Meta’s AI Spending Shock Tests the Market After the Fed Hold—and Microsoft Complicates the Bear Case

Wall Street entered July 29, 2026 expecting the Federal Reserve to be the day’s dominant event. It was not. The central bank left its target range unchanged at 3.50% to 3.75%, but the more revealing test arrived after the closing bell, when Meta Platforms and Microsoft delivered sharply different answers to the same question: can the enormous artificial-intelligence buildout begin producing enough visible financial return to justify its cost?

Meta’s answer unsettled investors. The company reported strong 28% revenue growth, but operating income fell, net income declined, and free cash flow dropped 91% from a year earlier to just $784 million. Meta also raised the lower end of its 2026 capital-expenditure forecast to $130 billion, leaving a range of $130 billion to $145 billion. Its shares fell about 5% in extended trading shortly after the release. Microsoft, by contrast, reported $90.0 billion in quarterly revenue, 43% Azure growth, a $678 billion commercial backlog, and more than 30 million paid Microsoft 365 Copilot seats. Its shares rose roughly 3% after hours.

The split matters because it turns a broad argument about an “AI bubble” into a more demanding company-by-company analysis. It is possible for the market to be right that AI infrastructure spending is excessive in parts of the ecosystem while also being right that Microsoft is converting the same investment cycle into faster cloud growth, contracted backlog, and software revenue. It is also possible for Meta’s core advertising business to benefit from AI while the cash cost of building future capacity overwhelms near-term free cash flow. The market is no longer asking whether AI is useful. It is asking who is earning a return, how quickly, and with what strain on the balance sheet and cash-flow statement.

That was the central risk identified hours earlier by technical analyst Tim Knight in a tastylive discussion. Knight argued that Meta’s earnings mattered more for the immediate AI trade than the Fed decision and more, in his view, than Microsoft’s report. A hard decline in Meta, he said, could throw “ice water” on AI-linked stocks. He also described bearish positions in technology, semiconductors, and the broader market, including long-dated index puts intended to reduce the psychological pressure of short-dated trading.

The first part of that thesis received support. Meta did fall after hours, semiconductor shares had already suffered a severe regular-session decline, and the Nasdaq 100 ended the day 11% below its June record. Yet the second part—the idea that Microsoft barely mattered—was harder to defend once the results arrived. Microsoft’s cloud performance provided precisely the evidence investors have been demanding: demand growth, paid adoption, backlog expansion, and a clearer commercial path from infrastructure investment to revenue.

Last updated: July 29, 2026, 5:10 p.m. Eastern Time. Microsoft’s earnings call was scheduled for 5:30 p.m. ET, after this research cutoff. This analysis therefore relies on the company’s published earnings release and information available by the cutoff.

The immediate answer: Meta mattered, but not in isolation

The dominant search question after the close was straightforward: did Meta’s earnings confirm that the AI trade is breaking down? The most accurate answer is narrower. Meta confirmed that investors are becoming less willing to overlook collapsing free cash flow merely because revenue is growing and management remains optimistic about long-term AI opportunities. It did not prove that enterprise AI demand is weak, that cloud growth is ending, or that every semiconductor and data-center investment will earn a poor return.

Meta’s result was especially damaging to the broad bullish narrative because the company’s operating business was not collapsing. Revenue reached $60.80 billion, up from $47.52 billion a year earlier. Ad impressions across the Family of Apps increased 14%, while average price per ad rose 12%. Daily active people across Meta’s family of services averaged 3.60 billion in June, up 3% year over year. Those are not the statistics of a platform losing relevance.

The problem was the conversion of growth into profit and cash. Costs and expenses jumped 55% to $42.03 billion. Operating income fell 8% to $18.78 billion, and operating margin contracted from 43% to 31%. Net income declined 14% to $15.85 billion, while diluted earnings per share fell 13% to $6.18. Free cash flow, a non-GAAP measure that Meta defines as net cash from operating activities minus purchases of property and equipment and principal payments on finance leases, fell to $784 million from $8.55 billion.

A company can report strong revenue growth and still worry investors when the incremental dollar of sales arrives alongside a much larger increase in infrastructure and operating cost. That is the central financial issue in the current AI cycle. The debate is not about whether Meta can sell more advertising. It plainly can. The debate is whether management can turn a vast expansion in computing capacity, AI talent, data centers, networking equipment, and energy commitments into durable returns that exceed the company’s cost of capital.

Microsoft’s quarter showed why blanket conclusions are hazardous. Revenue increased 18% to $90.01 billion. Operating income rose 18% to $40.60 billion. GAAP net income climbed 31% to $35.77 billion, though part of that increase reflected investment-related effects that require adjustment. Azure and other cloud-services revenue rose 43%, well ahead of the roughly 40% consensus estimate cited by Reuters. Microsoft Cloud revenue increased 27% to $59.3 billion, and commercial remaining performance obligations—the contracted revenue backlog not yet recognized—increased 84% to $678 billion.

Microsoft also reported more than 30 million paid Microsoft 365 Copilot seats, up from 20 million in the previous quarter. That metric does not settle the long-term economics of generative AI, but it is direct evidence of monetization rather than a promise that monetization will emerge later. For investors evaluating the AI buildout, the contrast is difficult to ignore: Meta is primarily demonstrating better engagement and ad performance while absorbing an extraordinary capital bill; Microsoft is showing those infrastructure costs alongside cloud usage, contractual commitments, and paid software adoption.

The result is a selective rather than universal bear case. Meta’s after-hours decline validated the claim that its report could influence sentiment across AI-linked assets. Microsoft’s rally challenged the idea that hyperscaler spending should be treated as one undifferentiated bubble. The same technological cycle can produce winners, overbuilders, suppliers with exceptional pricing power, and customers that discover their returns too late.

What Tim Knight argued before the reports

Knight’s argument combined technical analysis, market psychology, and a macro view of AI spending. He described a market in which bearish opportunities were shifting from one sector to another rather than appearing as a uniform decline. He had reduced some concentrated technology exposure, moved toward broader S&P 500 put options, and extended the maturity of his largest positions to January 2027. The longer expiry, he said, allowed him to sleep more easily even though he was paying for more time than he might need.

That comment contains an important distinction between a market forecast and a trading structure. A trader can be directionally correct and still lose money if timing, volatility, position size, or option decay works against the position. Extending an option’s expiration reduces the immediate pressure of time decay relative to a very short-dated contract, but it increases the premium paid and does not eliminate the possibility of a complete loss. Long-dated puts can also decline even when the underlying market is flat or modestly lower if implied volatility falls or if the move is too slow to overcome the premium embedded in the contract.

Knight’s sector thesis focused heavily on technology and semiconductors. He discussed the Technology Select Sector SPDR Fund, known by the ticker XLK; the Invesco QQQ Trust, which tracks the Nasdaq-100; and the VanEck Semiconductor ETF, SMH. He viewed completed chart patterns and breaks below support as evidence that the trend had turned bearish. The semiconductor fund was his strongest conviction, reflecting a belief that the “AI unraveling” had only begun.

He also argued that Meta was a more consequential earnings event than Microsoft because Meta’s result could shape the narrative around AI spending across the market. That claim was not irrational. Meta has no cloud business comparable to Azure that can sell excess infrastructure externally at scale. Its payoff depends heavily on improving advertising, recommendations, engagement, messaging, business tools, and future products enough to justify investment that is consuming a far larger share of cash generation. A weak Meta report therefore speaks directly to the market’s most uncomfortable question: what if infrastructure is being built faster than profitable demand develops?

Still, Knight’s dismissal of Microsoft underestimated how much the market needed a credible counterexample. Microsoft’s quarter did not merely beat an earnings estimate. It showed that one of the world’s largest buyers of AI infrastructure was experiencing faster Azure growth, a larger contracted backlog, and paid adoption of its flagship AI productivity product. Those facts do not guarantee attractive returns on every dollar of capital expenditure, but they weaken a sweeping claim that hyperscaler spending is already unraveling across the board.

The better interpretation is that Knight correctly identified the event risk and the market’s vulnerability, while assigning too little informational value to Microsoft. Meta was the more fragile narrative. Microsoft was the more important test of whether the commercial side of the AI buildout could offset the capital burden. Both mattered, but for different reasons.

The Fed held rates steady, yet the decision was not neutral

The Federal Reserve’s July decision was widely expected in its headline form. The Federal Open Market Committee maintained the federal-funds target range at 3.50% to 3.75%. The surprise was the degree of dissent. Beth Hammack, Neel Kashkari, and Lorie Logan preferred a quarter-point increase, producing a 9–3 vote. The official statement said economic activity was expanding at a solid pace, productivity growth and capital investment were strong, job gains had kept pace with the workforce, and unemployment had changed little.

The inflation language was blunt. The Committee said inflation remained elevated relative to the 2% goal, partly because supply shocks had raised prices in sectors including energy, and declared that it “will deliver price stability.” Fed Chair Kevin Warsh later said the central bank would not waver in returning inflation to target. According to Reuters, market pricing after the announcement assigned roughly a 57% probability to a September rate increase.

That combination—a hold today, a more divided committee, and an explicit willingness to tighten later—was not an easy backdrop for expensive growth stocks. Valuation in technology and AI-sensitive shares depends partly on the present value of cash flows expected far into the future. Higher bond yields and a higher expected policy path increase the discount rate applied to those future profits. They also raise financing costs for data centers, utilities, suppliers, and customers participating in the buildout.

