Inside the July 29 Stock Market Selloff: A Divided Fed, a Bond Market Revolt and an AI Trade Cracking at the Edges

Last updated: July 30, 2026, 6:00 a.m. Eastern Time

The Federal Reserve did nothing on Wednesday, and that turned out to be the most consequential thing it could have done. At 2:00 p.m. Eastern on July 29, 2026, the Federal Open Market Committee voted 9–3 to leave the federal funds target range at 3.50% to 3.75% for a fifth consecutive meeting. Three regional Reserve Bank presidents dissented — not because they wanted easier policy, but because they wanted rates higher. Within two hours, the 30-year Treasury yield had touched 5.21%, a level it had not seen since 2007, and the Dow Jones Industrial Average had given up an intraday rally and closed 1,153.18 points lower.

That is the short answer to the question most investors were typing into search boxes Wednesday night: the market fell because the bond market decided the Fed is behind on inflation, and equities were repriced accordingly. But it is not the complete answer, and the incomplete version misses what made this session genuinely unusual. Three separate stress points converged inside a six-hour window. A central bank under a new chairman deliberately withheld guidance about what comes next. Iran fired ballistic missiles toward American forces in Jordan, sending crude up nearly 8% after its steepest three-day collapse since 2020. And underneath both, the trade that has carried this market for two years — artificial-intelligence infrastructure — kept unwinding, with the Philadelphia Semiconductor Index on pace for its worst month since 2008.

Any one of those would have made for a difficult day. Together they produced the Dow’s worst session since April 2025, a Nasdaq-100 that slipped into correction territory, and a closing hour in which the S&P 500 undercut its own morning lows on accelerating volume. Then, after the bell, Microsoft and Meta Platforms reported quarterly results that split the market’s view of AI spending cleanly down the middle: one company was rewarded for showing what the money bought, and the other was punished for showing what the money cost.

What follows is a reconstruction of that session and an examination of what the numbers actually support — drawn from the Fed’s own statement and press-conference transcript, company earnings releases, Treasury yield data, and the market commentary that shaped how the day was read in real time. Where the widely repeated version of events is imprecise, this article says so.

Key Takeaways

• Main development: The FOMC held its target range at 3.50%–3.75% on July 29, 2026 by a 9–3 vote, with Cleveland’s Beth Hammack, Minneapolis’s Neel Kashkari and Dallas’s Lorie Logan all dissenting in favor of a quarter-point increase.
• Key figure: The 30-year Treasury yield rose to 5.21%, its highest since 2007, after climbing as much as roughly 14 basis points intraday; the 10-year finished near 4.70%, up from 4.61% on July 28.
• Market response: At the July 29 close, the Dow fell 1,153.18 points (2.19%) to 51,594.14, the S&P 500 lost 1.52% to 7,316.15 and the Nasdaq Composite dropped 1.74% to 24,442.94. Industrials were the worst sector, down 3.42%.
• The second shock: Brent crude jumped 7.9% to settle at $90.74 and West Texas Intermediate rose 6.6% to $84.46 after Iranian ballistic missiles were fired toward U.S. positions in Jordan and U.S. and Saudi forces struck Iran-aligned groups in Iraq.
• Why it matters: Chair Kevin Warsh has removed forward guidance from the Fed’s communications on purpose. Wednesday was the first serious test of what happens when a market accustomed to being told what comes next has to price policy on its own.
• What comes next: June personal consumption expenditures inflation, the advance estimate of second-quarter GDP and weekly jobless claims are all scheduled for 8:30 a.m. Eastern on July 30. Amazon and Apple report after Thursday’s close. The next FOMC meeting is September 15–16.

What the Fed actually decided — and what it pointedly refused to say

The FOMC statement released on July 29 runs to four short paragraphs. That is not an accident, and it is the single most important thing to understand about how markets behaved for the rest of the afternoon.

Under Kevin Warsh, who was presiding over only his second FOMC gathering — the previous one, as he noted, had concluded 42 days earlier — the Fed has stripped its post-meeting communication back to something closer to a bulletin than a policy essay. The July statement noted that economic activity “is expanding at a solid pace despite elevated uncertainty that owes, in part, to the conflict in the Middle East,” that “productivity growth and capital investment are strong,” and that job gains “have kept pace with the workforce.” On prices, it said inflation “remains elevated relative to the Committee’s 2 percent goal, in part reflecting supply shocks that have driven price increases in certain sectors, including energy,” and then delivered a single declarative line: “The Committee will deliver price stability.”

There is no sentence about what the Committee expects to do at its next meeting. There is no reference to being data-dependent, to assessing incoming information, or to the balance of risks. There is no calibrated hint. For a generation of traders whose models were built partly on parsing changes to Fed adjectives, the absence is the message.

The dissents were the other half of the story. Voting against the action were Beth M. Hammack, president of the Federal Reserve Bank of Cleveland; Neel Kashkari, president of the Minneapolis Fed; and Lorie K. Logan, president of the Dallas Fed. All three preferred to raise the target range by a quarter percentage point at this meeting. Three dissents in a single direction is rare in the modern Fed. Three dissents demanding tighter policy, at a moment when a sitting president has been publicly pressuring the central bank to cut, is rarer still.

The context the statement left out

Inflation has run above the Fed’s 2% target for more than five years, a stretch that began when the post-pandemic reopening overheated the U.S. economy in early 2021. The rate peaked above 9% in mid-2022, fell substantially in response to eleven rate increases across 2022 and 2023, and then stopped falling. Progress since has been grudging and, over the past several months, has been complicated by an energy shock that has nothing to do with domestic demand.

That is the bind the three dissenters were responding to. Rate increases do not manufacture crude oil. But a central bank that watches five years of above-target inflation get topped up by an energy shock risks discovering that households and businesses have quietly revised their expectations of what “normal” inflation looks like — and expectations, unlike oil supply, are something monetary policy can affect.

Christopher Waller, a member of the Fed’s Board of Governors, put the argument in unusually blunt terms in a speech earlier in July: “Sternly staring at inflation until it melts before our withering gaze is not an option.” That line circulated widely among rates traders in the run-up to the meeting, and it framed the question that dominated Wednesday’s press conference.

Warsh’s experiment: what happens when the Fed stops narrating

The preliminary transcript of Chairman Warsh’s opening statement makes clear that the sparse communication style is a deliberate program rather than the awkwardness of a new arrival still learning the job.

“As before, the policy statement conveys just the facts,” Warsh said. “It’s steering clear of forecasting, a choice we consider especially prudent at these uncertain times. Uncertainty, however, does not mean a lack of clarity.”

He then addressed something that has become the central credibility question of his early tenure — whether five years of overshoot has taught the public that the Fed’s real target is higher than the one it advertises. “For some households, businesses, and market professionals, five years of high inflation have left a mistaken impression that is hard to shake: that the Fed’s implicit inflation target was somehow above 2 percent,” he said. “Let me reiterate: There is no soft inflation target, there is no soft implicit target — not on this Committee’s watch. There is only a target, and it is 2 percent.”

The most revealing passage came when Warsh turned to the bond market’s behavior since the June meeting. He noted that “nominal and real yields are materially higher across the Treasury curve” and that some of the increases in market interest rates between the two meetings rank “around the top decile or so” of intermeeting moves over the past two decades. Rather than treating that as a problem, he presented it as evidence the new approach is working.

“Market participants are learning to play the ball, not the referee — and market prices will continue to respond in the direction and magnitude they see fit. This is, in my view, a change for the better — and we are just getting started. After all, the central bank need not always and everywhere be the center of attention.”

That is a coherent philosophy, and it has intellectual pedigree. A central bank that constantly telegraphs its next move invites markets to price the central bank rather than the economy, which can create a feedback loop where the Fed ends up validating expectations it created. Warsh is arguing that if investors are forced to form their own views about inflation and growth, asset prices will carry more information and less reflexivity.

The complication is what happened next. Within about two hours of that statement, long-dated Treasury yields were at nineteen-year highs, the dollar was weaker, market-based measures of inflation expectations had risen, and the S&P 500 had closed at its low of the session. If the intended message was that the Committee is resolute about 2%, the received message appears to have been closer to the opposite.

The question Warsh could not put away

Reporters in the room repeatedly pressed the same point: if the Committee’s own assessment is that inflation remains elevated, and if three voting members thought the evidence justified acting now, why not act? Warsh’s answers did not resolve the tension, and the press conference is where the intraday rally died.

It is worth being precise about what is and is not knowable here. Warsh did not say the Committee would raise rates in September. He did not rule it out. He did not offer a reaction function, a threshold, or a set of conditions. Interpretations of his reluctance — that he is privately dovish, that he wanted to avoid a market shock, that he is holding fire until he has more data, that he is preserving optionality — are inferences, not reporting. The transcript supports none of them individually and all of them equally, which is precisely the ambiguity the market had to price.

What can be said with confidence is that a hold plus three hawkish dissents plus no guidance is a combination markets found harder to digest than any of those elements alone. A hold with dovish framing would have rallied bonds. A hike would have been a clean hawkish signal. A hold with three hawkish dissents and no roadmap left investors to conclude that a September increase is likely, that the Committee is internally divided about the timing, and that no one at the Fed is going to help them narrow the range of outcomes in the meantime.

The bond market’s verdict: a nineteen-year high in the long end

The clearest read on Wednesday came from Treasuries, and the shape of the move matters as much as its size.

According to the Federal Reserve’s H.15 daily interest rate release, the constant-maturity Treasury curve closed Tuesday, July 28, with the 2-year at 4.26%, the 10-year at 4.61%, the 20-year at 5.11% and the 30-year at 5.09%. By Wednesday afternoon the 30-year had pushed to 5.21% — at one point rising as much as roughly 14 basis points to nearly 5.23% — while the 10-year finished around 4.70%. The short end did not participate. Two-year yields drifted lower during the press conference even as the long end sold off.

That combination has a name: bear steepening. When yields rise and the curve steepens simultaneously, it usually means investors are demanding more compensation for holding duration — more inflation risk, more term premium, more uncertainty about the path of policy over a decade or three — rather than repricing the immediate policy rate. A market that simply expected a September hike would push up 2-year yields. A market that suspects the Fed will get to 2% late, if at all, pushes up 30-year yields.

A useful anchor: on July 28, the 10-year Treasury inflation-protected security yielded 2.41% against a 4.61% nominal 10-year, implying a breakeven inflation rate of roughly 2.20 percentage points over the coming decade. The 30-year TIPS yield was 2.92%. Those real yields are historically high. Real 30-year money near 3% is a meaningfully restrictive discount rate for any asset whose value depends on cash flows arriving in the 2030s — which describes, with unfortunate precision, most of the AI infrastructure complex.

