMD also remains connected to the prospect of gaining market share in AI accelerators.

Memory and equipment companies are more cyclical. Micron, SanDisk, Lam Research, Teradyne, and Intel are facing questions about capacity, oversupply, and the breadth of semiconductor demand beyond the most advanced AI systems.

The rankings contain companies on both sides of this divide.

Astera Labs and Comfort Systems represent the less visible physical layer of AI development.

Astera provides connectivity technology for rack-scale systems. Comfort Systems supplies the cooling and mechanical infrastructure required by high-density computing facilities.

These companies offer exposure to the expansion of AI infrastructure without depending entirely on which model, application, or chip architecture ultimately dominates.

AT&T and Philip Morris occupy the lower-growth, income-focused end of the rankings.

Their earnings are less directly connected to AI capital expenditure, semiconductor cycles, or hyperscaler guidance. This can make their share-price behavior less extreme during periods of technology-led volatility.

They are unlikely to provide the same upside as high-growth AI companies. Their role is primarily to introduce a different earnings and cash-flow profile.

Institutional positioning has behaved like a pendulum over recent weeks.

Early in July, capital flowed into healthcare, utilities, and consumer staples, while technology and semiconductor stocks underperformed.

Leadership then shifted back towards technology. Energy and technology recorded some of the strongest gains, supported by semiconductor and communication-services companies, while defensive sectors weakened.

More recently, chip and memory stocks have experienced renewed selling. Investors have questioned elevated AI valuations and reduced exposure to businesses perceived as cyclical or commodity-like.

At the same time, demand for AI infrastructure companies such as Astera Labs and data-center enablers such as Comfort Systems has remained relatively resilient. This suggests that institutional capital is increasingly distinguishing between the visible applications of AI and the infrastructure required to operate them.

The AI stack is not a single trade. It is a spectrum that includes hyperscalers, semiconductor designers, foundries, connectivity, memory, equipment, cooling, and power.

Each layer carries a different combination of growth, cyclicality, valuation, and risk.

The previous chart highlighted the divide between AI semiconductor designers and equipment manufacturers. This week’s chart widens the view to include the major layers of the AI ecosystem.

Research into semiconductor equipment weakness, Astera Labs’ earnings, and Comfort Systems’ data-center exposure all point to the same conclusion.

Most AI-focused portfolios are concentrated near the top of the table. Yet the lower infrastructure layers are responsible for the physical work required to make AI systems function.

The chips receive the headlines. The network, cooling and electrical systems keep them operating.

Below is a structured qualitative assessment of the nineteen companies in the rankings. It combines valuation, recent momentum, risk, and the most important near-term catalyst based on the supplied market context.

The rankings form a clear barbell.

At one end are the dominant AI platforms and infrastructure leaders: Alphabet, Microsoft, Meta, Apple, Amazon, and Nvidia. Their valuations reflect strong expectations for growth, monetization, and sustained capital expenditure.

At the other end are memory and equipment companies already priced for more difficult conditions: Micron, SanDisk, Teradyne, Lam Research, Intel, and AMD. These businesses carry higher cyclical and execution risk, but their valuations increasingly reflect negative expectations.

Between the two groups sit the enabling businesses: TSMC, Broadcom, Astera Labs, and Comfort Systems. They provide the manufacturing, connectivity, and physical infrastructure required by the wider AI ecosystem.

AT&T and Philip Morris occupy a different category. Their cash flows have limited direct exposure to AI spending and provide a counterweight to the higher-growth, higher-volatility companies elsewhere in the rankings.

The central distinction is no longer between AI and non-AI companies. It is between businesses with scarce capabilities, businesses exposed to cyclical supply, and businesses responsible for the infrastructure that connects the two.