The Fed’s problem is complicated by the source of current inflation pressure. Oil prices surged on July 29 as conflict in the Middle East intensified and U.S. crude inventories fell. Brent settled at $90.74 a barrel, up 7.91%, while West Texas Intermediate gained 6.56% to $84.46. A central bank cannot produce oil or reopen a shipping lane. Raising interest rates in response to an energy supply shock can suppress demand, but it cannot directly repair the supply disruption. That is why some economists argue that the Fed should avoid overreacting to temporary commodity spikes, while others worry that repeated shocks will become embedded in inflation expectations and wages.

AI investment adds another layer. The Fed statement described capital investment and productivity as strong, while Warsh has emphasized the possibility that AI could raise the economy’s productive capacity. Yet the same investment boom can increase near-term demand for power, construction labor, electrical equipment, cooling systems, networking hardware, land, and financing. The economy can therefore experience a period in which AI is potentially disinflationary in the long run but inflationary in the buildout phase.

For markets, this means the Fed and AI earnings are not separate stories. They are linked through capital intensity, energy demand, financing costs, and the duration of expected profits. A company spending $100 billion or more annually on infrastructure faces a very different hurdle rate when Treasury yields are rising and the central bank is debating additional tightening. The longer the payoff lies in the future, the more sensitive the valuation becomes to interest rates and execution risk.

Wall Street was already weakening before Meta and Microsoft reported

The regular session established a bearish foundation before either earnings release. The S&P 500 fell 1.52% to 7,316.15, its lowest close in about a month. The Nasdaq Composite lost 1.74% to 24,442.94, while the Dow Jones Industrial Average declined 2.19% to 51,594.14. The Nasdaq-100 fell 2.1%, leaving it 11% below its June record. Information technology was among the weakest sectors, and semiconductor shares extended an already severe retreat.

ETF moves illustrated the concentration of the decline. The SPDR S&P 500 ETF Trust fell about 1.5%. QQQ lost roughly 2.0%. XLK declined approximately 2.6%, and SMH dropped about 4.8%. Those figures were not simply reactions to the Fed. They reflected a broader repricing of AI-linked assets after weeks of concern about capital spending, crowded positioning, geopolitical risk, oil, and earnings expectations that had become difficult to exceed.

Semiconductor investors had already received a warning from South Korea. SK Hynix shares fell 10% even after the memory-chip producer reported a sixfold increase in quarterly profit, because the result failed to satisfy exceptionally high expectations. Vertiv, a supplier of power and cooling equipment for data centers, fell 17% after missing quarterly revenue estimates. The market was signaling that exposure to the AI buildout was no longer enough. Companies had to beat elevated forecasts and provide confidence that demand, margins, and order visibility remained intact.

This is a common transition in investment cycles. During the early phase, a convincing narrative and rapid revenue growth can lift nearly every company associated with the theme. As valuations rise and capital floods into supply, the market becomes more discriminating. Investors begin separating firms with contractual demand from those relying on forecasts, profitable customers from speculative customers, and infrastructure that is immediately productive from capacity built for demand that may arrive later.

The July selloff also demonstrated why a single index can conceal important rotation. The day before, the Dow had advanced sharply while many technology shares weakened. A price-weighted industrial index can rise because a handful of large components rally even as market-cap-weighted technology benchmarks deteriorate. Knight’s observation that traders were “scurrying” from sector to sector captured this unevenness. The market was not moving as one unit. Defensive sectors, industrial names, financial stocks, energy, and selected consumer businesses could behave differently from semiconductors and hyperscalers.

That matters for broad S&P 500 puts. The index is diversified, but it is also float-adjusted market-cap weighted. Its largest companies exert disproportionate influence. A severe decline in mega-cap technology can pull the index lower even if many smaller constituents hold up. The reverse is also true: strength in a few giants can keep the headline index resilient while the average stock weakens. A trader relying on the S&P 500 therefore needs to understand both index concentration and market breadth.

Why semiconductors became the pressure point

Semiconductors sit at the center of the AI investment cycle because nearly every stage of the buildout depends on chips. Advanced accelerators perform model training and inference. High-bandwidth memory feeds data rapidly to those processors. Networking chips connect clusters. Central processors, storage controllers, power-management components, and analog devices support the system around them. Semiconductor-manufacturing equipment companies provide the tools needed to fabricate increasingly complex devices.

This centrality created exceptional revenue growth and pricing power for parts of the industry. It also produced a dangerous feedback loop in expectations. Customers announced larger capital budgets, suppliers raised forecasts, investors extrapolated growth, and market values increased. High valuations then required not merely good results but continued acceleration. Any evidence that customers were delaying orders, financing was tightening, utilization was low, or spending was shifting could generate an outsized price response.

SMH is an especially concentrated expression of that cycle. VanEck describes the fund as tracking 25 U.S.-listed companies involved in semiconductor production and equipment. Its holdings include both designers and manufacturers, but the fund’s performance can be dominated by a small number of very large positions. It therefore carries more thematic and company-specific risk than a broad-market ETF.

The semiconductor bear case has several layers. First, hyperscaler customers may eventually slow capital spending if free cash flow deteriorates or if AI revenue develops more slowly than expected. Second, supply can catch up with demand, reducing scarcity pricing. Third, customers may design more of their own chips, shift workloads toward cheaper inference hardware, or adopt more efficient models. Fourth, geopolitical restrictions and export controls can alter addressable markets. Fifth, the financing structures supporting data-center development may become more expensive as credit investors demand higher yields.

The bullish case is equally substantial. Demand remains constrained by available capacity in several areas. Microsoft said capacity limits were still restraining cloud growth. Paid Copilot adoption and Azure growth suggest that enterprise demand is not hypothetical. Meta’s advertising metrics imply that AI is already improving recommendation and monetization systems. If models become cheaper to run, lower unit cost could expand total usage rather than reduce aggregate chip demand—a version of the economic effect in which efficiency stimulates more consumption.

The question is therefore not whether chips are necessary. It is whether the industry has priced a demand curve that is too smooth, too permanent, and too profitable. Semiconductor equities can decline sharply even while end demand remains healthy if valuations had assumed near-perfect execution. That distinction is crucial. A falling SMH chart does not prove that AI adoption has stopped. It can mean that the market’s expected return on the theme was too high.

XLK, QQQ, and SMH are different trades

Technical discussions often group technology ETFs together, but their portfolios answer different questions. XLK tracks the information-technology sector of the S&P 500. It includes software, hardware, semiconductor, and technology-services companies classified within that sector. It does not include every company investors casually call “Big Tech.” Meta and Alphabet, for example, are classified in communication services rather than information technology. Amazon is classified in consumer discretionary.

That classification detail is important when evaluating Knight’s claim that Meta could determine the direction of XLK. Meta is not an XLK holding, but its earnings can influence XLK indirectly by changing sentiment toward AI spending, digital advertising, cloud infrastructure, semiconductors, and mega-cap growth. Microsoft, Apple, Nvidia, Broadcom, Micron, AMD, and other technology-sector companies are more direct drivers of the fund.

QQQ tracks the Nasdaq-100, an index of the largest nonfinancial companies listed on Nasdaq. It crosses sector boundaries and therefore includes technology, communication-services, consumer, healthcare, and other companies. Meta can have a direct impact on QQQ because it is a Nasdaq-listed constituent. Microsoft also carries substantial weight. QQQ is often described as a technology fund, but it is more accurately a concentrated large-cap growth portfolio shaped by Nasdaq listing rules.

SMH is narrower. It is designed to track U.S.-listed semiconductor production and equipment companies. That makes it more sensitive to chip demand, memory pricing, equipment orders, foundry utilization, export controls, and AI infrastructure spending. A 4.8% daily decline in SMH can occur while the broader S&P 500 falls much less because the fund has little exposure to banks, healthcare, staples, utilities, or other areas that might stabilize a diversified index.

These differences affect both analysis and risk management. A bearish view on Meta does not automatically imply the same magnitude of downside in XLK. A bearish view on semiconductor capital spending is more directly expressed through SMH than through QQQ. A broad concern about mega-cap valuation and economic risk may be more appropriately studied through the S&P 500 or Nasdaq-100. Each instrument contains a different mixture of company risk, sector risk, and macro sensitivity.

ETF moves are based on market data shortly after the July 29, 2026 close and are rounded. Fund composition can change over time.

What a “right triangle” breakdown can and cannot tell investors

Knight’s technical case centered on price patterns he described as right triangles. In chart analysis, a triangle generally represents a period of compression in which price trades between converging or nearly converging boundaries. A break below support is interpreted as evidence that sellers have gained control. A bearish engulfing candle, another pattern he cited, occurs when a down-period price range covers or “engulfs” the prior up-period body, suggesting a reversal in momentum.

These patterns can be useful as risk-management frameworks because they define observable levels. A trader can identify where the thesis would be invalidated, where a stop might be placed, and whether the market is making lower highs or lower lows. They also help separate opinion from action. Instead of shorting because a headline feels negative, the trader waits for a price break and evaluates whether the market confirms the narrative.

But chart patterns are not causal explanations. A triangle does not know whether Azure growth is 35% or 43%, whether Meta’s free cash flow is $784 million, or whether the Fed will raise rates in September. The pattern summarizes the behavior of buyers and sellers before the facts are fully known. It can fail when new information changes those expectations.

Breakdowns are also vulnerable to false signals. Price can move below support during a volatile session, trigger stops, and then recover into the prior range. An earnings beat, geopolitical de-escalation, lower inflation reading, or dovish Fed communication can reverse a technically bearish setup. Conversely, an apparently resilient support level can collapse rapidly when several risks arrive together.

The strongest use of technical analysis in this context is therefore conditional rather than prophetic. A trader might say: the trend is bearish while price remains below a defined level; the position will be reduced if price re-enters the prior range; and exposure will be sized so that a gap against the position is survivable. That is different from claiming that a completed pattern guarantees a target.