A brief translation for readers who do not live in the rates market: a basis point is one-hundredth of a percentage point, so a 14 basis point move in the 30-year is a rise of 0.14 percentage points. Bond prices move inversely to yields, which is why a rising-yield day is described as a selloff even though nothing about the government’s ability to pay changed. And a yield is not a coupon — the 30-year bond issued years ago at a much lower coupon still pays that coupon; what changed Wednesday is the price investors will pay for it.

The practical consequence extends well beyond Wall Street. The 30-year Treasury yield is the anchor for long-term borrowing costs across the economy. The bank prime loan rate stood at 6.75% and the Fed’s discount window primary credit rate at 3.75% as of July 28. Mortgage rates, corporate bond issuance costs and the discount rate applied to every long-duration equity in the market all take their cue from the long end. A 5.2% 30-year is not a market technicality. It is a repricing of the cost of the future.

Why the intraday rally failed

Equities did not fall in a straight line. The initial reaction to the 2:00 p.m. statement was constructive; the Nasdaq briefly turned positive and was up roughly half a percent at one point after the release. The reversal came during the question-and-answer session, and it accelerated into the final hour.

The mechanics are worth spelling out because they explain the psychology of the close. If a market rallies off a headline and then trades back below the low it made before that headline, everyone who bought the initial move is offside. Forced selling from those positions tends to arrive late in the session, which is exactly the pattern the tape showed: the S&P 500 undercut its early-session lows and finished at 7,316.15, essentially on the day’s low, against an intraday high of 7,450.84. That is a 134-point round trip in a single afternoon.

Volume told the same story. Sessions where indexes close near their lows on higher volume than the prior day are, in the vocabulary of institutional flow analysis, distribution days — evidence that large holders are reducing exposure rather than individuals trimming positions. Whether one uses that framework or not, the observable fact is that the last thirty minutes produced the sharpest selling of the day, and the long end of the Treasury curve spiked in the final minutes alongside it.

The shock that arrived before the Fed: Iran, Jordan and a 7% jump in crude

Every account of Wednesday that begins at 2:00 p.m. is starting the story late. The session opened badly, and it opened badly because of the Middle East.

Early Wednesday, Iran’s Islamic Revolutionary Guard Corps said it had fired ballistic missiles at a U.S. airbase and a U.S. Central Command facility in Jordan. Jordanian air defenses intercepted the incoming missiles. The launches came hours after the U.S. military said it had knocked down a separate Iranian barrage aimed at American forces elsewhere in the region, and they were accompanied by joint U.S. and Saudi strikes on Iran-aligned groups in Iraq. The exchange ended a brief pause in fighting that markets had begun, tentatively, to price as the start of de-escalation.

Oil responded violently, and the violence was amplified by positioning. Crude had fallen roughly 16% over the preceding three sessions — the steepest three-day drop since 2020 — as traders bet on diplomacy. Wednesday reversed a meaningful chunk of it. Brent crude futures rose 7.9% to settle at $90.74 a barrel. West Texas Intermediate advanced 6.6% to settle at $84.46. Less than a week earlier, Brent had briefly traded above $100.

The underlying supply situation explains why headlines move this market so violently. Following U.S. and Israeli strikes on February 28, Iran closed the Strait of Hormuz, the channel through which roughly a fifth of the world’s seaborne oil and natural gas passes. That produced what has been widely described as the largest disruption to oil supply in history. Prices have oscillated with the state of the conflict ever since, but the baseline has shifted: a barrel of crude now costs roughly $10 to $15 more than it did at the same point a year earlier. Iranian-backed Houthi forces in Yemen have separately been attacking shipping in the Red Sea, targeting tankers carrying Saudi crude through the Bab el-Mandeb Strait.

For the Fed, this is the awkward part. Monetary policy is a demand instrument, and an energy shock is a supply problem. Raising rates does not reopen Hormuz. But the FOMC’s own statement acknowledged that inflation remains elevated “in part reflecting supply shocks that have driven price increases in certain sectors, including energy” — and then said the Committee will deliver price stability anyway. Those two sentences, side by side, describe the corner the Committee is in.

Carl Weinberg, chief economist at High Frequency Economics, framed the choice sharply in a note ahead of the meeting: “Sure, it is possible that the latest rise in prices is a transient blip that will reverse in a heartbeat. Then again, it seems equally that the war with Iran will get worse, that the Strait of Hormuz and Bab al-Mandab will remain blockaded for months or longer, and that energy prices will continue to trend up.” His question — whether a central bank should set policy on the hope that oil reverses, or act to minimize the chance inflation stays above target — is the question the three dissenters answered differently from the nine.

Where the damage landed

The index-level numbers understate how uneven the session was.

The Dow’s 2.19% decline was worse than either broad index, which is unusual and diagnostic. Blue-chip industrials do not normally lead a tech-driven selloff. On Wednesday they did, and the reason was not the Fed.

Caterpillar fell 6.91%, the index’s worst performer, after Baird downgraded the stock to Neutral from Outperform. The analyst’s argument had nothing to do with interest rates: a growing wave of state and local restrictions on data center construction, the note argued, could slow demand for Caterpillar’s high-margin power generation business starting in 2027, after an expected near-term boom. That is worth pausing on. The downgrade was a bet against the durability of AI infrastructure spending, delivered through a 100-year-old machinery company. Deere dropped 4.52% on falling agricultural commodity prices and cautious guidance.

What made the industrial weakness sting was the timing. Just a day earlier, Sherwin-Williams and Boeing had delivered well-received earnings reports and roughly half the Dow’s components were higher. Both companies were among the index’s worst decliners on Wednesday. A market that gives back yesterday’s good news within twenty-four hours is not processing new information about those companies; it is repricing risk appetite wholesale.

Elsewhere, the day’s biggest S&P 500 percentage losers read like a list of AI-adjacent capital spending proxies and rate-sensitive cyclicals: Lennox International down 20.97% to $430.02, Masco down 11.09%, KLA down 10.80% to $170.19, Micron Technology down 9.94% to $739.00 on nearly 70 million shares, and Super Micro Computer down 9.67% to $25.70.

The gainers were the mirror image. Garmin rose 16.24% to $294.83, GE HealthCare gained 12.15%, Cognizant Technology Solutions added 11.25%, SBA Communications rose 6.49% and EPAM Systems gained 6.46%. Exxon Mobil and Chevron rose with crude. Not one of those is an AI capital spending story. On a day when the market fell more than 1%, the leadership board was populated almost entirely by companies whose earnings do not depend on hyperscaler budgets.

Among the largest companies, the closes were relatively contained compared with the carnage further down the market: Apple finished at $338.19 (−0.6%), Amazon at $226.65 (−1.8%), Alphabet at $335.76 (+0.9%), Meta Platforms at $585.61 (−1.31%), Microsoft at $390.54 (−0.71%), Nvidia at $190.01 (−3.6%) and Tesla at $298.32 (−3.0%). The mega-cap complex was not where Wednesday’s damage concentrated. That distinction belonged to the semiconductor and memory names, and to anything selling equipment into a data center.

How far from the highs?

Context matters here, and the various indexes were in very different places entering the session.

The Nasdaq-100 entered correction territory on Wednesday, having fallen roughly 10% from its June high, dragged there by a chip complex in outright decline. The Nasdaq Composite closed at 24,442.94, having spent the session breaking below the 25,000 level that had held earlier in the month. The S&P 500, at 7,316.15 against a 52-week high of 7,620.90, was approximately 4.0% below its peak — a figure calculated from those two published levels. The Dow was closing in on its 50-day moving average for the first time in weeks but had not broken it decisively.

That dispersion is itself informative. When the technology-heavy indexes are 10% off their highs and the broad market is 4% off, the market is not experiencing a general derating. It is experiencing a rotation away from one theme, with the rest of the tape absorbing collateral damage. Wednesday was the first session in a while where the collateral damage looked as though it might be becoming the main event.

A short glossary for the charts everyone was watching

Market commentary on days like Wednesday leans heavily on technical vocabulary that can obscure more than it reveals. Because several of these terms recur throughout this article, they are worth defining plainly.

• 50-day and 200-day moving averages. The average closing price over the prior 50 or 200 sessions, recalculated daily. Neither has predictive power on its own. They matter because enough institutional participants use them as reference points that they become self-reinforcing decision levels — places where systematic strategies add or reduce exposure. On Wednesday the Dow closed slightly below its 50-day line for the first time in this pullback, and the Nasdaq Composite’s 200-day average sat close to the round 24,000 level.
• Overhead supply. The pool of shares held by investors who bought at higher prices and are waiting to sell at break-even. As a beaten-down stock rallies, it runs into successive layers of these sellers, which is why recovering from a large gap down usually takes longer than the decline did. This is a real, observable phenomenon in order books, not a superstition.
• Base and buy point. A “base” is a sideways consolidation after an advance. A “buy point” or “pivot” is the price above which a stock is considered to have resolved that consolidation upward. Whether these patterns have genuine forecasting value is contested in academic literature; what is not contested is that a large community of active traders acts on them, which affects short-term flow.
• Relative strength. A stock’s performance measured against a benchmark rather than in absolute terms. In a declining market, a stock that falls less than the index is showing relative strength — a description of what has already happened, not a forecast of what will.
• Bear steepening. Long-dated bond yields rising faster than short-dated ones. Typically read as the market demanding more compensation for inflation and duration risk rather than repricing near-term policy.

None of these tools tells anyone what a stock will do next. They are a shared language for describing where supply and demand have been meeting. Used as description, they are useful. Used as prophecy, they are not, and the July 29 session offered a compact demonstration of why: several stocks that looked technically constructive at 1:00 p.m. were technically broken by 4:00 p.m., with no change in their underlying businesses.

Seagate: the best earnings report nobody could buy

If one company captured the disconnect between fundamentals and price action on Wednesday, it was Seagate Technology.

Seagate reported fiscal fourth-quarter results after the close on July 28, covering the quarter ended July 3, 2026, and the numbers were exceptional by any standard. Revenue reached $3,629 million, up from $2,444 million a year earlier — growth of 48.5%. GAAP diluted earnings per share came in at $5.58 against $2.24 in the year-ago quarter. On a non-GAAP basis, diluted EPS was $5.71 versus $2.59, an increase of roughly 120%. Non-GAAP gross margin expanded 570 basis points to 52.7%. Operating cash flow was $1.3 billion and free cash flow $1.1 billion, the company’s strongest quarterly free cash flow in more than a decade.

A note on a widely repeated error: several accounts of Seagate’s quarter, including live market commentary on Wednesday afternoon, transposed the two headline growth rates and described the company as posting 120% revenue growth and 48% earnings growth. The correct framing is the reverse. Revenue grew about 48%; non-GAAP earnings per share grew about 120%. The distinction matters, because the gap between those two numbers is the story — it is operating leverage, and it is what a pricing-driven upcycle looks like on an income statement.

