July was defined by a violent reallocation out of AI infrastructure and back into value, followed by a two-session reversal that recovered a substantial portion of it. Both halves matter. The rotation was real and the snapback was equally real, and a month-end table alone tells almost none of the story.

Through July 29, the S&P 500 Index had declined 2.39% while the equal-weighted S&P 500 advanced 1.39%, a gap approaching 4% that captured the shift beneath a placid headline. The Russell 3000 Value Index gained 2.71% against an 8.22% loss for the Russell 3000 Growth Index, a spread of nearly 11% inside a single month. Leadership inverted. Energy rose 10.43%, Financials added 5.73%, Consumer Staples climbed 5.16%, and Health Care gained 4.78%, while Technology shed 12.57%, Consumer Discretionary fell 4.83%, and Industrials declined 4.63%. The Nasdaq 100 Index dropped 10.17%, and the Nasdaq Composite Index fell 6.74%.

The Bloomberg Magnificent 7 Index (Nvidia, Microsoft, Alphabet, Amazon, Apple, Meta, and Tesla) lost only 3.1% over that stretch, holding up far better than the broader technology complex and identifying where the selling concentrated: semiconductors and AI infrastructure rather than the mega-cap platforms. Size offered no shelter, as the Russell 2000 Index fell 3.85%, eroding part of a first half gain of 22.69%. International developed markets outperformed on a relative basis, with the MSCI World ex USA Index up 0.11%, while emerging markets absorbed the semiconductor shock directly and the MSCI EM Index tumbled 9.23%.

The Philadelphia Semiconductor Index, a gauge of the largest chipmakers, rallied 8% on July 30, lifting the Nasdaq 100 by 3.2% and both the S&P 500 and the MSCI World Index by 1.5%. Microsoft surged 17% in the session on strong cloud growth, and Oracle advanced on an expanded agreement to run Google’s Gemini models, gains that outweighed a disappointing forecast from Meta Platforms. The recovery extended through the final session, when Amazon jumped 15.3% after reporting its fastest quarterly cloud-unit growth since 2021 and guiding to roughly $220 billion of 2026 capital spending, a figure investors treated as evidence that the AI buildout will earn its cost. Apple was the conspicuous exception, falling 7.4% on a sales forecast weighed down by component shortages.

The S&P 500 finished July down 0.06%, recovering nearly the entire decline it carried into the last two sessions. The Nasdaq 100 closed the month off 6.59% and the Nasdaq Composite off 3.19%, each roughly 3.5% better than its July 29 mark. The Magnificent 7 swung from a 3.1% loss to a 2.07% gain, a move of more than 5% in two days, and the Russell Top 50 Index ended up 1.02%. The equal-weighted S&P 500 finished up 1.01%, still slightly ahead of the cap-weighted index for the month.

The value-to-growth spread compressed and remained wide. Russell 3000 Value closed up 3.64% and Russell 3000 Growth closed down 4.81%, an 8.45% gap against the eleven points that stood on July 29.

The Federal Open Market Committee left the target range at 3.50 to 3.75% on July 29, but the vote was 9 to 3, with three regional presidents (out of twelve) dissenting in favor of a quarter-point increase. Chair Kevin Warsh reiterated the absolute nature of the 2% objective while declining to guide markets toward the next move. Federal funds futures now imply better than a 70% probability of a hike in September, according to Bloomberg.

Duration, meaning the sensitivity of a bond’s price to a move in yields, bore the cost, and unlike equities it never got the bounce. Through July 29 the Bloomberg US Aggregate Bond Index had fallen 1.08%, the iShares 20+ Year Treasury Bond ETF had lost 3.77%, and US corporate high yield had edged down 0.42%. Long-dated Treasuries underperformed in the aftermath of the Federal Reserve (Fed) decision, carrying the 30-year to its highest yield since 2007 and the 10-year note above 4.7%, its highest since January 2025. The equity rally into month end pushed yields higher still, and the Aggregate finished down 1.30% while long Treasuries closed off 4.47%. High yield ended down only 0.25%, evidence that the move was a rates event rather than a growth scare.

That backdrop explains the sector table as much as the AI rotation does. Financials benefited directly from the steeper curve, while Utilities fell 2.18% and Industrials fell 2.91%, the classic pattern when long rates rise.

Two distinct momentum regimes have defined equity leadership over the past decade. The first ran from January 2017 to February 2020 and was led by the FAANGs: Facebook, now Meta, along with Amazon, Apple, Netflix and Google. The second was set in motion by the launch of ChatGPT in November 2022, which supplied both the catalyst and the narrative, though the wave itself is measured from the January 2023 low and has been led by the Magnificent Seven. The pandemic, and the rate shock that followed it, gave the first wave a clean end date. The second has no such marker, and June 2026 should be read as the peak in the current data rather than a confirmed top. What remains unresolved is whether AI capital expenditures will be monetized; i.e. turned into sustainable profits.

Both cohorts reached the top of the market for the same reason. They grew earnings far faster than the index and did so from business models the market judged defensible. The FAANGs owned distribution in search, social, retail and streaming, monetized attention and commerce at scale, and converted that position into cash flow with little incremental capital. The Magnificent Seven arrived with the same platform advantages and added a new claim, that the capital, data and distribution required to build artificial intelligence at scale were concentrated in their hands, which meant the incumbents would capture most of what the technology created rather than being disrupted by it.