Knight’s own behavior showed an awareness of this uncertainty. He extended option maturities, reduced some exposure, and discussed the psychological burden of concentrated positions. Those choices do not prove the forecast is correct, but they acknowledge that being early can be financially and emotionally expensive. Markets can remain irrational relative to a trader’s model for longer than a short-dated option remains alive.

Meta’s second quarter: strong sales, weaker economics

Meta’s second-quarter report was not a conventional earnings collapse. The advertising engine remained powerful. Revenue of $60.80 billion was near the top of the company’s prior $58 billion to $61 billion guidance range and represented 28% year-over-year growth. Ad impressions increased 14% and average price per ad rose 12%, indicating that growth came from both additional inventory and better monetization.

The deterioration appeared below the revenue line. Costs and expenses increased by $14.95 billion from the prior-year quarter to $42.03 billion. The resulting $18.78 billion of operating income was $1.67 billion lower than a year earlier even though revenue increased by $13.29 billion. That is a severe negative operating-leverage outcome: the company generated substantial additional sales, but incremental costs more than absorbed the benefit.

Part of the cost increase was connected to a $2.4 billion charge related to legal proceedings. It would be misleading to treat every dollar of the margin decline as recurring AI expense. Yet the broader spending trend is unmistakable. Meta raised the lower end of expected 2026 total expenses from $162 billion to $165 billion, maintaining the upper end at $169 billion. It also narrowed capital-expenditure guidance to $130 billion to $145 billion from $125 billion to $145 billion.

Capital expenditure does not flow through the income statement all at once. Purchases of servers, buildings, network equipment, and related infrastructure are recorded as assets and expensed over time through depreciation. That accounting treatment means the immediate pressure can appear more dramatically in cash flow than in reported operating profit. As the asset base grows, depreciation expense rises in future periods, creating a delayed income-statement burden even if capital spending later stabilizes.

Meta’s free cash flow illustrates that timing. The company still produced significant operating cash flow, but purchases of property and equipment and lease-related payments consumed nearly all of it. Free cash flow of $784 million was down from $8.55 billion a year earlier. A 91% decline does not mean Meta is insolvent or unable to fund operations. It means the margin of cash available for buybacks, dividends, acquisitions, debt reduction, or additional flexibility shrank dramatically in the quarter.

That distinction matters. Meta ended the prior quarter with a large cash position and remains highly profitable. It has access to capital markets and a dominant advertising franchise. The concern is not near-term survival. It is capital allocation. Investors must decide whether management is investing through a temporary bottleneck that will produce exceptional future returns or committing too much capital before the revenue model is sufficiently proven.

Why Meta’s free-cash-flow decline became the night’s defining number

Revenue growth often dominates an earnings headline because it is the clearest evidence of customer demand. For an infrastructure-heavy AI cycle, however, the more revealing question is how much of that revenue survives after the company pays to operate and expand the business. Meta’s second quarter demonstrated why free cash flow has moved to the center of the hyperscaler debate.

Free cash flow is commonly calculated as operating cash flow minus capital expenditures. It is not a standardized measure under U.S. generally accepted accounting principles, and companies can define it differently, particularly when finance leases or other asset purchases are involved. The concept remains useful because it approximates the cash generated after maintaining and building the productive asset base. When that figure falls sharply while revenue rises, investors naturally ask whether the investment cycle is creating durable future capacity or merely consuming current economics.

Meta’s $784 million of quarterly free cash flow was especially striking because the company’s advertising operations remained healthy. The quarter did not expose a collapse in user activity, ad demand, or pricing. Family daily active people reached 3.60 billion, up 3% from a year earlier. Ad impressions rose at a double-digit rate and average price per ad also increased. This was not a business that had stopped selling. It was a business spending much faster than its operating gains could replenish cash.

That produces an unusual analytical tension. A weak company may cut investment because it lacks confidence or financing capacity. Meta is doing the opposite. Management is spending from a position of strength because it believes AI infrastructure is necessary to improve recommendations, advertising tools, consumer products, and future computing platforms. The bearish interpretation is not that Meta has lost the ability to earn money. It is that management may be using a highly profitable legacy franchise to fund an uncertain infrastructure race with a less visible return profile.

The scale of the guidance matters. At the midpoint, Meta’s 2026 capital-expenditure range is $137.5 billion. That is not a one-quarter experiment. It implies a multiyear industrial program involving data centers, accelerators, networking, power, cooling, land, construction, and software development. Even if each individual project has a rational business case, the portfolio can still disappoint if demand arrives more slowly than planned, technology changes before assets are fully utilized, or competitive pricing prevents the company from capturing the economic value of its investment.

Conversely, the free-cash-flow decline could prove temporary if the infrastructure creates a step-change in monetization. Better recommendation systems can increase time spent in Meta’s apps. Improved ad ranking can raise conversion rates and let advertisers bid more confidently. Generative tools can reduce the cost of producing campaigns, expand the population of businesses capable of advertising, and increase the number of creative variants tested. If these gains compound across billions of users and millions of advertisers, modest improvements in efficiency can produce very large revenue effects.

The second-quarter report therefore did not settle the AI-return debate. It sharpened it. The income statement showed that Meta’s core machine remains formidable. The cash-flow statement showed how much the company is willing to sacrifice today to defend and expand that machine. The after-hours selloff reflected the market’s discomfort with the gap between those two facts.

Meta’s advertising results are evidence of AI value, but not a complete payback calculation

Meta has a stronger near-term AI monetization story than a simple “spending with no revenue” narrative suggests. Its advertising system has long used machine learning to predict which content and advertisements are most relevant to each user. Generative models and more capable recommendation systems can improve that process. When Meta reports rising impressions and higher average prices in the same quarter, it indicates that advertisers are willing to pay for additional inventory and that the platform is still converting engagement into revenue.

AI-assisted creative tools are also economically important even when they do not carry a separate subscription price. A small business that can generate images, text, and campaign variations inside Meta’s advertising interface may spend more because the cost and expertise required to launch a campaign decline. An advertiser that receives better targeting or measurement may allocate a larger share of its budget to Meta. Those effects appear inside advertising revenue rather than as a line labeled “AI sales.”

This makes Meta different from a software company selling a distinct AI seat or a cloud provider billing customers for model training and inference. The return is embedded in the performance of the existing platform. That can be an advantage because Meta does not need to persuade every user to buy a new product. It can deploy AI across products with enormous existing distribution. It can also make the return harder to audit from outside the company. Investors see revenue growth, but they cannot precisely separate the contribution of AI from changes in the economy, ad pricing, engagement, competition, or product design.

Management can point to better recommendations and advertiser outcomes, but a rigorous payback calculation would require additional disclosure: incremental revenue attributable to particular AI systems, the cost of training and serving them, utilization rates for the infrastructure, useful asset lives, and the degree to which older systems have been displaced. Public financial statements do not provide that level of granularity. The absence of such detail is not unusual, but it leaves room for both optimistic and skeptical interpretations.

A further complication is that AI can be both offensive and defensive spending. Some investment may generate new revenue. Some may be necessary simply to prevent users, creators, and advertisers from migrating to competitors with better products. Defensive investment can still be rational, but its return is measured partly in revenue not lost rather than revenue newly created. That makes the counterfactual impossible to observe directly.

Meta’s 28% revenue growth offers meaningful evidence that the company is not pouring capital into a stagnant franchise. Yet the decline in operating margin and free cash flow shows that strong demand alone does not guarantee attractive incremental returns. The key question is not whether AI helps Meta. It almost certainly does. The question is whether the next dollar of infrastructure spending will produce enough additional cash over its life to justify the cost and risk.

Capital expenditure today becomes depreciation tomorrow

The accounting treatment of data-center investment can delay the full impact on reported earnings. When Meta buys servers or constructs a facility, most of the cash outlay is capitalized on the balance sheet rather than immediately recognized as an operating expense. The company then records depreciation over the estimated useful life of the asset. That spreads the expense across future reporting periods.

This is sensible accounting because infrastructure is expected to produce benefits for more than one quarter. It also means investors should not assume that a current operating margin fully reflects the economic cost of the current buildout. The cash leaves first. Depreciation follows later. If capital spending keeps rising, each new group of assets adds another layer of future depreciation.

Useful-life assumptions matter. AI accelerators may remain technically functional for years, but their economic productivity can decline faster if newer chips deliver materially better performance per watt or per dollar. A company can redeploy older equipment to less demanding workloads, but rapid technological change raises the risk that accounting lives exceed the period of peak economic value. Shortening useful lives would accelerate depreciation; keeping them longer would preserve near-term earnings but could increase the risk of future impairments or weaker returns.

Power and cooling infrastructure may have much longer lives than chips. Buildings, substations, land improvements, networking equipment, and servers therefore create a mixed asset base with different replacement cycles. The distinction matters because not all “AI capex” has the same obsolescence risk. A well-located data-center campus with secured power can remain valuable even as the processors inside it are replaced. A specialized chip purchased near the top of a cycle may depreciate economically much faster.

Lease financing can also complicate comparisons. A company may build and own facilities directly, enter long-term leases, or use joint ventures and project financing. Each structure changes the timing and presentation of cash flows, assets, liabilities, and expenses. It does not eliminate the economic obligation to pay for capacity. Analysts therefore need to look beyond a single capital-expenditure line and consider lease commitments, depreciation, interest, and any guarantees or residual exposure.

The practical implication is that Meta’s investment debate will extend beyond 2026. Even if capital spending reaches a plateau, depreciation and operating costs associated with the installed base can continue rising. The bullish case requires revenue and gross profit to scale faster than those expenses. A temporary pause in cash outflow would not by itself prove that the investment generated an adequate return.