For the full fiscal year, Seagate generated $12.2 billion in revenue, up 34%, with non-GAAP diluted EPS of $15.58 and record free cash flow of $3.1 billion. The company retired $1.4 billion in debt and returned $810 million to shareholders across the year, and declared a quarterly dividend of $0.74 per share payable October 7, 2026 to holders of record on September 24.

The guidance was arguably better than the quarter. Seagate told investors to expect fiscal first-quarter 2027 revenue of $4.1 billion, plus or minus $100 million, and non-GAAP diluted EPS of $7.30, plus or minus $0.20. That is guidance for sequential revenue growth of roughly 13% off an already record base, and for earnings per share nearly 28% above the quarter just reported. Chief executive Dave Mosley attributed the momentum to “robust cloud data center demand and disciplined execution,” and pointed to the ramp of the company’s Mozaic platform built on heat-assisted magnetic recording, which now accounts for about 40% of nearline exabyte shipments.

Now the price action. Seagate opened Wednesday sharply higher and spent the morning as one of the few green names in a red semiconductor complex, trading up about 2% at midday. It closed up roughly 2.3% — well below where it had traded overnight, and, more importantly, still beneath its 50-day moving average, which the stock had broken during the memory-sector rout of the preceding two weeks.

Consider what that means. A company beat consensus on revenue and earnings, expanded gross margin by nearly 15 percentage points year over year, generated record free cash flow, and guided the next quarter meaningfully above where the current one landed — and the stock added 2%, then faded with the tape into the close. That is not a market rewarding fundamental performance. That is a market with a different question on its mind.

How the market got here: a timeline

July 29 was the collision point of several trajectories that had been running for months. Assembling them in order makes the session look less like a shock and more like an arrival.

• Early 2021. U.S. inflation moves above the Fed’s 2% target as the economy reopens from pandemic restrictions. It has not sustainably returned to target since.
• Mid-2022. Inflation peaks at just over 9%. The Fed executes eleven rate increases across 2022 and 2023, bringing the rate down substantially before progress stalls.
• February 28, 2026. U.S. and Israeli forces strike Iran. In response, Iran closes the Strait of Hormuz, through which roughly one-fifth of the world’s seaborne oil and gas passes, producing what has been described as the largest supply disruption in the history of the oil market. Energy prices surge.
• Spring 2026. Kevin Warsh becomes chairman of the Federal Reserve, appointed by President Trump. He begins removing forward guidance from FOMC communications. By the July meeting he has been in the role roughly nine weeks.
• April 2026. Meta Platforms raises its 2026 capital expenditure guidance to as much as $145 billion, roughly double its 2025 spending and more than its 2024 and 2025 outlays combined. The stock falls. Microsoft issues the fiscal fourth-quarter guidance against which its July results would later be measured, on April 29.
• April 10, 2026. Snowflake bottoms after a 56% peak-to-trough decline, part of a broader derating of enterprise software on fears of AI disintermediation. The subsequent recovery becomes one of the few durable uptrends in the technology complex.
• June 5, 2026. Meta falls 6.6% to close at $584.95 on a report that the company could raise tens of billions of dollars to fund its AI buildout. Year to date, the stock is down more than 14%.
• June 15, 2026. Memory and storage stocks surge on news of a U.S.-Iran peace agreement. Western Digital rises 13%, Micron 8% and SanDisk 6%. The relief proves temporary.
• June 17, 2026. The FOMC meets and holds. It is Warsh’s first meeting as chairman. The next meeting will be 42 days later.
• Late June 2026. The Nasdaq-100 sets the high from which it would fall 10% by July 29.
• July 2, 2026. Memory and storage stocks slide on supply-glut fears. SanDisk falls 11%, Seagate 7% and Micron 4% in a single morning — an early tremor in what became a month-long unwind.
• July 10, 2026. SK Hynix American depositary receipts begin trading in New York in what is described as the largest U.S. initial public offering ever by a foreign company. The ADR falls roughly 23% over the following three weeks.
• July 13, 2026. A brokerage note out of Seoul triggers what is reported as the largest single-day decline in SK Hynix’s history, dragging Micron, SanDisk and Western Digital down about 6% each.
• July 15, 2026. Warsh testifies before the Senate Banking Committee for the semiannual monetary policy report, declaring he has “no tolerance” for elevated inflation.
• Mid-July 2026. China’s ChangXin Memory Technologies prepares an initial public offering sized around $8.6 billion. Memory stocks fall again on the prospect of another wave of Chinese capacity.
• Roughly July 16–28, 2026. Meta Platforms falls for nine consecutive sessions, its longest daily losing streak since the 2012 IPO, shedding approximately $223 billion in market capitalization.
• July 22–24, 2026. Brent crude briefly trades above $100 a barrel on intensifying fighting, then begins retreating on hopes of de-escalation. Over three sessions crude falls roughly 16%, the steepest three-day drop since 2020.
• July 27, 2026. CXMT’s Shanghai debut reignites oversupply concerns. SanDisk falls 12%, Micron 5% and SK Hynix 8%.
• July 28, 2026. Overnight declines in Samsung, SK Hynix and Kioxia hit U.S. memory and storage stocks. The Nasdaq-100 approaches correction territory. Boeing and Sherwin-Williams deliver well-received Dow earnings and nearly half the index’s components rise. Seagate reports fiscal fourth-quarter results after the close. Treasury closes: 2-year 4.26%, 10-year 4.61%, 30-year 5.09%.
• July 29, 2026, before the open. Iran’s Islamic Revolutionary Guard Corps fires ballistic missiles at a U.S. airbase and a Central Command facility in Jordan; Jordanian defenses intercept them. U.S. and Saudi forces strike Iran-aligned groups in Iraq. Oil surges. The Kospi falls 6%.
• July 29, 2026, 2:00 p.m. ET. The FOMC holds rates 9–3 with three hawkish dissents and no forward guidance. Equities briefly rally, with the Nasdaq up about half a percent, before reversing during the press conference.
• July 29, 2026, 4:00 p.m. ET. The Dow closes down 1,153.18 points at 51,594.14; the S&P 500 finishes at 7,316.15, essentially on its low; the Nasdaq Composite ends at 24,442.94. The 30-year Treasury yield settles near 5.21%.
• July 29, 2026, after the close. Microsoft, Meta, Qualcomm, Lam Research, Fortinet, Starbucks, Arm, Glaukos and Silicon Motion all report.
• Overnight into July 30, 2026. U.S. forces strike roughly a dozen Iranian targets. Asian equities rebound, with the Nikkei up about 2%. U.S. futures point modestly higher ahead of June PCE inflation and second-quarter GDP.

The industry underneath the trade: why storage and memory are not the same business

A recurring source of confusion in coverage of the July selloff is the treatment of “memory and storage” as a single sector. The distinction is worth drawing carefully, because it explains why Seagate rose on a day Micron fell 10%.

Hard disk drives store data on spinning magnetic platters. Two companies — Seagate and Western Digital — account for essentially all of the enterprise market. The technology is mature, the capital intensity is high but the capacity additions are slow and incremental, and the competitive structure is a duopoly. Seagate’s differentiator is heat-assisted magnetic recording, marketed as the Mozaic platform, which allows substantially higher areal density per platter. That technology now accounts for roughly 40% of the company’s nearline exabyte shipments. Nearline storage is the high-capacity, moderately accessed tier that cloud providers use for data they need to keep but do not need instantly — which describes an enormous and growing share of what AI systems produce.

DRAM is volatile working memory. It is dominated by Samsung, SK Hynix and Micron, with high-bandwidth memory — the stacked DRAM used alongside AI accelerators — the fastest-growing and most profitable variant. NAND flash is non-volatile solid-state storage, supplied by Samsung, SK Hynix, Micron, SanDisk and Kioxia. Both DRAM and NAND are built in fabrication plants whose capacity can be expanded meaningfully within an eighteen-to-thirty-month planning horizon, and both have a long history of brutal boom-bust cycles driven by exactly that expandability.

Controllers are the chips that manage flash memory. Silicon Motion is the largest independent supplier, which is why it trades in the same industry group as the memory makers despite having a fundamentally different, more asset-light business model.

The strategic point is that AI demand hits all of these differently. Training clusters consume high-bandwidth DRAM voraciously. Inference at scale consumes NAND. And the data those systems generate and retain — logs, embeddings, checkpoints, model artifacts, training corpora — lands overwhelmingly on the cheapest durable medium available, which remains the hard drive. Seagate’s exabyte demand is therefore a second-derivative play on AI activity rather than a direct one, and it is supplied by an industry that cannot double its output by writing a check.

That is the structural argument for why the HDD names held up while the memory names collapsed. It is a real argument, and it is also the kind of nuance that stops mattering the moment a sector-level unwind becomes indiscriminate. On July 2, Seagate fell 7% alongside SanDisk’s 11% decline. Correlation within the group rises when volatility rises, which is the general problem with owning a differentiated company inside a distressed sector.

The question on the market’s mind: what the memory unwind is really about

To understand why Seagate’s results did not matter much on Wednesday, you have to look at what happened to the rest of the storage and memory complex over the preceding three weeks.

The trigger was South Korea. SK Hynix reported second-quarter results that included a record operating profit, up 557% year over year, and a record operating margin of 76%. By any historical standard for a memory manufacturer — an industry famous for destroying capital across cycles — those are extraordinary figures. The stock fell anyway, because the print missed elevated expectations and, more consequentially, because management guided 2026 capital expenditure up 50% to at least $31 billion.

That capex number is the crux. In a commodity industry, a 50% increase in supply-side investment by the market leader is not a signal of confidence that investors reward. It is a signal that the current pricing environment is attracting enough capital to end itself. Memory upcycles have historically been terminated not by demand collapsing but by supply arriving. The market read SK Hynix’s guidance as the sound of the next downcycle being funded.

Two other developments compounded it. China’s ChangXin Memory Technologies completed a widely watched Shanghai listing — an offering sized around $8.6 billion — that could bankroll another wave of domestic DRAM expansion. And SK Hynix’s own U.S.-listed American depositary receipts, which began trading in New York on July 10 in what was described as the largest-ever U.S. initial public offering by a foreign company, fell roughly 23% from the listing price in under three weeks. Barclays cut its price target on the shares to $300 from $330 while maintaining an Overweight rating.

The damage spread quickly. Micron, Samsung Electronics and SK Hynix collectively shed a staggering amount of market value in July — roughly $113 billion, $173 billion and $176 billion respectively, by one tally — and each fell more than 20% from recent closing highs, the conventional threshold for a bear market. The Roundhill Memory ETF, whose top three holdings are Samsung at about 25%, SK Hynix at about 24% and Micron at about 24%, dropped roughly 20% across five sessions. SanDisk fell about 36% over five days. On Wednesday alone, SanDisk was down 7% at midday to $1,019 and Micron closed down 9.94% at $739.00.