The BlackRock data-center partnership shows how Meta is widening its financing toolkit

One day before the earnings report, Reuters reported that Meta and BlackRock had formed a joint venture tied to a roughly $14 billion data-center project in El Paso, Texas. BlackRock would hold an 80% interest and Meta 20%, supported by approximately $12.5 billion of debt. Meta was expected to contribute land and construction valued at about $2.3 billion, receive a distribution of about $1 billion, and lease capacity from the venture.

The structure illustrates how hyperscalers can expand capacity without funding every dollar through traditional company-owned capital expenditure. Institutional capital can finance long-lived physical assets while the technology company commits to use the capacity. For investors seeking infrastructure-like returns, a lease backed by a large investment-grade tenant can be attractive. For Meta, the arrangement can reduce the immediate cash burden and diversify funding sources.

It would be wrong, however, to treat external financing as free capacity. Lease payments remain an economic cost. Depending on contract terms and accounting classification, the arrangement can create recognized assets and liabilities, recurring expenses, or substantial future commitments. The financing changes who supplies the capital and when cash is paid; it does not remove the need for the underlying project to earn an acceptable return.

The partnership also signals that the AI buildout is becoming a capital-markets story, not merely a technology story. Data centers are increasingly connected to private credit, infrastructure funds, utility investment, power purchase agreements, land markets, and municipal development. The returns and risks are distributed across a wider set of investors. A slowdown in AI demand would therefore affect more than semiconductor shares. It could influence debt spreads, project-finance assumptions, construction pipelines, and power-market planning.

For Meta shareholders, the relevant question is whether such structures improve capital efficiency without obscuring total obligations. Transparent disclosure of lease terms, guarantees, utilization, and expected economics will become more important as financing grows more complex. A company can reduce reported near-term capex while still committing itself to many years of payments. Economic leverage matters even when it does not resemble a conventional bond issuance.

Microsoft delivered the counterexample to an indiscriminate AI bear case

Knight said Microsoft barely registered compared with Meta. As a market call about the night’s downside catalyst, that instinct had merit: Meta fell after hours and became the immediate source of anxiety. As a fundamental judgment about the AI trade, however, Microsoft’s report was too important to dismiss.

Microsoft reported $90.01 billion of revenue for its fiscal fourth quarter, an 18% increase from the prior year. Operating income rose 18% to $40.60 billion. GAAP net income increased 31% to $35.77 billion, and diluted earnings per share rose 32% to $4.81. Microsoft Cloud revenue reached $59.3 billion, up 27%.

The standout figure was Azure and other cloud-services growth of 43%, above the roughly 40% rate expected by analysts surveyed by Visible Alpha, according to Reuters. That result suggested that demand remained strong enough to absorb expanding capacity. Microsoft’s commercial remaining performance obligation reached $678 billion, up 84%, providing a large contracted or committed revenue backlog, although the timing of recognition varies and not every dollar is guaranteed to become revenue on the same schedule.

Microsoft also said paid Microsoft 365 Copilot seats exceeded 30 million, compared with 20 million in the previous quarter. A seat count is not the same as disclosed revenue or profit, and pricing may vary across customers and bundles. It nevertheless provides a more visible unit of AI adoption than an abstract claim that users are experimenting. Enterprise customers are moving from pilots toward paid deployment at material scale.

The report therefore offered the strongest available rebuttal to the idea that hyperscaler investment is uniformly outrunning monetization. Microsoft is spending heavily, but it can point to accelerated cloud growth, expanding contracted backlog, and paid AI seats. Its AI infrastructure serves external customers that pay for computing capacity as well as Microsoft’s own software products. That diversified revenue architecture gives the company several ways to monetize the same physical and software platform.

None of this makes Microsoft’s spending riskless. Capital expenditure was about $41 billion in the quarter, up approximately 70% from a year earlier, according to Reuters. The company was expected to spend about $190 billion in calendar 2026. Capacity needs, power availability, model economics, and customer willingness to pay remain critical. But the quarter showed that a blanket claim of “AI unraveling” misses meaningful differences among companies.

Why Azure growth matters more than a one-night share-price move

After-hours prices are informative but incomplete. They reflect the first wave of interpretation, often before executives have answered analyst questions and before investors have fully modeled guidance. Microsoft’s roughly 3% extended-trading rise showed relief, but the more durable evidence was operational.

Azure growth answers a central question in the AI cycle: is deployed capacity finding paying customers? A 43% growth rate at Microsoft’s scale indicates strong demand for cloud infrastructure and associated services. Some of that growth is conventional computing, databases, cybersecurity, and enterprise migration rather than generative AI. Microsoft does not provide a complete revenue breakdown that isolates every AI workload. Still, the acceleration is inconsistent with a broad collapse in customer demand.

Remaining performance obligations offer another lens. The $678 billion balance includes contracted revenue expected to be recognized over time. It provides visibility, but it should not be treated as cash in the bank. Contracts can contain termination rights, variable components, or recognition schedules extending for years. The figure can also rise when companies sign large multiyear commitments that require continued investment to serve. Backlog is evidence of demand, not a guarantee of margin.

The increase in paid Copilot seats matters because the enterprise-software model can distribute AI at high scale. Microsoft already has relationships, identity systems, security controls, and workflows embedded across corporate customers. It can sell AI as an add-on, bundle, consumption service, or feature within products customers use every day. Distribution reduces the cost of customer acquisition compared with a new entrant building an enterprise sales organization from scratch.

Microsoft can also monetize the same infrastructure through several layers: Azure compute, model access, developer tools, data platforms, security products, and end-user applications. This does not mean each layer will retain premium margins. Customers may resist high prices, open-source models may lower costs, and competition from Amazon, Google, Oracle, and specialized providers may intensify. But a multilayered revenue model creates more opportunities to recover infrastructure spending.

That is why Microsoft complicates the semiconductor short thesis. Strong cloud demand can support purchases of accelerators, networking equipment, memory, and power systems even if one platform company disappoints. Chip suppliers are exposed to the aggregate capital budgets of many customers, not only Meta. A decisive bearish case therefore requires evidence that industry-wide orders, utilization, or customer budgets are weakening—not merely that one company’s quarterly free cash flow declined.

Microsoft’s quarter still contained reasons for caution

A strong earnings beat is not proof that every dollar of future AI spending will earn an attractive return. Microsoft’s capital intensity is increasing rapidly. Roughly $41 billion of quarterly capital spending is extraordinary even relative to the company’s cash generation. Building capacity ahead of demand can support growth, but it also raises fixed costs and the risk of underutilization if customers slow deployments.

Cloud growth can also be constrained by supply rather than demand. When a provider says capacity is tight, revenue may accelerate as new facilities come online. That is encouraging, but it can temporarily conceal the economics of incremental supply. Investors still need to know how much revenue each unit of capital produces, how quickly new assets reach utilization targets, and whether gross margins remain resilient as depreciation and energy costs rise.

Microsoft’s GAAP earnings also benefited from investment-related gains. The company disclosed a $3.2 billion gain related to Anthropic and other discrete items that provided a net benefit of approximately $0.27 per share. Its non-GAAP results excluded the impact of OpenAI investments. Those adjustments do not invalidate the quarter, but they demonstrate why headline EPS should be separated from operating performance. Investment marks can be volatile and may reverse.

The company’s consumer businesses were less uniformly strong. More Personal Computing revenue declined 4%. Windows OEM and Devices revenue fell 7%, while Xbox content and services revenue decreased 10%. Microsoft is increasingly defined by cloud and enterprise software, but weaker consumer segments show that not every part of the portfolio is participating equally.

Concentration in a few giant customers and AI partners may create additional risks. Large cloud contracts can be strategically valuable but expensive to serve. Model developers can negotiate pricing, shift workloads, raise capital elsewhere, or build more infrastructure themselves. Partnerships can blur the line between customer, supplier, investee, and competitor. The economic substance of those relationships can be harder to evaluate than ordinary third-party sales.

Finally, a successful product does not automatically justify any valuation. Even excellent businesses can produce poor investment returns when expectations are too high. Microsoft’s report strengthened the operating case, but the stock’s future performance still depends on the price investors pay, the durability of growth, and the eventual margin on AI services.

The AI investment cycle is separating buyers, builders, and beneficiaries

The phrase “AI trade” compresses several distinct business models into one label. That simplification was useful when most related shares rose together. It is less useful as spending grows and investors demand evidence of returns.

Hyperscalers and platform companies

Microsoft, Amazon, Alphabet, and Meta commit the largest capital budgets. They purchase chips, build data centers, secure electricity, hire researchers, and develop products. Their risk is that spending grows faster than profitable demand. Their advantage is distribution, existing cash flow, and the ability to monetize across many services.

Semiconductor designers and manufacturers

Chip designers, foundries, memory producers, networking suppliers, and equipment makers benefit from the construction phase. Their near-term revenue can grow before end customers prove the full economic return. This “picks-and-shovels” position can be attractive, but it is cyclical. If hyperscalers pause orders after a period of overbuilding, suppliers can face inventory corrections, pricing pressure, or lower utilization.

Data-center and power infrastructure

Utilities, electrical-equipment manufacturers, cooling providers, construction companies, real-estate developers, and infrastructure funds participate in the physical buildout. Their demand can be supported by long-term contracts, but projects face permitting, interconnection delays, community opposition, and uncertainty over future load. A data center announced today may require years before becoming fully operational.

Enterprise software and application developers

Application companies attempt to turn model capability into workflow value. Their economics depend on whether customers will pay enough to cover model-inference costs and whether incumbents bundle comparable features. Some applications may create new categories; others may become features inside larger platforms.

End users

Businesses and consumers ultimately determine whether the investment earns a return. Productivity gains must become measurable savings, higher revenue, better decisions, or improved experiences. Adoption can be broad without being profitable if usage is subsidized or if competition drives prices toward cost.

This segmentation explains why Meta and Microsoft can report divergent economics on the same night. They are both major AI investors, but their monetization paths differ. Meta primarily improves a consumer advertising platform and explores new products. Microsoft sells infrastructure and enterprise software directly. A bearish thesis that treats both as interchangeable overlooks the structure of their cash flows.

What Reuters’ hyperscaler cash-flow analysis adds to the debate

Reuters reported in July that major technology companies’ capital spending was on course to grow faster than free cash flow, with projected incremental capital expenditure exceeding incremental operating cash flow by 2027. The analysis estimated roughly $534 billion of additional capital investment for every $340 billion of added operating cash flow across the group—about $1.57 of investment for each $1 of additional cash generation.

Those projections are estimates rather than audited outcomes, and company-level results will vary. They nevertheless capture the core concern behind Knight’s short thesis. A capital cycle becomes dangerous when participants extrapolate demand, compete for scarce inputs, and commit resources faster than the underlying cash flows mature.

The same data can support a less bearish interpretation. Companies with unusually strong balance sheets may rationally invest ahead of demand because the strategic cost of lacking capacity is high. A dollar invested today may produce cash over many years, so comparing current capex with current incremental operating cash flow can understate future value. Early infrastructure is often inefficient before utilization rises.

The decisive issue is return on invested capital over the asset life. That calculation requires assumptions about revenue growth, margins, useful lives, maintenance spending, and the residual value of infrastructure. No single quarterly free-cash-flow figure can answer it. But when several companies spend hundreds of billions simultaneously, the burden of proof rises. Investors become less willing to accept “capacity will be needed” without evidence of pricing power and utilization.

Why semiconductor shares fell harder than the broad market

SMH’s approximately 4.8% decline on July 29 was more than three times the S&P 500’s percentage loss. That relative weakness reflected more than Meta’s report, which arrived after the close. The semiconductor group had already been under pressure as investors questioned the durability of AI spending, absorbed company-specific earnings, and confronted a tighter-rate backdrop.

Semiconductors are highly sensitive to changes in expectations because the supply chain is long and capital intensive. Hyperscaler budgets support accelerator orders. Accelerator demand supports foundry capacity, advanced packaging, high-bandwidth memory, networking, and manufacturing equipment. When investors reduce the assumed growth rate at the top of that chain, the valuation impact can ripple through many companies.

The same chain can amplify upside. If Microsoft’s cloud demand remains above expectations and Meta maintains a $130 billion to $145 billion capex range, chip demand does not disappear. The market’s challenge is timing. Suppliers may have strong order books today while investors worry that 2027 or 2028 budgets will normalize. Stocks discount future earnings, so they can fall before reported revenue turns down.

Not every semiconductor company has the same exposure. Some are concentrated in leading-edge AI accelerators. Others sell memory, analog chips, automotive components, personal-computer processors, communications devices, or manufacturing equipment. A sector ETF diversifies company-specific risk but can obscure these differences. A slowdown in one end market may coexist with strength in another.

Valuation also affects sensitivity. When a stock price assumes years of exceptional growth, a modest reduction in expected revenue or margin can produce a large decline in present value. This does not require an outright collapse in demand. It requires only that the future be less extraordinary than the price implied.

Higher oil and tighter monetary policy created a difficult backdrop for long-duration growth stocks

The AI earnings debate did not occur in isolation. Crude oil surged on July 29 as Middle East conflict disrupted shipping expectations and U.S. inventories fell sharply. Brent settled at $90.74 a barrel, up 7.91%, while West Texas Intermediate settled at $84.46, up 6.56%, according to Reuters. U.S. crude inventories fell by 7.2 million barrels to 404.5 million, their lowest level since 2018.

Higher oil can affect technology valuations through several channels. It can raise headline inflation, transportation costs, and household expenses. If the shock persists, it can reduce the Federal Reserve’s willingness to cut rates and increase the possibility of a hike. Higher bond yields raise the discount rate applied to future corporate cash flows, which tends to weigh more heavily on companies whose valuation depends on profits expected far into the future.

Data centers themselves consume large amounts of electricity, though their direct exposure to crude oil varies by region and power source. The broader issue is energy scarcity. AI infrastructure requires reliable power, transmission capacity, backup generation, and cooling. A world of constrained energy supply can raise construction and operating costs even when oil is not the primary fuel used by a facility.

The Federal Reserve’s decision reinforced this pressure. Holding the federal-funds target at 3.50% to 3.75% was not a dovish pause because three policymakers preferred an increase and the statement emphasized elevated inflation. Reuters reported that market pricing moved to imply a 57% probability of a September rate increase after the announcement. The 30-year Treasury yield rose above 5.20% for the first time since 2007.

For cash-rich technology companies, higher rates do not create the same refinancing crisis they might create for a leveraged borrower. They still affect opportunity cost and valuation. A project that looked attractive when the risk-free rate was low must clear a higher hurdle when long-dated government bonds offer more than 5%. Investors can demand more immediate evidence of returns rather than rewarding distant optionality.

This is one reason the Fed and Meta should not be treated as competing explanations. The Fed set the valuation environment; Meta supplied company-specific evidence about cash returns within that environment. Microsoft supplied a counterexample. Oil added inflation risk. The market decline reflected the interaction of all four.

Oil’s rally was a warning against reading every equity move through AI

Knight’s discussion of crude was useful because it widened the frame beyond technology. Markets rotate. Capital can leave one sector and enter another. A portfolio positioned only around AI may miss simultaneous moves in energy, industrials, financials, or defensives.

Oil’s July 29 advance was driven by both physical and geopolitical factors. Reuters reported that only a small number of commodity ships had transited the Strait of Hormuz during the week, while U.S. inventories fell substantially. Backwardation—when near-term futures trade above later contracts—can indicate that buyers are willing to pay a premium for immediate supply. It is not a perfect measure of shortage, but steep backwardation is generally consistent with a tight prompt market.

A persistent oil shock can reshape earnings across sectors. Producers may benefit from higher realized prices. Airlines, logistics companies, chemicals producers, and consumers face higher costs. Inflation expectations can rise, pressuring bonds and interest-sensitive equities. The effect depends on duration: a short spike has a different economic impact from a sustained disruption.

For the AI trade, the relevant lesson is that macro risk can invalidate a company-specific model. Meta might deliver excellent advertising growth and Microsoft might beat cloud expectations, yet valuations can still compress if long-term yields rise sharply. Conversely, an easing of geopolitical tensions and lower energy prices could support growth stocks even if earnings are merely adequate.

Technical traders often describe this as the need to respect price rather than force a single narrative onto the tape. Fundamental investors should make a similar distinction. The same stock decline can reflect weaker company economics, higher discount rates, risk reduction, positioning, or all of them at once.

The VIX was no longer near 15 by the close

Knight described the Cboe Volatility Index as being near 15 when explaining why he favored longer-dated index puts. That statement reflected the market level during the discussion, not the final reading after the day’s selloff and the Fed decision. By late July 29, Cboe data showed the VIX around 20.66, up roughly 13.5% from the previous close.

The distinction is important because option prices depend partly on implied volatility. The VIX is derived from S&P 500 option prices and represents the market’s estimate of expected volatility over approximately the next 30 days. It is not a forecast of direction, and a level of 20 does not mean the index will fall 20%. It indicates that option prices embed a wider expected range of outcomes than they do at a level of 15.

Buying puts when implied volatility is low can be attractive because the insurance is relatively inexpensive compared with periods of stress. Once volatility rises, the same protection generally costs more, all else equal. A trader who purchased before the jump may benefit from both a decline in the underlying index and an increase in implied volatility. A trader entering after the jump faces a higher premium and the risk that volatility falls even if the index does not rally much.

Volatility can also be uneven across maturities. Knight’s reference to January 2027 puts concerns options with much longer lives than the 30-day horizon summarized by the VIX. The relevant implied volatility is the level embedded in those specific contracts. The futures curve and term structure can flatten around major events as near-term uncertainty rises. That relationship affects the relative cost of short- and long-dated protection.

The change from roughly 15 during the segment to above 20 by the close strengthened the claim that market anxiety had increased. It also weakened the immediate “cheap insurance” argument for anyone who had not already entered. Timing matters even when the broader thesis remains unchanged.

What January 2027 index puts actually provide

A put option gives its holder the right, but not the obligation, to sell an underlying asset at a specified strike price before or at expiration, depending on the contract style. An index or ETF put can gain value when the underlying market falls, when implied volatility rises, or when both occur. The maximum loss for a long put is generally the premium paid, while the potential gain increases as the underlying falls below the strike, subject to contract terms and expiration.

Extending expiration to January 2027 buys time. A short-dated bearish thesis can fail even if the eventual direction is correct because the decline arrives after the option expires. A longer-dated contract reduces that specific timing risk. It also costs more because the seller is assuming risk for a longer period.

Time is not free. Options lose time value through a process commonly described by the Greek letter theta. The rate is not constant and generally accelerates as expiration approaches, though it varies by moneyness and volatility. A long-dated option may decay more slowly day to day than a near-term contract, but the buyer has committed more premium and can still lose money if the market remains stable, rises, or falls too little.

Strike selection changes the trade. An at-the-money put has greater immediate sensitivity to index moves than a far out-of-the-money put, but it costs more. A deep out-of-the-money put may provide inexpensive protection against a severe crash while losing most or all of its value in an ordinary correction. Without the strike and premium, “January 2027 puts” describes a maturity, not a complete strategy.