The broader read-across is the number that should get attention: the Philadelphia Semiconductor Index fell 19% in July, on pace for its worst month since 2008, with every single member trading below its 50-day moving average. That is not rotation within a sector. That is a sector-wide liquidation.

The split inside the storage trade

What makes Wednesday analytically interesting is that the selling was not indiscriminate. Seagate and Western Digital — both hard disk drive companies — were flat to higher while the NAND and DRAM names were being taken apart. Western Digital finished the midday session roughly unchanged near $465.

That divergence carries information. Hard drives and flash memory both store data, but they are different businesses with different capital cycles. The hyperscaler demand for mass-capacity nearline storage that Seagate serves is driven by the sheer volume of data AI systems generate and retain, and Seagate’s supply is constrained by a technology roadmap (HAMR) that competitors cannot easily replicate on short notice. DRAM and NAND, by contrast, are being flooded with new fabrication capacity. Same end market, opposite supply dynamics.

Morningstar analyst William Kerwin captured the tension in his assessment of SanDisk, calling the NAND boom “tremendous, but finite” while maintaining a $1,000 price target and a “Very High” uncertainty rating. That phrase — tremendous but finite — is a reasonable summary of what the market spent July trying to price.

The risk to the bullish storage case is that this distinction proves too subtle to survive a real derating. If hyperscalers cut capital spending plans, Seagate’s exabyte demand falls with everyone else’s, HAMR advantage or not. The stock’s inability to hold gains on an exceptional quarter suggests a meaningful cohort of investors is already positioned for that outcome.

Teradyne: a 20% gain that evaporated by the closing bell

The single cleanest illustration of Wednesday’s mood was Teradyne, and it is worth working through carefully because the headline numbers and the stock reaction appear irreconcilable until you read the guidance.

Teradyne posted record second-quarter revenue of $1,329 million, up 104% from the year-earlier quarter, with earnings up more than 300%. GAAP earnings per share came in at $2.38 and non-GAAP EPS at $2.47, against a Zacks consensus of $2.04. Operating income was $448 million for an operating margin of 33.7%, and gross margin improved 250 basis points year over year to 59.8%. Both revenue and non-GAAP EPS exceeded the high end of the company’s own guidance range. Record memory test revenue was driven by strength in DRAM and a resurgence in NAND final test.

The stock opened dramatically higher — up close to 20% at one point during Wednesday’s session — and closed roughly flat to slightly lower. An entire day’s 20-point advance vanished.

Commentary at the time attributed the reversal to the market’s refusal to reward good news. That is partly right and analytically incomplete. Look at the outlook: Teradyne guided third-quarter revenue to $1,200 million to $1,300 million, with non-GAAP net income of $1.85 to $2.15 per diluted share. The midpoint of that revenue range, $1,250 million, is roughly 6% below the $1,329 million the company just reported. The midpoint of the EPS guidance, $2.00, is about 19% below the $2.47 just delivered.

In other words, Teradyne told investors that the record quarter they were celebrating was, in management’s own projection, the peak — at least for now. In a market already convinced that semiconductor capital equipment demand is cresting, a guide for sequential decline is not a footnote. It is confirmation. The 20% opening pop was a reaction to the trailing numbers; the fade was a reaction to the forward ones. Both were rational.

This is a useful corrective to the reflex of treating every faded rally as evidence of irrational market conditions. Sometimes the tape is telling you something the headline did not.

Lam Research and the chip-equipment paradox

Lam Research reported results for the quarter ended June 28, 2026, and the numbers were, if anything, more impressive than Teradyne’s on a forward basis.

Revenue reached $6.72 billion, up 30% year over year and up from $5.84 billion in the March quarter. U.S. GAAP diluted EPS was $1.81 and non-GAAP diluted EPS $1.82, roughly 8% above consensus. Non-GAAP gross margin was 52.0% and non-GAAP operating margin 38.4%, both records, as was revenue. And the company guided the current quarter to approximately $8.1 billion in revenue — a figure roughly 13.6% above where analysts had modeled it, and about 21% above the quarter just reported.

That is the opposite of Teradyne’s shape. Lam is not guiding to a peak; it is guiding to acceleration.

The stock’s position entering the report tells you how little that mattered to sentiment beforehand. Lam had fallen more than 40% from recent highs and was trading roughly 25% below its 50-day moving average, though it remained above its 200-day line — better structural footing than most of its peers. Shares rose about 6% in the extended session following the release.

Two readings of that combination are defensible. The optimistic one: a 6% move on a guide that beats consensus by nearly 14% is an underreaction, and it reflects a market that has stopped listening rather than a market that has correctly assessed the business. The skeptical one: equipment orders are booked well in advance, so a strong near-term guide tells you about decisions hyperscalers and foundries made months ago, not about decisions they are making now. If the AI capital spending cycle is turning, semiconductor capital equipment is a lagging indicator by construction, and a great September quarter is entirely compatible with a poor 2027.

Both readings can be true simultaneously, which is why the stock is 40% off its high while reporting record results.

KLA and the rest of the equipment complex

The same day Lam guided to $8.1 billion, KLA fell 10.80% to $170.19. That divergence within a single industry group, on a single day, is worth noting: whatever the market was doing to semiconductor capital equipment on July 29, it was not applying a uniform multiple compression. Individual company results still moved individual stocks. The sector-level bear market and the company-level results were running on separate tracks, and where they intersected — as in Teradyne — the forward guidance decided the outcome.

After the bell: Microsoft’s milestone and Meta’s bill

Then the closing bell rang and the day’s actual news arrived.

Microsoft: $100 billion Azure, and a gain from Anthropic

Microsoft reported fiscal fourth-quarter results for the quarter ended June 30, 2026, and the headline was the one the company clearly wanted: Azure revenue exceeded $100 billion for the full fiscal year for the first time, growing 41%. At that scale, Azure trails Amazon Web Services and remains larger than Google Cloud.

The quarter itself, per Microsoft’s earnings release:

• Revenue of $90.0 billion, up 18% (up 17% in constant currency), against a consensus near $87.6 billion.
• Operating income of $40.6 billion, up 18%.
• Net income of $35.8 billion, up 31% on a GAAP basis; $35.3 billion and up 22% on a non-GAAP basis.
• Diluted EPS of $4.81 GAAP, up 32%; $4.74 non-GAAP, up 23%.
• Microsoft Cloud revenue of $59.3 billion, up 27%, with commercial remaining performance obligation up 84% to $678 billion.
• Intelligent Cloud revenue of $39.3 billion, up 32%, with Azure and other cloud services up 43%.
• Productivity and Business Processes revenue of $37.8 billion, up 14%.
• More Personal Computing revenue of $12.9 billion, down 4%, with Windows OEM and Devices down 7% and Xbox content and services down 10%.
• $10.2 billion returned to shareholders through dividends and buybacks in the quarter.

For the full fiscal year, revenue was $331.8 billion, up 18%; operating income $155.2 billion, up 21%; net income $133.7 billion, up 31%; and diluted EPS $17.95, up 32%.

Two items in that release deserve more scrutiny than they typically receive. First, Microsoft disclosed that discrete items produced a benefit of $0.27 on diluted EPS relative to the guidance it gave on April 29, and identified those items: a $3.2 billion gain from the company’s investment in Anthropic and lower-than-expected expenses tied to a Voluntary Retirement Program, partially offset by severance expense and impairment charges in Xbox. Microsoft stated that adjusting for those items, it still exceeded expectations across revenue, operating income and diluted EPS — a claim that is consistent with the revenue beat but which readers should note is the company’s own characterization.

Second, Microsoft’s non-GAAP figures exclude the impact of its investments in OpenAI. In the June quarter that adjustment reduced net income by $480 million, taking GAAP net income of $35,766 million down to an adjusted $35,286 million. For the full year the adjustment was larger and ran the other way relative to the prior year’s comparison: GAAP net income of $133,749 million adjusts to $128,786 million, a swing of $4,963 million. GAAP diluted EPS of $17.95 becomes $17.28 adjusted. Investors comparing Microsoft’s earnings growth across years need to be clear which basis they are using, because the OpenAI stake is now large enough to move the reported number by several percentage points in either direction.

That is not a criticism of the disclosure — Microsoft is unusually explicit about it. It is a caution about the shorthand. A headline that reads “Microsoft earnings up 32%” describes a GAAP figure flattered by a $3.2 billion mark on a private AI company and depressed by an equity-method drag from another. The operating business grew 18% on the top line and 18% at the operating income line. Those are the numbers that describe what Microsoft actually did.

Chief executive Satya Nadella framed the year around scale and adoption: “This year, Azure revenue surpassed $100 billion for the first time, and Microsoft 365 Copilot reached over 30 million paid seats.” Chief financial officer Amy Hood pointed to Microsoft Cloud revenue of $59.3 billion, up 27%, and on the call guided Azure to roughly 45% constant-currency growth in the fiscal first quarter — above the roughly 41.4% consensus tracked by StreetAccount.

The stock’s path through the evening is itself instructive. Microsoft was up about 3% in the minutes after the release, when the initial numbers were being digested. It ended the extended session up roughly 8%, and was trading up nearly 8% in Thursday premarket dealing from a July 29 close of $390.54. The gain built as the call progressed — which is to say, the Azure guidance mattered more than the reported quarter.

Context is required. Even after an 8% move, Microsoft remains in a well-defined downtrend, trading below a declining 50-day moving average and below its 10-week line on the weekly chart; market commentary on Wednesday afternoon placed the shares roughly 30% below their highs. One good quarter does not reverse a decline of that depth, and the stock was rejected at its 40-week moving average once already on the way down.

Meta: a revenue beat, and free cash flow of $784 million

Meta Platforms reported second-quarter results at 4:22 p.m. Eastern, and the market’s verdict was immediate and negative. Shares were down about 5% within minutes and extended to a decline of roughly 7% to 8% as the evening progressed. By Thursday premarket, Meta was down more than 8% from its July 29 close of $585.61.

The revenue line was not the problem. Meta reported revenue of $60.80 billion, up 28% year over year and ahead of consensus estimates that clustered between roughly $59.5 billion and $60.2 billion depending on the compiler. Ad impressions rose 14% and the average price per ad increased 12% — a healthy combination indicating both volume and pricing power. Family daily active people grew 3%. Headcount was down 1% year over year to 75,472 as of June 30.

Everything below the revenue line was the problem.

Diluted earnings per share came in at $6.18 against consensus estimates in the $7.13 to $7.22 range — a substantial miss. Meta attributed the pressure to $3.58 billion in one-time legal and severance charges, a higher effective tax rate, and rising expenses. Some of that is genuinely non-recurring. Some of it is the cost of running a company that is simultaneously expanding an AI infrastructure footprint and litigating an expanding set of regulatory matters.