Position size is equally important. A put with a defined maximum loss can still be financially damaging if the premium represents too much of a portfolio. Repeatedly buying protection that expires unused can create a persistent drag. Conversely, too little protection may not materially offset losses elsewhere. The effectiveness of a hedge depends on the relationship between contract exposure and the assets being hedged.

Index puts also introduce basis risk. SPY tracks the S&P 500, QQQ tracks the Nasdaq-100, and XLK tracks an information-technology sector index. A portfolio concentrated in semiconductors may not move in the same proportion as SPY. A broad-market hedge may cushion a systemic decline but leave substantial sector-specific exposure. A direct SMH put is more targeted but can be more volatile and may carry different pricing.

Knight framed the longer maturity partly as an investment in his own ability to sleep. That is a legitimate risk-management consideration. A strategy that is theoretically optimal but psychologically unsustainable may lead to impulsive exits or oversized losses. The cost of time can be viewed as payment for flexibility. It should still be evaluated like any other expense: against the probability, magnitude, and timing of the risk being insured.

Why selling profitable puts can still be rational

Knight regretted closing SMH puts before the sector fell further. That reaction is familiar to traders, but hindsight can distort the evaluation of a decision. A profitable exit is not automatically wrong because a position later becomes more profitable. The proper question is whether the exit followed a sound process based on information and risk available at the time.

Options can reverse quickly. A position showing a large gain before earnings may lose value if the underlying rallies, implied volatility collapses, or the market interprets news differently than expected. Taking profit reduces exposure to those outcomes. It also sacrifices additional upside. No exit captures every dollar of a move.

Partial exits can balance these objectives. A trader may sell enough contracts to recover the initial premium and retain a smaller position. Another approach is to roll to a later expiration or a different strike, locking in some gain while maintaining bearish exposure. Each adjustment has transaction costs, tax implications, and new risk.

The emotional language in the segment is analytically useful because it reveals a hidden source of market risk: the trader’s own behavior. Regret can encourage revenge trading or larger re-entry at worse prices. Fear can produce premature exits. Overconfidence can turn a successful thesis into an oversized position. A written plan for entry, invalidation, profit-taking, and maximum loss is often more valuable than a perfect retrospective chart.

For readers, the responsible lesson is not that Knight should have held every put. It is that a good market thesis needs an exit framework before volatility arrives. Otherwise every outcome can feel like a mistake: selling too early, holding too long, hedging too little, or paying too much for protection.

The S&P 500 support test was also a concentration test

Knight contrasted the Nasdaq futures’ break with the S&P 500’s repeated defense of support. That difference can arise from index composition. The S&P 500 includes technology leaders but also financials, healthcare, industrials, energy, consumer companies, utilities, real estate, and materials. The Nasdaq-100 is more heavily exposed to large growth companies and excludes financial firms.

Yet the S&P 500 is float-adjusted market-cap weighted, so its largest companies have disproportionate influence. A sharp move in Meta or Microsoft can affect the broad index even though hundreds of other constituents are unchanged. This concentration means the S&P can look diversified by company count while remaining sensitive to a small group of mega-cap earnings reports.

The July 29 decline was not limited to one company. Technology fell sharply, but industrials also dropped more than 3%, and the Dow fell 2.19%. That breadth matters. A narrow technology pullback can be offset by rotation into banks, energy, or defensives. A broader selloff suggests that rates, growth expectations, or risk appetite are affecting multiple sectors.

Support levels are most informative when combined with breadth. If an index holds while fewer stocks remain above moving averages, the apparent resilience may depend on a handful of leaders. If the index falls but many constituents stabilize, the decline may be concentrated and closer to exhaustion. Advance-decline data, equal-weighted indexes, sector performance, credit spreads, and volume can add context to a single price line.

A close below a widely watched level can trigger systematic selling, stop orders, and dealer hedging. It can also attract buyers who view the move as temporary. The level is not magical. Its importance comes from the number of participants using it to organize risk.

Knight’s phrase that one bad earnings night could “seal the deal” described this reflexive dynamic. A negative report can push the index through support; the break itself can then generate additional selling. Meta’s after-hours decline moved in that direction. Microsoft’s rise worked against it. The next regular session would reveal which signal dominated once both reports were incorporated into cash-market trading.

What Knight’s bearish thesis got right

The strongest part of the thesis was its focus on capital intensity. By July 2026, the market had moved beyond asking whether AI adoption was real. It was asking who would earn an adequate return on an unprecedented infrastructure buildout. Meta’s results validated that concern. Revenue growth remained excellent, but costs rose faster, operating margin fell by 12 percentage points, and free cash flow nearly disappeared.

The thesis also correctly identified semiconductors as a high-beta expression of changing expectations. SMH’s 4.8% decline showed how quickly the supplier complex can reprice when investors question hyperscaler budgets or long-term demand. The sector’s valuations and exposure to a concentrated group of buyers make it vulnerable to even modest downward revisions.

Knight was also right that Meta had market-wide relevance beyond its own share price. Meta is one of the largest AI spenders, a major customer of semiconductor and data-center suppliers, and a bellwether for whether AI can improve a mature consumer platform. A cash-flow shock at that scale informs assumptions across the ecosystem.

His emphasis on flexibility also held up. Rather than relying on the Fed announcement alone, he watched how markets reacted and used defined technical levels. He extended maturities to reduce timing pressure and acknowledged the psychological cost of concentrated trades. Those are process strengths independent of whether the final market direction proves correct.

Where the thesis was too broad

The phrase “AI unraveling” risks collapsing heterogeneous businesses into one narrative. Microsoft’s results showed why that is dangerous. Azure grew faster than expected, cloud revenue expanded 27%, commercial backlog surged, and paid Copilot seats exceeded 30 million. That is evidence of monetization, not merely spending.

Meta’s own advertising metrics also weaken the most extreme version of the bear case. The company grew revenue 28%, impressions 14%, and average ad price 12%. AI may be contributing to those outcomes through recommendation and advertising tools. The problem is the cost and timing of the return, not an absence of business value.

The semiconductor argument can likewise become too uniform. Strong Microsoft demand and Meta’s maintained capex range continue to support orders. A share-price correction may reflect valuation compression rather than an imminent collapse in industry revenue. Suppliers with constrained capacity or dominant products can remain profitable even as their stock multiples decline.

Dismissing Microsoft as irrelevant also understates index mechanics. Microsoft is one of the largest companies in the S&P 500, Nasdaq-100, and XLK. A post-earnings rally can materially offset weakness elsewhere. Meta may carry the more dramatic message, but Microsoft carries substantial index weight and provides a fundamental benchmark for AI monetization.

Finally, a long-dated broad-market put is not a pure bet against hyperscalers. It is a bet on the path of the entire index. Strong earnings in healthcare, financials, industrials, or energy could support the S&P even if semiconductors weaken. Monetary easing, geopolitical de-escalation, or productivity gains could also lift the market. The instrument broadens the thesis beyond its original evidence.

The strongest bearish interpretation

The bearish case begins with overinvestment. Hyperscalers compete to secure chips, talent, land, and power because no company wants to fall behind. This strategic race can produce individually rational spending and collectively excessive capacity. Each management team assumes demand will justify its projects, but industry supply may grow faster than customers’ willingness to pay.

Meta’s quarter offers a preview. The core business grows rapidly, yet incremental expenses exceed incremental revenue and free cash flow falls 91%. If this pattern spreads, investors may stop valuing AI spending as an asset and begin valuing it as a claim on future cash. Lower free cash flow can reduce buybacks and increase dependence on debt, leases, or external partners.

Higher interest rates intensify the problem. The Fed has not declared victory over inflation, energy prices are rising, and long-dated Treasury yields exceed 5%. A higher discount rate reduces the present value of profits expected years from now. It also raises the hurdle rate for infrastructure projects. Even successful AI products may not earn enough to justify the capital committed at peak-cycle prices.

Semiconductor suppliers could then face a classic inventory and order correction. Hyperscalers may continue taking delivery on existing commitments while slowing new orders. Revenue can remain strong for several quarters even as stocks decline in anticipation. Equipment companies and memory suppliers can be especially sensitive to changes in utilization and capacity plans.

In this scenario, Meta’s result is not an isolated miss. It is the first highly visible sign that the AI capital cycle has entered an accountability phase. The market narrows its support to companies with clear, near-term monetization and punishes those asking for patience. Multiples contract across the sector, and the broad indexes fall because mega-cap weights are so large.

The strongest bullish interpretation

The bullish case begins with demand rather than spending. Microsoft’s Azure growth, $678 billion commercial obligation balance, and expanding paid Copilot seats indicate that enterprises are committing real budgets. Meta’s advertising revenue shows that AI-enhanced recommendations and ad systems can improve an established business at enormous scale.

Infrastructure is built before revenue can be served. Capacity constraints have limited cloud growth, so elevated capex may be a response to verified demand rather than speculative overbuilding. New facilities can take years to permit, construct, connect to power, and equip. A company that waits for complete certainty may permanently lose customers and developer ecosystems.

Meta’s free-cash-flow decline may therefore represent a trough during a front-loaded investment period. Its balance sheet and advertising cash engine allow it to invest when smaller competitors cannot. If the spending produces better engagement, more effective ads, valuable assistants, and new computing interfaces, future revenue may scale across an installed base of billions without a proportional increase in distribution costs.

Rapid technological progress can also lower unit costs. New chips and optimized models may perform more inference per watt and per dollar. Software improvements can reduce the computation required for a task. Higher utilization can spread fixed costs over more revenue. Today’s expensive workload may become economically attractive as the stack matures.

In this scenario, market weakness is a valuation reset within a durable growth cycle, not the beginning of an industry collapse. The companies with the strongest distribution and balance sheets emerge with wider competitive moats. Semiconductor demand remains structurally high even if annual growth becomes less explosive.

Five questions that will decide which interpretation wins

1. Does AI-related revenue grow faster than depreciation and operating costs?

Revenue growth alone is insufficient. Investors need evidence that gross profit and operating cash flow expand faster than the cost of running the installed infrastructure. Rising depreciation will be a central test from 2026 onward.

2. Do utilization rates remain high as capacity arrives?

Capacity shortages support pricing and margins. Excess capacity weakens both. Companies disclose limited utilization data, so investors will infer conditions from cloud growth, lead times, pricing, supplier orders, and management commentary.

3. Will enterprises renew paid AI products after initial deployments?

Thirty million paid Copilot seats demonstrate adoption. Renewal rates, usage, expansion, and measurable productivity determine durability. A purchased license that is rarely used has a different economic value from a product embedded in daily work.

4. Can Meta convert consumer AI engagement into incremental cash?

Better recommendations already support the ad business, but the company’s larger infrastructure program likely assumes more. Investors will look for evidence from ad efficiency, messaging commerce, assistants, creator tools, and any new services that management chooses to monetize.

5. Does the macro environment permit patient capital?

A lower-rate environment gives long-duration projects more time. Persistent inflation, high energy costs, and rising bond yields shorten the market’s patience. The Fed’s next decisions may therefore matter more to the valuation of AI spending than to the operational demand for AI itself.

A concise timeline of the week’s market setup

• July 22, 2026: Reuters analysis highlighted growing pressure on big technology free cash flow as AI capital spending expanded faster than current cash generation.
• July 23, 2026: Alphabet’s spending and cash-flow profile added to concern that hyperscalers were entering a more capital-intensive phase.
• July 28, 2026: Reuters reported Meta’s approximately $14 billion El Paso data-center partnership with BlackRock, illustrating the scale and evolving financing of AI infrastructure.
• July 29, 2026, regular session: U.S. equities fell sharply. The S&P 500 lost 1.52%, the Nasdaq Composite 1.74%, and the Dow 2.19%. SMH fell approximately 4.8%.
• July 29, 2026, 2 p.m. ET: The Federal Reserve held the federal-funds target at 3.50% to 3.75% in a 9–3 vote. Three policymakers preferred a quarter-point increase.
• July 29, 2026, after the close: Meta reported 28% revenue growth but a 91% decline in free cash flow and raised the lower ends of spending expectations. Shares fell about 5% in extended trading.
• July 29, 2026, after the close: Microsoft reported 43% Azure growth, stronger-than-expected cloud demand, and more than 30 million paid Copilot seats. Shares rose about 3% in extended trading.

The chronology matters because the market was already weak before either company released earnings. Meta intensified a bearish narrative; it did not create the entire decline. Microsoft simultaneously provided evidence against the broadest version of that narrative.

What happens next

The first test is the next regular trading session, when cash-market investors can process both earnings reports and the Federal Reserve decision together. After-hours prices can change substantially once conference-call details, analyst models, and broader liquidity enter the market.

Meta’s earnings call and subsequent filings will be important for the composition of spending, the expected trajectory of depreciation, the timing of new capacity, and management’s explanation of returns. Investors will also examine whether legal charges or other temporary items explain enough of the margin decline to make the underlying trend less severe.

Microsoft’s call will be evaluated for fiscal 2027 guidance, capacity constraints, cloud margins, capital spending, and the conversion of backlog into revenue. The market will seek more detail on paid Copilot adoption and the degree to which AI contributes to Azure growth.

The Fed’s September meeting has become a larger risk after the 9–3 decision. Incoming inflation, labor-market, and energy data will shape whether policymakers move from a hold to a hike. Any sustained disruption to oil shipping could influence that path.

For semiconductors, company guidance and hyperscaler purchase commitments will matter more than one ETF chart. Investors will watch order growth, advanced packaging capacity, memory pricing, equipment lead times, and whether customers maintain 2027 plans.

For the broad market, the technical question is whether the S&P 500 can reclaim broken levels and whether breadth improves. The fundamental question is whether earnings outside mega-cap technology can offset higher discount rates and concentrated AI-spending risk.

Why the next phase will be judged by return on invested capital

Free cash flow is an important warning signal, but it is not the final measure of whether the AI buildout creates value. A company can spend heavily today and still make an excellent investment if the assets generate sufficiently large cash flows over many years. The broader measure is return on invested capital: the after-tax operating profit produced relative to the capital required to produce it.

That framework changes the debate from “capex is high” to “what return will the capex earn?” A $10 billion project that produces $2 billion of durable annual after-tax operating profit may be attractive. The same project producing $300 million may destroy value even though it adds revenue. The answer depends on useful life, maintenance needs, financing cost, taxes, and the risk that technology or customer demand changes before the investment is recovered.

For Meta, the calculation begins with the advertising franchise. If better models increase engagement, conversion, and advertiser spending, the incremental profit can be substantial because the platform already reaches billions of people. But the denominator is expanding quickly as Meta commits to data centers, chips, power, and related assets. Revenue gains that would once have produced operating leverage may now be absorbed by depreciation, energy, technical staff, and infrastructure operations.

Microsoft’s position is different because it can charge external customers for cloud consumption and sell AI software on top of the same infrastructure. This provides more visible revenue attribution, but the company also assumes the cost of serving demanding workloads and must compete on price and performance. A large backlog is encouraging only if contracts convert at margins high enough to compensate for the capital needed to fulfill them.

Return on invested capital cannot be observed cleanly in real time. New assets may be underutilized while campuses ramp. Revenue may be recognized over several years. Some investment protects an existing franchise rather than creating an identifiable new product. Shared infrastructure serves both AI and conventional workloads. Management teams also have discretion over allocation methods and asset lives. The calculation will therefore remain approximate from outside the companies.

Several indicators can still improve the estimate. Investors can compare the growth of operating profit with the growth of property and equipment, monitor depreciation as a share of revenue, examine free cash flow after lease payments, and track whether cloud or advertising margins stabilize. They can also compare capital spending with revenue growth over rolling multiyear periods rather than judging one quarter in isolation.

The greatest risk is a permanently lower cash-return profile. A company may continue growing revenue while becoming structurally more capital intensive. That does not make the business bad, but it can justify a lower valuation multiple. Investors historically paid premium multiples for software and digital advertising partly because incremental revenue required relatively little physical capital. If AI shifts those businesses toward the economics of infrastructure, the valuation framework may need to change even when growth remains strong.

Three plausible market paths from here

Scenario one: selective consolidation

In the most balanced scenario, AI demand remains strong but investors distinguish sharply among business models. Microsoft and other companies with measurable cloud or subscription revenue retain support. Meta and other aggressive spenders face pressure until cash conversion improves. Semiconductor leaders remain profitable, but their valuations compress as growth normalizes. The broad market experiences volatility without a deep recessionary decline.

This path would be consistent with the July 29 evidence. It accepts both Meta’s cash-flow warning and Microsoft’s demand strength. Capital does not abandon AI; it moves toward companies able to demonstrate pricing, utilization, and recurring revenue.

Scenario two: a capital-cycle correction

In the bearish scenario, customer adoption remains real but fails to match the capacity being built. Cloud providers begin to report lower utilization or weaker incremental bookings. Hyperscalers delay projects, renegotiate supplier commitments, or rely more heavily on financing structures. Semiconductor orders slow, data-center credit becomes more expensive, and depreciation continues rising after capex peaks.

Under this path, share prices can fall well before reported revenue contracts. Markets anticipate the margin pressure and assign lower multiples to the entire chain. A broad-market hedge performs better because mega-cap concentration transmits the correction to the S&P 500 and Nasdaq-100.

Scenario three: productivity-led acceleration

In the bullish scenario, enterprise deployments move rapidly from experimentation to production. AI products reduce labor time, increase sales conversion, improve software development, and create new services. Paid seats expand, usage deepens, and infrastructure utilization remains high. Model and chip efficiency lower unit costs while new capacity removes supply constraints.

Meta’s current cash-flow trough then looks like front-loaded investment, and Microsoft’s quarter becomes an early indicator of broader demand. Semiconductor revenue remains elevated, operating leverage returns as growth outruns depreciation, and higher productivity eventually helps offset inflationary pressure.

No single quarter can choose among these outcomes. The value of the July 29 reports is that they established measurable starting points. Meta must show that spending can restore cash growth. Microsoft must show that contracts and seats become durable high-margin revenue. The Fed must determine whether inflation permits a rate environment in which long-duration investment can be valued patiently.

How readers can interpret future earnings without chasing every headline

The next wave of AI earnings will contain large numbers and dramatic after-hours moves. A disciplined reading can focus on a small set of linked questions.

• Demand: Are customers paying, renewing, expanding usage, and signing long-term commitments?
• Capacity: Is growth limited by unavailable infrastructure, or is new capacity arriving faster than demand?
• Unit economics: Are prices and gross profit sufficient to cover inference, power, depreciation, support, and sales costs?
• Cash conversion: Does operating cash flow grow after capital expenditures and lease obligations?
• Asset productivity: Is revenue growing faster than the installed capital base over a multiyear period?
• Financing: Are projects funded from internal cash, debt, leases, joint ventures, or commitments that create future fixed payments?
• Valuation: How much success is already embedded in the share price?

These questions prevent a common analytical error: treating a revenue beat as proof of excellent economics or a capex increase as proof of waste. Both can be true or false depending on the return earned. The market’s challenge after July 29 is not to choose between enthusiasm and skepticism as identities. It is to demand evidence at each stage of the cash-flow chain.