The number that did the real damage was further down. Meta generated $31.86 billion in cash flow from operations during the quarter and $784 million in free cash flow. Read that again. A company producing nearly $32 billion of operating cash in three months converted less than a billion of it into free cash flow, because capital expenditure absorbed essentially all of it. Meta ended the quarter with $90.26 billion in cash, cash equivalents and marketable securities.

Free cash flow is what a business generates after paying to maintain and expand its asset base. It is the pool from which dividends, buybacks and debt reduction are funded. Meta has, for the moment, spent it. Whether that is visionary or reckless depends entirely on what the assets earn, and nobody — including Meta — can yet demonstrate that return.

The guidance did not help. Meta expects third-quarter revenue of $61 billion to $64 billion, against consensus near $62.7 billion — a range whose midpoint is roughly in line but whose low end implies a sharp deceleration. Full-year 2026 total expenses were guided to $165 billion to $169 billion, raised from a prior $162 billion to $169 billion. And full-year capital expenditures were guided to $130 billion to $145 billion, up from a prior range of $125 billion to $145 billion.

Note the mechanics of that capex revision. The top of the range did not move. The bottom did. Meta did not tell investors it will spend more than it previously said was possible; it told them it will no longer spend as little as it previously said was possible. In a market that has spent two months demanding evidence of spending discipline, raising the floor is the least welcome way to narrow a range.

Meta also flagged litigation exposure in unusually direct language, stating in the release that it continues “to see scrutiny on youth-related issues in several markets” and has “a number of youth-related trials scheduled for this year in the U.S., which may ultimately result in a material loss.” That is a company-authored disclosure of potential material loss, not an established liability, and it should be read as such — but it is a notable escalation in tone.

Chief executive Mark Zuckerberg’s framing was characteristically forward-looking: “AI is accelerating our core business today, powering our next generation of products, and opening the door to entirely new enterprise opportunities. The results are already showing, and I’m optimistic about the potential ahead.” The revenue growth and ad pricing data give that claim real support. The free cash flow line gives the market its counterargument.

The nine-day slide that preceded the print

Meta did not arrive at its earnings date in good standing. The stock had fallen for nine consecutive trading sessions into the report — the longest daily losing streak in its history since the May 2012 initial public offering — shedding roughly $223 billion in market capitalization along the way. It had already fallen 6.6% on June 5, closing at $584.95, on a report that the company could raise tens of billions of dollars to fund its AI push, and it was down more than 14% year to date at that point.

The pattern through 2026 has been consistent. In April, Meta raised its capital expenditure guidance and the stock fell. In June, reports of external financing to support the buildout sent it lower again. In July, the shares declined for nine straight days on no company-specific news at all. And on July 29, the company beat on revenue, missed on earnings, raised the floor of its spending range, and fell again.

One reasonable interpretation: the market has stopped treating Meta’s AI investment as an option with unlimited upside and started treating it as a liability with a known cost and an unknown payoff. Investors are, in effect, applying a discount rate to a promise. Another interpretation, equally defensible: Meta is building infrastructure that will be extremely valuable in three years, and the market’s twelve-month attention span is mispricing it. Which of those is correct will be settled by data that does not exist yet — specifically, evidence of incremental revenue attributable to the incremental compute.

The capex backlash is no longer confined to technology

The most underappreciated development of July 29 was not in the Nasdaq. It was Baird’s downgrade of Caterpillar.

The argument was not about machinery pricing, dealer inventories or China. It was that state and local governments are increasingly restricting data center construction, and that this political friction could reduce demand for Caterpillar’s power generation equipment beginning in 2027 — after the near-term boom already in the order book. The stock fell 6.91%, the worst performance in the Dow.

Follow the chain. Hyperscalers announce enormous capital spending programs. Those programs require data centers. Data centers require enormous quantities of electricity and on-site generation, which is what Caterpillar sells. Communities object to the electricity consumption, the water use, the noise and the tax abatements. Legislatures respond. And a Dow industrial trades down 7% on a Wednesday afternoon in July.

That is a materially different risk than the one the market has been debating. The AI capex debate to date has been about whether returns justify the spending — a question for management teams and shareholders. The Caterpillar downgrade introduces a question about whether the spending will be permitted at the intended scale. Regulatory and community resistance is not a variable that hyperscaler earnings calls typically address, and it is not one that responds to demonstrations of AI’s usefulness.

Combined with Lennox International’s 20.97% decline and Masco’s 11.09% drop on the same session — both housing and building products names sensitive to construction activity and long-term rates — the industrial complex was pricing something broader than a bad afternoon.

The rest of the after-hours tape

Qualcomm: the diversification story meets the memory bill

Qualcomm reported fiscal third-quarter revenue of $9.9 billion with non-GAAP earnings per share of $2.21, meeting the high end of its own revenue guidance but falling short of Wall Street’s earnings expectation. Handset chip sales came in at $5.1 billion, down 20% year over year, which management characterized as a bottoming in the China market.

The diversification narrative delivered. Automotive revenue set a record with 61% year-over-year growth, and the company raised its automotive annualized revenue run-rate outlook to approximately $7 billion exiting fiscal 2026, up from $6 billion. Qualcomm also doubled its fiscal 2029 non-handset revenue target to $40 billion.

The guidance is where the stock lost the room. Qualcomm sees fiscal fourth-quarter revenue of $9.7 billion to $10.5 billion with non-GAAP diluted EPS of $2.05 to $2.25 — in-line revenue with light earnings, which management tied to the supply crunch in computer components, memory in particular. The company also warned of accelerating declines in Apple-related revenue.

There is an elegant irony in that guidance. The memory price inflation that produced Seagate’s 570 basis points of gross margin expansion and Teradyne’s record memory test revenue is the same memory price inflation that is compressing Qualcomm’s margins. One industry’s pricing power is another’s cost of goods sold. Shares closed the regular session at $155.68, down 4.42%, and traded lower again after hours.

Arm Holdings: records, then a guidance haircut

Arm Holdings reported record fiscal first-quarter results with revenue up 22% year over year to $1.29 billion and non-GAAP EPS up 29% to $0.45, exceeding the high end of its guidance. Licensing and royalty revenue both hit records, and free cash flow rose 343%. Management disclosed that demand for its new AGI CPU has exceeded $2 billion, more than double the company’s initial fiscal 2027–2028 opportunity estimate of $1 billion, while flagging supply constraints across wafers, substrates, testing and memory.

Shares fell about 6% anyway, on guidance the market found underwhelming. Arm also faces a reported Federal Trade Commission inquiry, which some observers have argued is the more consequential development for the company than a single quarter’s numbers. As with any investigation, an inquiry is not a finding, a charge or a penalty, and readers should not treat it as one.

Silicon Motion: a clean beat, a red tape

Silicon Motion, which supplies controllers for the NAND market and is tracked in the same industry group as Seagate, Micron and Western Digital, reported second-quarter earnings of $2.43 per share against estimates of $2.18, on revenue of $451 million versus roughly $411 million expected. Net income was $136.1 million. The stock was down about 2.4% in the extended session at the time of the market close commentary — a modest decline relative to the sector, and one that left the shares comfortably above their 200-day moving average.

Fortinet: the day’s cleanest positive surprise

Fortinet did something almost nobody managed on Wednesday: it looked good before the news and better after it. During the regular session the security software company reversed higher above its 50-day moving average on strong volume, gaining about 2% on a day when the market fell more than 1.5%. Then it reported.

Second-quarter revenue was $2.05 billion, up 26% year over year against a Street consensus near $1.89 billion. Product revenue — the hardware-and-license line that leads the subscription business by several quarters — was $773 million, up 52%. Billings rose 33% to $2.37 billion. Earnings came in at $0.90 per share against consensus of $0.75.

The guidance was raised across the board: third-quarter revenue of $2.01 billion to $2.10 billion against a $1.95 billion consensus; full-year 2026 revenue of $8.02 billion to $8.18 billion, implying roughly 19% growth; full-year billings of $9.35 billion to $9.55 billion; and full-year EPS of $3.41 to $3.47, up from a prior $3.10 to $3.16. Shares rose roughly 10% to 12% after hours.

The distinction between billings and revenue matters here and is often blurred. Billings represent what customers were invoiced during the period, including amounts for services that will be delivered later; revenue is what has been recognized as earned. Billings growth of 33% against revenue growth of 26% indicates the business is accelerating faster than the income statement currently shows, because the deferred portion will be recognized in future quarters. Product revenue growth of 52% is the leading edge of the same signal: Fortinet is shipping a lot of appliances, and appliances pull subscription attachments behind them.

Starbucks: four quarters into a turnaround

Starbucks reported fiscal third-quarter results that made a strong case for chief executive Brian Niccol’s “Back to Starbucks” program, now roughly two years into his tenure after he arrived from Chipotle.

Global comparable store sales rose 7.9%, the fourth consecutive quarter of growth, led by transactions rather than price. In the U.S., comparable sales also rose 7.9%, with transactions up 4.2% and average ticket up 3.6%. That split is the important one: a retailer growing comps on traffic is winning customers, while a retailer growing comps on ticket alone is testing their tolerance for price increases. Starbucks did both, weighted toward the former.

Reported net revenues fell 1% to $9.3 billion, but that decline is an artifact of the company’s sale of a controlling stake in its China business rather than a demand signal. GAAP earnings per share of $0.91 rose 86%; adjusted EPS of $0.85 climbed 70%.

Management raised full-year fiscal 2026 adjusted EPS guidance to a range of $2.55 to $2.65 from a prior $2.25 to $2.45 — an increase of roughly 10% at the midpoint. It also now projects global same-store sales growth of nearly 6% and U.S. growth above 6%, versus prior guidance of at least 5% for both. Shares rose about 7% after hours and subsequently reached a 52-week high.

Starbucks was, in other words, the day’s proof that this market will still pay for growth — provided the growth arrives from customers walking through doors rather than from capital expenditure on speculative infrastructure.

Glaukos: revenue up 50%, still losing money

Glaukos, an ophthalmic medical technology company, reported second-quarter revenue of $185.6 million, up from $124.1 million a year earlier — growth of roughly 50% and well above the $150.9 million analyst consensus. U.S. glaucoma revenue was $118.5 million, including approximately $74 million from iDose TR, the company’s intraocular drug delivery implant. Epioxa contributed roughly $11 million in its first full quarter of commercial launch.

Management raised full-year 2026 net sales guidance to $680 million to $700 million from a prior $620 million to $635 million, and guided iDose TR revenue of $275 million to $285 million for the year.

The company is not profitable. Glaukos posted a net loss of $18.4 million in the quarter, or $(0.31) per diluted share. That is an important qualifier on a stock that rose roughly 9% after hours following a strong regular session on heavy volume. Commentary during Wednesday’s close suggested the company is expected to turn its first profit in 2027; that is a projection, and the quarterly results confirm it has not happened yet.