The information investors still do not have

Both companies disclosed enough to move markets, but not enough to calculate a definitive AI return. Meta does not separately report the revenue, operating profit, or cash flow attributable to its newest models, recommendation systems, assistants, or generative advertising tools. Microsoft reports cloud and product growth, but it does not provide a complete bridge from AI consumption to revenue, gross margin, capital employed, and cash return. The missing detail is understandable because infrastructure is shared, products are bundled, and competitive information is sensitive. It nonetheless limits the certainty of outside analysis.

Investors also lack standardized definitions. One company may describe a workload as AI-related because it uses machine learning; another may reserve the label for generative models. Capital expenditure can include land, buildings, conventional cloud servers, accelerators, networking, and offices. Remaining performance obligations can include contracts with different durations and cancellation terms. Paid seats can carry different prices, discounts, and usage levels. Comparing headline figures without these differences can create false precision.

The best available approach is triangulation. Financial statements show cash, profit, assets, and obligations. Product metrics show adoption. Supplier results show orders and capacity. Customer surveys show perceived value. Power and construction data show the physical pace of deployment. No source is sufficient alone. A credible conclusion should become stronger only when several independent indicators point in the same direction.

This limitation also argues against extreme language. Meta’s quarter cannot prove that every AI dollar is wasted, and Microsoft’s beat cannot prove that the industry will earn exceptional returns. The results establish contrasting evidence within one investment cycle. The responsible interpretation remains conditional, updated as utilization, pricing, renewals, margins, and free cash flow become visible.

Frequently asked questions

Why did Meta stock fall after its second-quarter 2026 earnings?

Meta’s revenue grew 28% to $60.80 billion, but costs rose 55%, operating margin fell to 31% from 43%, and free cash flow dropped 91% to $784 million. The company also raised the lower ends of its 2026 expense and capital-spending ranges. Shares fell about 5% in extended trading as investors focused on the cash cost of the AI buildout rather than the strong advertising growth alone.

Did Meta miss revenue expectations?

The central problem was not a collapse in revenue. Meta’s quarterly revenue landed near the top of its prior $58 billion to $61 billion guidance range. Advertising remained strong, with impressions up 14% and average price per ad up 12%. The concern was that expenses and capital investment grew so rapidly that operating profit, margin, net income, earnings per share, and free cash flow all weakened despite the sales increase.

How can Meta report billions in profit but only $784 million in free cash flow?

Net income and free cash flow measure different things. Capital expenditures for servers, buildings, and equipment are generally capitalized and depreciated over time for income-statement purposes, but the cash is paid when assets are purchased or constructed. Meta therefore remained highly profitable under accounting rules while its current infrastructure payments consumed most of the cash generated by operations. Lease-related payments and working-capital movements can also affect cash conversion.

What is Meta’s 2026 capital-expenditure guidance?

Meta narrowed its expected 2026 capital expenditures to $130 billion to $145 billion, compared with an earlier range of $125 billion to $145 billion. Raising the lower end indicated greater confidence that spending would remain elevated. The range covers the company’s broad infrastructure program, not a single product or facility.

Why did Microsoft stock rise after earnings while Meta fell?

Microsoft gave investors more direct evidence that infrastructure spending was translating into paid demand. Azure and other cloud-services revenue grew 43%, above the roughly 40% Visible Alpha consensus cited by Reuters. Microsoft Cloud revenue reached $59.3 billion, commercial remaining performance obligation rose to $678 billion, and paid Microsoft 365 Copilot seats exceeded 30 million. Those results did not eliminate spending risk, but they strengthened the monetization case.

Was Microsoft’s earnings growth entirely operational?

No. Microsoft reported strong operating growth, but GAAP net income also benefited from investment-related items. The company disclosed a $3.2 billion gain related to Anthropic and other discrete items that provided a net benefit of about $0.27 per share. Microsoft also presented non-GAAP figures excluding the effect of OpenAI investments. Investors should separate cloud and software performance from volatile investment marks.

What did the Federal Reserve decide on July 29, 2026?

The Federal Open Market Committee held the federal-funds target range at 3.50% to 3.75% in a 9–3 vote. Beth Hammack, Neel Kashkari, and Lorie Logan preferred to raise the range by 25 basis points. The statement described inflation as elevated and highlighted energy-supply risks. The split vote and Chair Kevin Warsh’s inflation emphasis made the hold more hawkish than a routine pause.

Why are higher interest rates negative for AI stocks?

Higher rates increase the return investors can earn on lower-risk assets and raise the discount rate used to value future profits. AI infrastructure projects also face a higher required return when financing and opportunity costs rise. Companies with strong current cash flow can still invest, but markets tend to demand clearer and faster evidence that long-dated projects will earn more than the higher hurdle rate.

What are XLK, QQQ, and SMH?

XLK is the State Street Technology Select Sector SPDR ETF, which tracks the information-technology companies in the S&P 500. QQQ is the Invesco ETF tracking the Nasdaq-100, a broader large-cap growth index that includes Meta as well as major technology companies. SMH is VanEck’s semiconductor ETF, focused on chip designers, manufacturers, and equipment companies. Their holdings and sensitivities differ, so they should not be treated as interchangeable “tech” instruments.

Does a bearish chart pattern guarantee that a stock or ETF will fall?

No. A breakdown can help define a trend, entry, stop, or invalidation level, but it does not guarantee an outcome. Earnings surprises, policy changes, geopolitical events, and changes in positioning can reverse price quickly. False breakouts and breakdowns are common. Technical analysis is most useful as a conditional risk framework rather than a promise about direction.

Are January 2027 puts a safe way to profit from a market decline?

No option strategy is guaranteed or risk-free. A long put has a defined maximum loss equal to the premium paid, but the buyer can lose the entire premium if the market does not decline enough before expiration. Long-dated options cost more, are affected by implied volatility and time decay, and may not match the exposure being hedged. The strike, premium, position size, and underlying instrument determine the actual risk.

Does Meta’s quarter prove that the AI investment boom is ending?

No. It proves that the cash cost and margin consequences are becoming more visible. Meta’s advertising growth remained strong, Microsoft’s cloud demand exceeded expectations, and both companies continued to invest. A genuine industry reversal would require broader evidence such as budget cuts, weaker supplier orders, slower cloud growth after capacity expands, low renewal rates, or persistent deterioration in returns on invested capital.

Final assessment: Meta validated the concern, Microsoft prevented a clean verdict

Tim Knight’s central insight was that Meta’s report could shape the market’s judgment of the AI buildout more than the day’s scheduled Fed drama. The after-hours reaction supported that view. Meta delivered one of the strongest revenue-growth rates among the world’s largest companies, yet its operating economics weakened and free cash flow fell to $784 million. The result placed a concrete number on the cost of competing at the frontier.

The bearish argument is strongest when it focuses on that mismatch. Capital spending is accelerating before outside investors can measure the full return. Depreciation and operating costs will persist after facilities are built. Higher oil prices and a more hawkish Federal Reserve raise the discount rate and shorten the market’s patience. Semiconductor shares, already priced for exceptional demand, are vulnerable to any reduction in long-term expectations.

But the evidence does not support treating the entire AI complex as one failing trade. Microsoft’s 43% Azure growth, $678 billion commercial obligation balance, and more than 30 million paid Copilot seats show that customers are committing substantial budgets. Meta’s own advertising results show that AI-enhanced systems can support engagement and monetization even when the cash payback remains difficult to isolate.

The most accurate conclusion is selective rather than absolute. The AI cycle has entered a phase in which distribution, pricing, utilization, and financing matter more than announcements. Companies with direct, recurring revenue and strong customer commitments have a more defensible case. Companies asking shareholders to fund larger infrastructure programs without equally visible cash returns will face more skepticism.

Knight’s technical warnings may prove prescient if broad indexes remain below support, semiconductor orders soften, and hyperscaler cash conversion deteriorates across several quarters. They will be weakened if cloud demand accelerates, paid AI adoption renews, and Meta’s spending begins to lift operating cash flow faster than depreciation and lease costs.

The July 29 reports did not mark the confirmed end of the AI boom. They marked the end of a simpler market story. Spending alone is no longer evidence of leadership. Revenue growth alone is no longer enough. The next stage will be decided by which companies can turn vast computing capacity into durable free cash flow before the cost of capital, competition, and technological obsolescence catch up.

Sources

• Federal Reserve monetary policy statement, July 29, 2026
• Reuters analysis of the July 2026 Federal Reserve decision
• Reuters report on the July 29 U.S. market close and after-hours earnings reactions
• Meta second-quarter 2026 earnings release
• Reuters analysis of Meta’s second-quarter results and capital spending
• Meta first-quarter 2026 earnings release
• Microsoft fiscal 2026 fourth-quarter earnings release
• Reuters analysis of Microsoft’s fiscal fourth-quarter results
• Reuters analysis of hyperscaler capital spending and free cash flow
• Reuters report on Meta and BlackRock’s El Paso data-center partnership
• Reuters report on the July 29 crude-oil rally and U.S. inventories
• Cboe Volatility Index data and methodology
• State Street Technology Select Sector SPDR ETF information
• Invesco QQQ ETF information
• VanEck Semiconductor ETF information
• S&P Dow Jones Indices methodology and S&P 500 information

This article is provided for general informational purposes and does not constitute financial, investment, tax, or legal advice.

Affiliate disclosure: Businessfinance.news may earn compensation from qualifying actions completed through selected links on this website, at no additional cost to the reader. Affiliate relationships do not influence our editorial reporting, analysis, or conclusions.