The narrow leadership that survived the day

Amid the wreckage, a handful of groups held up, and the composition of that list says something about where money is hiding.

Software: the laggard that stopped lagging

Enterprise software spent much of 2026 as the market’s problem child, hit by the argument that AI would disintermediate seat-based licensing models. That thesis appears to be losing force. The iShares Expanded Tech-Software Sector ETF finished Wednesday up about six-tenths of a percent — a fourth consecutive advance — even after being dragged down from its highs in the final hour.

Snowflake was the clearest example. The data cloud company rose 4.6% on Wednesday, clearing what chart-based investors identify as a cup-with-handle buy point, and closed off its highs with the rest of the tape. It has been one of the few software names showing a relative strength line near its highs while peers attempt to emerge from bottoming patterns.

The fundamental backdrop supports the price action. Snowflake reported first-quarter fiscal 2027 product revenue of $1.334 billion with growth accelerating to 34% year over year, an acceleration analysts have attributed substantially to Cortex Code, the company’s AI coding agent for data work. The stock fell 56% from peak to trough before bottoming on April 10, 2026, and has since recovered nearly all of that decline. Wells Fargo raised its price target on the shares to $500 in July. Snowflake’s next earnings report is due at the end of August.

A caution worth stating plainly: a price target is one firm’s opinion about a plausible future value, published at a moment in time, not an appraisal. Snowflake’s recovery from a 56% drawdown is a fact. Where it trades in twelve months is not.

Biotechnology and medical products

Healthcare has quietly become one of the better places to be. GE HealthCare’s 12.15% gain led Wednesday’s advancers, and medical products has produced a steady stream of stocks holding above their moving averages while technology unwinds.

Arcutis Biotherapeutics was among the names being watched into its own report. The dermatology company’s lead product, Zoryve, is a topical roflumilast franchise approved for plaque psoriasis, atopic dermatitis and seborrheic dermatitis. In the first quarter of 2026 Arcutis reported net revenues of $105.4 million, up 65% year over year, and reaffirmed full-year 2026 revenue guidance of $480 million to $495 million.

Here a correction is warranted to commentary circulating on Wednesday. Arcutis has not yet turned profitable on a GAAP basis. The company reported a first-quarter 2026 net loss of $11.3 million — a loss that narrowed by 55% from the first quarter of 2025, which is a meaningful improvement but is not the same thing as 55% earnings growth, and is not the same thing as profitability. Screens that rank stocks on adjusted earnings measures can produce a “turned profitable” flag on a company still reporting statutory losses, and readers evaluating this or any similar name should check which basis a given metric uses. Arcutis is scheduled to report second-quarter results after the close on August 5, 2026.

Transports, and the ones that broke

The transportation complex held support at its 50-day moving average on Wednesday despite falling 1.4%, a resilience made more notable by the fact that trucking stocks within the group were breaking down individually while airlines held their own. When a sector index holds a level that a majority of its components have already lost, the index is being carried by a shrinking number of names. That is a fragile kind of strength, and it is a pattern worth monitoring rather than celebrating.

The former leaders that are not leading

Robinhood Markets fell about 3% on Wednesday, extending a pullback that has taken the stock more than 50% off its highs. Its earnings growth has decelerated over recent quarters and revenue growth has slowed as well, though second-quarter revenue was expected to pick up somewhat. The company’s fortunes have also become more closely tied to cryptocurrency markets than in the past, and Bitcoin’s recent softness — trading near $63,564 on Wednesday and around $64,208 early Thursday — has not helped.

Former leadership that stops leading is one of the more reliable characteristics of a maturing market advance. It is not, by itself, a forecast of anything. But a market whose speculative retail-facing names peaked months ago and have since halved is a market whose risk appetite changed before the index did.

What the real-time commentary got right, and what needs correcting

Because a great deal of what investors absorb about a session like Wednesday comes from live market commentary produced within minutes of the close, it is worth assessing how that first draft held up against the documented record.

What held up well. The observation that the session’s selling accelerated after the press conference rather than after the statement is confirmed by the intraday path of both equities and the long end of the Treasury curve. The characterization of a yield curve steepening on rising long rates and falling short rates is consistent with the H.15 data and with reported intraday moves. The identification of the industrial complex as an unexpected source of weakness — with Boeing, Sherwin-Williams, Goldman Sachs, JPMorgan and Caterpillar all falling hard a day after the Dow looked healthy — matches the sector data showing industrials down 3.42%. The point that this market is “not rewarding even good earnings reports,” illustrated by Teradyne, is empirically accurate as a description of price action, even if the specific case has a guidance explanation. And the framing of Meta’s problem as overhead supply from investors who bought higher is a fair description of the mechanics facing any stock that has gapped down repeatedly.

What needs correcting or qualifying. Four items stand out.

The Seagate growth rates were transposed. Revenue grew 48.5%; non-GAAP earnings per share grew about 120%. Commentary that reversed those figures understated the operating leverage that is the most striking feature of the quarter.

Arcutis has not turned profitable. The company reported a GAAP net loss of $11.3 million in the first quarter of 2026. The 55% figure describes the reduction in that loss, not earnings growth.

The Iran incident description was close but imprecise. Iran’s Islamic Revolutionary Guard Corps fired ballistic missiles at a U.S. airbase and a U.S. Central Command facility in Jordan; Jordan intercepted them. Separately, U.S. and Saudi forces struck Iran-aligned groups in Iraq. Characterizing the event solely as missiles fired at “a military base in Jordan” omits both the American target and the coalition response that helped drive the oil move.

The Fed chair’s name and the vote were reported correctly, but the significance of the dissents deserves more weight than a live segment can give it. Three dissents in favor of tightening, from three regional presidents with meaningfully different analytical traditions, is the most hawkish internal configuration the FOMC has displayed in years. It is the single best predictor available of what September might bring — better than anything the chairman said.

None of these corrections undermine the analytical value of same-day market commentary, which serves a different purpose than the record. But they illustrate why the first account of a trading day should be treated as a sketch rather than a source.

The bull case, stated at its strongest

An honest assessment of Wednesday requires taking the constructive interpretation seriously, because there is one and it is not trivial.

The economy is not breaking. The FOMC’s own statement described activity expanding “at a solid pace,” with strong productivity growth and capital investment, job gains keeping pace with the workforce and an unemployment rate that “has changed little.” Warsh went further in his opening remarks, calling business investment growth “the most striking feature of the economy” and citing four-quarter growth of nearly 20% in AI-related high-tech equipment and software. Whatever is happening in the semiconductor tape, it is not a demand collapse showing up in the national accounts.

Corporate earnings are, in aggregate, excellent. Consider what was reported inside a single 24-hour window: Seagate grew EPS 120%, Teradyne grew revenue 104%, Lam Research posted record revenue and margins and guided 21% higher sequentially, Fortinet grew billings 33% and raised guidance three ways, Microsoft grew revenue 18% at a $332 billion annual scale, Starbucks raised full-year EPS guidance by roughly 10%, Glaukos raised sales guidance by about 10%, Arm posted records with free cash flow up 343%, and Silicon Motion beat on both lines. Meta was the only outright earnings disappointment in the group, and even Meta grew revenue 28%.

The selling looks like positioning, not fundamentals. A market where the Philadelphia Semiconductor Index falls 19% in a month while its constituents report records is a market unwinding a crowded trade. Crowded-trade unwinds are painful and they end. They do not require a recession to end, only for the marginal seller to run out of shares.

Rate expectations may be overshooting. Roughly 76% of traders were pricing a September increase after the meeting, up from 59% a month earlier. If June PCE and subsequent inflation prints cool — and core inflation did cool in June, helped by slower growth in apartment rents — the hawkish repricing in the long end could partly reverse. Bond selloffs driven by expectation shifts unwind faster than those driven by supply.

The market is not expensive by its own recent standards. The S&P 500 finished about 4% below its 52-week high. That is a pullback, not a bear market. The 200-day moving average for the S&P sits far below current levels, which is a way of saying the index remains in a well-established uptrend that a 4% decline does not threaten.

The bear case, stated at its strongest

The bond market is making a credibility judgment, and it is not flattering. A 30-year yield at a nineteen-year high, rising inflation breakevens, a weaker dollar and falling equities all on the same afternoon is a coherent package with an unpleasant interpretation: investors believe the Fed will not act quickly enough, and they are demanding compensation for that belief. Warsh can describe the move as markets “playing the ball, not the referee,” but the ball in question is inflation expectations, and if those become unanchored the eventual policy response has to be larger.

The AI capital expenditure cycle is showing supply-side stress from three directions at once. SK Hynix is raising capex 50%. CXMT has raised billions in Shanghai to expand Chinese memory capacity. And Baird is downgrading Caterpillar because local governments are restricting the data centers themselves. Excess capacity, new entrants and permitting friction are the three classic ways a capital cycle ends, and all three are visible simultaneously.

Free cash flow is disappearing at the demand end. Meta generated $31.86 billion of operating cash flow and $784 million of free cash flow. That is the most concrete evidence available that the current level of AI investment is not self-funding on a quarterly basis at one of the most profitable advertising businesses ever built. Investors who have been told the spending is affordable now have a number.

Guidance is deteriorating even where results are not. Teradyne guided sequentially lower. Qualcomm guided earnings light on memory costs. Arm’s guidance disappointed. Meta raised the floor of its spending. Companies do not usually all get cautious in the same week by coincidence.

Breadth is narrowing. The equal-weighted S&P 500 made a new high on July 28 and then fell about nine-tenths of a percent on July 29 — a one-day reversal off a fresh high. The Russell 2000 broke below its 50-day moving average after holding up relatively well. The transports are holding an index-level support that many of their components have already lost. When the number of stocks participating in an advance shrinks, the advance becomes dependent on a smaller group, and the eventual failure of that group has outsized consequences.

Berkshire Hathaway is sitting on a record $397 billion in cash. That is not a market call — Berkshire’s cash pile reflects the difficulty of deploying capital at scale as much as any view about valuations. But it is a data point that a very patient buyer with very few constraints has not found much worth buying.

Historical comparisons, and their limits

Several analogies have been circulating. Each illuminates something and misleads about something else.

The 2018 hawkish-hold parallel. In late 2018, a Federal Reserve perceived as tightening into slowing growth triggered a sharp fourth-quarter equity decline that reversed almost entirely once policy expectations shifted in early 2019. The similarity is a market punishing a central bank for insufficient flexibility. The difference is decisive: in 2018 inflation was near target and the Fed had room to pivot dovish. In 2026 inflation has been above target for more than five years and the dissents are on the hawkish side. There is no comparable pivot available.

The 1994 bond rout. The comparison to 1994, when an unexpected tightening cycle produced one of the worst bond years on record, captures the duration pain but not the mechanism. In 1994 the Fed was surprising markets by hiking. In 2026 the Fed is surprising markets by not hiking, and the long end is selling off as a result. The direction of the surprise is inverted.

The 2000 capital-cycle bust. The telecommunications buildout of the late 1990s produced enormous capital spending on infrastructure whose revenue arrived years later than projected, and the equipment suppliers were destroyed long before the demand thesis was proven wrong. The parallel to today’s data center construction is genuine, and the Caterpillar downgrade rhymes with it uncomfortably. The important difference is that today’s spenders are extraordinarily profitable incumbents funding capex from operating cash flow rather than debt-financed startups. Meta’s free cash flow fell to $784 million; it did not go negative, and the company holds $90.26 billion in cash and marketable securities. That is a materially stronger balance sheet position than the 2000 comparison implies.

The 1970s supply-shock template. The current combination — an oil shock layered on top of persistent inflation, with a central bank under political pressure — has obvious echoes. The differences are that the U.S. economy is far less energy-intensive per unit of output than it was fifty years ago, that the Fed retains a formal 2% target it has repeatedly reaffirmed, and that long-term inflation expectations, at roughly 2.20 percentage points implied by ten-year breakevens as of July 28, remain close to target despite five years of overshoot. That last figure is the most important number in the entire debate, and it is the one to watch.

Warsh’s Fed: the institutional experiment underneath the market move

It is difficult to evaluate July 29 without engaging with what Kevin Warsh is attempting, because the market reaction was as much to the communication regime as to the rate decision.

Warsh was appointed by President Donald Trump, who has publicly and repeatedly pressed the Fed to cut rates. Warsh has taken the opposite position, telling Congress earlier in July, during Senate Banking Committee testimony on July 15, that he has “no tolerance” for elevated inflation. That is an unusual dynamic: a chairman appointed by a president who wants easier policy, running a committee whose dissents all point tighter.

The communication changes are substantive. Forward guidance — the practice of signaling the likely future path of policy — has been a core Fed tool since the financial crisis. It was designed to increase policy potency at the zero lower bound, when the Committee could not cut further and needed to influence expectations directly. Warsh’s argument, laid out in his July 29 opening statement, is that the tool has outlived its usefulness and now distorts price formation.

“I understand the desire for rolling forecasts and commentary from this Committee,” he said. “But for our part, we need to observe market reaction to developments, direct and unfiltered.”

He also disclosed the four questions that structured the Committee’s July discussion, which is itself a form of transparency even as the forecasts disappear. Those questions: whether five years of high inflation has permanently altered the current policy calculus (“has the past really passed?”); whether recent shocks — pandemic supply chains, military conflict, energy disruption, tariff increases and the AI investment surge — differ in their effects on output and employment as well as in their sources; whether shock-driven price increases in specific sectors such as memory and logic chips indicate a broader inflationary dynamic or merely reflect where the light is brightest; and how much accommodation the balance sheet is still providing if the policy rate is meant to be the primary instrument.

That last question deserves attention. The statement noted the Committee “is continuing its policy of maintaining ample reserves in the banking system.” If the Committee concludes the balance sheet is supplying more accommodation than intended, tightening could arrive through that channel rather than through the funds rate — a distinction that matters enormously for the long end of the curve, and one that has not been widely priced.

The evaluation of Warsh’s experiment cannot be made from one meeting. The fair reading of July 29 is that the regime is producing exactly what he said it would — larger, faster market reactions to data and events — and that the market has not yet decided whether the resulting volatility is a feature or a cost. His own framing invites the test: “where necessary and appropriate, we will not hesitate to act.” September will show what he meant.

What the decision means outside the market

Most people affected by Wednesday’s session do not own semiconductor stocks. The relevant transmission runs through borrowing costs.

As of July 28, the bank prime loan rate stood at 6.75% and the Federal Reserve’s discount window primary credit rate at 3.75%, per the Fed’s H.15 release. Those are anchored to the policy rate, which did not change. What did change is the long end — and long-term borrowing costs for households and businesses key off the 10-year and 30-year Treasury yields rather than the funds rate.

A 30-year Treasury yield at 5.21% is consistent with elevated fixed mortgage rates, higher long-dated corporate borrowing costs, and higher costs for state and municipal issuers funding infrastructure. If the yield holds at these levels, the practical effect on the housing market and on capital-intensive industries is meaningfully tighter conditions than the unchanged policy rate suggests. That is one plausible explanation for why Lennox International, a heating and cooling manufacturer, and Masco, a building products company, were among the worst performers on a day the Fed did nothing.

On the other side of the ledger, savers continue to earn yields that would have been unimaginable for most of the 2010s. Treasury bills were yielding between roughly 3.64% and 3.91% across the four-week to one-year range as of July 28. Rates on any specific deposit product vary by institution and change frequently; the figures here describe the Treasury market on a specific date, not the terms available on any particular account.

Inflation, meanwhile, continues to show up where households notice it most. A pound of ground beef reached $6.82 in the most recent data, a figure that has been cited as weighing on consumer confidence. Core inflation cooled in June, helped by decelerating apartment rents and a temporary drop in gasoline prices, but “cooled” means prices rose more slowly, not that they fell. That distinction is the source of a good deal of the gap between official inflation statistics and how the economy feels.

Risks and uncertainties

Several specific risks follow from the July 29 configuration, listed roughly in order of how quickly they could matter.

• Escalation in the Iran conflict. The Strait of Hormuz remains closed. Houthi attacks on shipping through Bab el-Mandeb continue. Overnight into Thursday, U.S. forces struck a dozen Iranian targets. Each escalation raises energy prices, which raises headline inflation, which constrains the Fed. A sustained return above $100 Brent would make a September hike close to unavoidable regardless of core inflation.
• A hawkish surprise in September. With roughly three-quarters of traders now positioned for a September increase, a 50 basis point move — or a rate increase combined with a change in balance sheet policy — would be a genuine shock. The absence of forward guidance means there is no mechanism for the Fed to prepare markets for it.
• Memory oversupply arriving faster than expected. SK Hynix’s capex increase and CXMT’s expansion are both multi-year builds, but memory pricing is famously reflexive and can turn on order-book signals long before physical capacity comes online.
• Hyperscaler capital spending revisions. Amazon and Apple report Thursday. If either signals restraint, the read-across to the equipment and storage complex would be immediate. If both raise spending, the Caterpillar-style permitting risk becomes more prominent rather than less.
• Meta’s litigation exposure. The company itself has disclosed that youth-related trials scheduled in the U.S. this year “may ultimately result in a material loss.” That is a company statement about possibility, not an established liability, and the outcomes are not predictable. It is nonetheless a disclosed risk that investors should weigh.
• Data center permitting and community opposition. The risk highlighted in Baird’s Caterpillar downgrade applies across the buildout chain: turbine and generator suppliers, electrical equipment makers, construction firms and utilities. It is a slow-moving, politically driven risk that does not resolve on an earnings call.
• Narrowing breadth. A market carried by fewer names is more fragile. The equal-weight S&P 500 reversing off a new high, the Russell 2000 losing its 50-day line, and transports holding index support that individual components have lost are all consistent with that dynamic.
• Concentration in the storage thesis. Seagate’s results depend heavily on cloud data center demand, which depends on a small number of very large customers. Customer concentration is a structural feature of the business, and it cuts both ways.
• Currency and international spillover. The Kospi’s 6% drop demonstrated that the memory unwind is a global risk event, not a U.S. sector story. A weaker dollar alongside rising U.S. yields is an unusual combination that can signal reduced foreign appetite for Treasuries.

What happened after the video, and overnight

Markets did not stand still after the closing bell commentary.

Microsoft’s after-hours gain expanded from about 3% in the first minutes to roughly 8% by the end of the extended session, as the Azure guidance from the earnings call was absorbed. Meta’s decline deepened from about 5% to roughly 7% to 8%. Fortinet’s gain expanded from 10% to about 12%. Those revisions are a reminder that the initial print and the conference call are two separate events, and the second one frequently matters more.

Overnight, U.S. forces launched fresh strikes against roughly a dozen Iranian targets, threatening a further escalation in a conflict that markets had briefly hoped was cooling. Oil prices nonetheless eased in Thursday morning trading — a reminder that a market that has already priced substantial conflict risk does not necessarily add to it on every headline.

Asian equities rebounded. The Nikkei 225 rose about 2%, a meaningful reversal after the region led Wednesday’s chip selloff.

By the early hours of Thursday, July 30, U.S. equity futures were pointing modestly higher: S&P 500 futures at 7,378.25, up 0.37%; Dow futures at 51,918, up 0.30%; Nasdaq-100 futures at 27,520, up 0.65%; and Russell 2000 futures at 2,923.70, up 0.28%. The Cboe Volatility Index was quoted at 19.82, down about 4%. Gold was at $4,121.20, up 0.59%. Meta was indicated down more than 8% in premarket dealing and Microsoft up nearly 8%. The 30-year Treasury yield was quoted at 5.20% — essentially unchanged from Wednesday’s multi-decade high, which is itself notable. An overnight bond rally would have suggested Wednesday’s move was an overshoot. There was none.

Market data in this section reflects levels quoted between approximately 4:20 a.m. and 5:30 a.m. Eastern Time on Thursday, July 30, 2026, ahead of the U.S. cash open and ahead of the morning’s economic releases. Prices in premarket trading are frequently thin and are not always indicative of where a security opens.

What happens next: the confirmed calendar

Several scheduled events will do more to resolve Wednesday’s questions than any amount of commentary.

• Thursday, July 30, 8:30 a.m. ET. The Bureau of Economic Analysis releases the June personal consumption expenditures price index — the Fed’s preferred inflation gauge — alongside the advance estimate of second-quarter gross domestic product. Weekly initial and continuing jobless claims are published at the same time. Consensus ahead of the release looked for GDP growth of roughly 2.3% at an annual rate and a core PCE increase of about 0.2% month over month, following 0.3% in the prior reading. Advance GDP estimates are preliminary and subject to two subsequent revisions.
• Thursday, July 30, after the close. Amazon and Apple report. For Amazon, the focus is Amazon Web Services growth and capital expenditure. For Apple, attention is on gross margin given rising memory chip costs — the same input inflation Qualcomm cited in its guidance.
• Wednesday, August 5. Arcutis Biotherapeutics reports second-quarter results after the close.
• Late August. Snowflake reports fiscal second-quarter results.
• September 15–16. The next scheduled FOMC meeting. Following July’s decision, roughly 76% of futures traders were positioned for a rate increase at that meeting, up from about 59% a month earlier, based on the CME FedWatch tool. Market-implied probabilities describe positioning, not outcomes, and they move continuously with incoming data.

Beyond the calendar, the week ahead carries an unusually heavy earnings load among mid-cap growth companies, which will provide a broader read on whether the market’s refusal to reward good results is confined to semiconductors or applies across the growth complex.

Frequently asked questions

Why did the stock market fall on July 29, 2026?

Three forces compounded. The Federal Reserve held rates steady with three officials dissenting in favor of a hike and offered no guidance about future policy, which pushed long-term Treasury yields to multi-decade highs. Iranian missiles fired toward U.S. positions in Jordan sent crude up nearly 8%. And an ongoing unwind in semiconductor and memory stocks accelerated. The Dow closed down 1,153.18 points, or 2.19%.

What did the Federal Reserve decide at the July 2026 meeting?

The FOMC voted 9–3 to maintain the federal funds target range at 3.50% to 3.75%, the fifth consecutive meeting without a change. Beth Hammack, Neel Kashkari and Lorie Logan dissented, each preferring a quarter-point increase. The statement contained no forward guidance about future meetings.

Who is Kevin Warsh and why does his approach matter?

Kevin Warsh is the chairman of the Federal Reserve, appointed by President Trump, and July 29 was his second FOMC meeting. He has eliminated forward guidance from the Fed’s post-meeting statements, arguing that markets should respond to economic data rather than to central bank commentary. The July session was the first major test of how markets behave under that regime.

Will the Fed raise interest rates in September?

Nobody knows, and the Fed has deliberately declined to say. Market pricing after the July meeting implied roughly a 76% probability of an increase at the September 15–16 meeting, up from about 59% a month earlier. Three FOMC members already voted for a hike in July. Those are the facts available; the outcome will depend substantially on inflation data between now and then.

Why did Treasury yields rise if the Fed did not raise rates?

Because the two are different things. The Fed sets the overnight policy rate; the market sets everything else. Long-dated yields rose because investors demanded more compensation for inflation risk and uncertainty over a 10- to 30-year horizon — a judgment about the Fed’s likely trajectory rather than its current setting. The 30-year yield reached 5.21%, its highest since 2007, while 2-year yields fell.

Did Microsoft beat earnings expectations?

Yes. Microsoft reported fiscal fourth-quarter revenue of $90.0 billion, up 18% and above consensus near $87.6 billion, with non-GAAP diluted EPS of $4.74, up 23%. Azure revenue exceeded $100 billion for the full fiscal year for the first time, growing 41%. Shares rose roughly 8% in extended trading. Note that GAAP results included a $3.2 billion gain on the company’s Anthropic investment.

Why did Meta stock fall after reporting revenue growth of 28%?

Meta beat on revenue but missed badly on earnings, reporting $6.18 per share against consensus estimates in the $7.13 to $7.22 range, weighed down by $3.58 billion in one-time legal and severance charges and a higher tax rate. More significantly, free cash flow fell to $784 million despite $31.86 billion in operating cash flow, and the company raised the low end of its 2026 capital expenditure guidance to $130 billion from $125 billion. Shares fell roughly 7% to 8% after hours.

How strong were Seagate’s earnings?

Very strong. Fiscal fourth-quarter revenue was $3,629 million, up 48.5% year over year, with non-GAAP diluted EPS of $5.71 against $2.59 a year earlier — growth of about 120%. Non-GAAP gross margin expanded 570 basis points to 52.7% and free cash flow reached $1.1 billion. The company guided fiscal first-quarter 2027 revenue to $4.1 billion plus or minus $100 million and non-GAAP EPS to $7.30 plus or minus $0.20.

Why are semiconductor stocks falling if their earnings are so good?

Because the market is pricing the next cycle rather than the current one. SK Hynix guided 2026 capital expenditure up 50% to at least $31 billion, China’s CXMT completed a large Shanghai listing that could fund further memory capacity, and investors have grown skeptical about how long hyperscaler spending can grow at recent rates. The Philadelphia Semiconductor Index fell 19% in July, on pace for its worst month since 2008, with every member below its 50-day moving average.

What caused oil prices to jump on July 29?

Iran’s Islamic Revolutionary Guard Corps fired ballistic missiles at a U.S. airbase and a U.S. Central Command facility in Jordan, which Jordan intercepted, and U.S. and Saudi forces struck Iran-aligned groups in Iraq. Brent crude rose 7.9% to settle at $90.74 and WTI rose 6.6% to $84.46, partially reversing a 16% three-day decline that had been the steepest since 2020.

Is the stock market in a correction?

The Nasdaq-100 entered correction territory on July 29, having fallen roughly 10% from its June high. The S&P 500 finished about 4.0% below its 52-week high of 7,620.90 — a pullback, not a correction, by the conventional 10% definition. Different indexes are in materially different positions, which is itself a defining feature of this market.

What should investors watch next?

The June PCE inflation report and second-quarter GDP, both released the morning of July 30; Amazon and Apple earnings after Thursday’s close, particularly their capital expenditure commentary; the behavior of the 30-year Treasury yield around 5.20%; and whether the market begins rewarding strong earnings again or continues to sell them. None of this constitutes a recommendation to take any particular action.

Final assessment

The most important thing that happened on July 29 was not the rate decision. It was the divergence between what companies reported and what their shares did.

Within a single trading day and the evening that followed, American corporations disclosed earnings growth of 120% at Seagate, revenue growth of 104% at Teradyne, record margins and a 21% sequential revenue guide at Lam Research, 33% billings growth and a triple guidance raise at Fortinet, 43% Azure growth at Microsoft, a 10% profit guidance increase at Starbucks, and free cash flow up 343% at Arm. And the Dow fell 1,153 points, the semiconductor index extended its worst month since 2008, and a company that just posted its best quarterly free cash flow in over a decade could manage a 2% gain before fading into the close.

That gap is the story, and it has a coherent explanation. The market is no longer trading on what these businesses earned. It is trading on the discount rate applied to what they will earn, and on whether the capital being deployed today produces the cash flows being promised for the 2030s. Both of those variables moved against equities on Wednesday. The 30-year Treasury yield hit a nineteen-year high, which mechanically reduces the present value of distant earnings. And Meta demonstrated, with an $784 million free cash flow figure that no amount of framing can soften, exactly what the AI buildout costs a company that is paying for it out of pocket.

The strongest evidence for the constructive case is that the underlying economy is not cracking. The Fed’s own assessment describes solid growth, strong capital investment and stable employment. Corporate profitability is exceptional. Inflation breakevens, at roughly 2.20 percentage points over ten years as of July 28, indicate that markets still expect the Fed to get back to target eventually. Crowded-trade unwinds are violent and they conclude.

The strongest evidence against it is that three separate constraints on the AI capital cycle became visible in the same week — oversupply from SK Hynix and CXMT, cash flow exhaustion at Meta, and permitting resistance flagged in Baird’s Caterpillar downgrade — while the bond market simultaneously signaled that it does not believe the central bank is moving fast enough on inflation. Those two pressures compound each other. Higher long rates make speculative capital spending harder to justify, and doubts about capital spending returns remove the growth story that justified paying up for duration.

What genuinely changed on July 29 is that the Federal Reserve stopped mediating between markets and reality. Warsh said explicitly that he wants to observe market reactions “direct and unfiltered.” He got one. The reaction was a 1,153-point decline in the Dow, a nineteen-year high in the 30-year yield, and an equity market that punished every company whose future depends on borrowed time and cheap capital while rewarding the ones selling coffee, firewalls and medical devices.

What remains uncertain is nearly everything about September. The Committee gave no signal, three of its members have already voted to tighten, and the data that will decide the question arrives over the next six weeks starting Thursday morning. The one thing the July meeting settled is that the Fed will not be softening the blow in advance. Investors who have spent fifteen years being told what comes next now have to work it out for themselves, in a market where the cost of being wrong about long-duration assets has just gone up.

Sources

• Federal Reserve Board — FOMC statement, July 29, 2026
• Federal Reserve Board — Transcript of Chairman Warsh’s press conference opening statement, July 29, 2026 (preliminary)
• Federal Reserve Board — H.15 Selected Interest Rates (Daily), release of July 29, 2026
• Seagate Technology — Fiscal fourth quarter and fiscal year 2026 financial results
• Microsoft — FY26 Q4 earnings press release
• Lam Research — Financial results for the quarter ended June 28, 2026
• Teradyne — Second quarter 2026 results, Form 8-K exhibit filed with the SEC
• Starbucks — Q3 fiscal year 2026 results
• Glaukos — Second quarter 2026 results and raised 2026 sales outlook, Form 8-K
• Fortune / Associated Press — “Fed, in a nail-biter July meeting, decides to leave rates unchanged,” Paul Wiseman
• Fortune — Fed expected to hold as Iran-Jordan clash reignites war fears and oil jumps
• CNBC — 30-year Treasury yield hits highest level since 2007 after Fed keeps rates unchanged
• Bloomberg — Bond yields at 19-year high send Warsh a credibility warning
• CNN Business — The bond market to Kevin Warsh: what are you doing about inflation?
• CNBC — Brent oil jumps back above $90 as U.S.-Iran strikes resume
• Business Recorder / Reuters — Oil jumps nearly 7% on escalating Middle East airstrikes
• The Motley Fool — Stock Market Today, July 29: stocks slide on hawkish Fed and increased Middle East tensions
• CNBC — Chip stocks shed more than $1 trillion as selloff hits companies powering the AI boom
• 24/7 Wall St. — SanDisk sinks 7%, Micron slides 6% as memory selloff intensifies; Seagate bucks the trend
• CNBC — Caterpillar downgraded on growing data center buildout backlash
• Benzinga — Meta reports mixed Q2 results, raises low end of capex guidance range
• CNBC — Meta’s stock drops on disappointing guidance and dwindling free cash flow
• Crypto Briefing — Meta stock hits longest daily losing streak in history
• CNBC — Microsoft beats Q4 cloud expectations as full-year Azure revenue tops $100 billion
• Benzinga — Fortinet stock soars on Q2 results and raised guidance
• Investing.com — Qualcomm fiscal Q3 2026 presentation: diversification accelerates amid headwinds
• Investing.com — Arm fiscal Q1 2027 presentation: AI surge drives record results
• Associated Press — Silicon Motion Q2 earnings snapshot
• Yahoo Finance / Zacks — Seagate reports Q4 earnings: what key metrics have to say
• Benzinga — Snowflake stock rises as Wells Fargo lifts target to $500
• Arcutis Biotherapeutics — Notice of second quarter 2026 financial results, August 5, 2026
• Yahoo Finance — Stock market today, Thursday July 30: futures edge up ahead of inflation data
• Investing.com — Core PCE price index, GDP and jobless claims due Thursday
• Bureau of Economic Analysis — Personal consumption expenditures price index
• CME Group — FedWatch tool, market-implied probabilities for FOMC decisions
• Investor’s Business Daily — Stock Market Today closing-bell commentary with Alissa Coram and Ken Shreve

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.