With the AI complex selling off and the narrative under scrutiny, the FAANG comparison deserves closer examination. The resemblance is real, but it holds in the backward-looking features investors cite as reassurance and breaks in the forward-looking ones that will determine how the wave ends.

The analogy is real at the surface. Both waves began with a small cluster of platform businesses pulling away from the index during a period of accommodative financial conditions, and in both cases the outperformance rested on genuine earnings delivery rather than on multiple expansion alone, which is the point of separation from the dot-com episode.

The magnitudes are extraordinary in both cases and larger in the second. On an equal-weighted basis, the FAANG cohort compounded at roughly 28.2% annually across the first wave against 7.5% for the equal-weight S&P 500, an excess of about 20.7% a year. The Magnificent Seven compounded at roughly 47.4% annually from the January 2023 low through June 2026 against 14.5% for the equal-weight index, an excess closer to 32.9%. Amazon is the clean through-line, growing earnings per share 575% across the first wave and again at extraordinary rates from 2023 through 2026 as Amazon Web Services margins recovered and advertising scaled, according to Bloomberg.

Each wave also produced a single engine name whose returns dwarfed the cohort: Netflix at 190% in the first and Nvidia at 1,305% in the second. Neither was the largest or most profitable member of its group. Each carried a thesis that resolved to yes or no rather than to a rate of growth. For Netflix, whether international subscribers would scale fast enough to cover a content budget funded by debt. For Nvidia, whether the hardware and software stack constitutes a durable architectural monopoly or merely a favorable position in a capital spending cycle. Momentum regimes concentrate their most extreme outcomes in the name where the market is repricing a binary question, because each quarter of confirming evidence removes an entire branch of the outcome tree rather than adjusting an estimate.

The earnings behind the second wave remain real, and that is not a small point. Blended second-quarter earnings growth for the S&P 500 stands at 37.9%, according to FactSet, with Technology at 64.6% and Communication Services at 112.4%. Ten of eleven sectors are growing profits. Anyone reaching for the 1999 comparison has to explain away the income statements first.

The analogy breaks first on capital intensity, and this is the difference that should govern how investors read the current drawdown.

The first wave was capital-light. Apple’s capital expenditure fell from roughly 5.4% of sales to 2.7% while operating cash flow grew from $64.5 billion to $80.7 billion, the textbook profile of an asset-light compounder. Netflix was the exception, funding content through the bond market and running negative operating cash flow into 2019.

The second wave inverts that picture. Microsoft’s capital expenditure to sales ratio rose from 12.0% in fiscal 2022 to 22.9% in fiscal 2025, Alphabet’s from 11.1% to 22.7%, Meta’s from 27.0% to 34.7%, and Amazon’s from 12.4% to 18.4%. That is not software reinvesting at the margin. It is infrastructure-scale deployment closer in character to semiconductor fabrication plants or telecom buildouts, and it is still accelerating. Amazon raised full-year capital expenditure guidance to $220 billion from $200 billion, with most of the increase directed at artificial intelligence. The scale has reached the point of consuming operating cash flow within the quarter. Alphabet generated $73.3 billion in free cash flow across 2025, then reported negative $5.9 billion in the second quarter of 2026, its first negative quarter, on the capital spending surge.

Two consequences follow, and both are structural rather than cyclical. The first is timing. Returns on the first wave were observable in the current period, while returns on the second are underwritten on a multi-year lag. The second is flexibility. A company monetizing an installed base can slow spending when conditions deteriorate. A company defending a position through capital deployment cannot slow spending without conceding the position. That asymmetry explains why capital expenditure guidance has become the number investors watch most closely, and why Amazon’s July 31 guidance was received as reassurance rather than as a warning about free cash flow.

The FAANG names were recognizably the same kind of business, monetizing attention or commerce at scale, with cumulative earnings per share growth ranging from 53% at Apple to 575% at Amazon.

The Magnificent Seven is not a category so much as a label. Microsoft is a mature enterprise software incumbent whose capital intensity nearly doubled across the wave, yet its forward price-to-earnings multiple stayed essentially flat at roughly 23 times, and its total return of 99.5% came in at less than half the cohort median of 216.5%, set by Amazon at the midpoint. Tesla is a capital-intensive automaker that reached 195 times forward earnings at the June 2026 peak, with earnings per share down 53% in 2024 and a further 47% in 2025 on Bloomberg consensus, making it the most expensive and most fundamentally impaired name in the group.

Mag 7 Internal dispersion running from Nvidia’s 1,305% to Microsoft’s 99.5%, is wide enough that treating the seven as one analytical unit conceals more than it reveals.

Concentration compounds the problem. The group peaked near 30 to 33% of the S&P 500, roughly double the 15 to 17% the FAANGs reached in February 2020, with three names individually above 6% of the index. This is not diversification. The seven cannot be underwritten with a single thesis, yet they trade as a bloc, linked by passive flows, a shared narrative and, increasingly, by cash flows that run between them. Index investors hold a position that resists analysis as one unit and behaves as one anyway.

The FAANG claim was a description. These companies owned distribution and were monetizing it, and the analytical question was how long the monetization would compound.

The Magnificent Seven claim is a forecast. It holds that incumbents will capture most of what a technology creates, and it is made about a buildout that is not finished. Descriptions can be verified from a filing. Forecasts have to be underwritten, and their value moves with the market’s willingness to underwrite them.

That distinction explains the valuation behavior of July better than anything in the fundamentals. The forward twelve-month price-to-earnings ratio for the S&P 500 stood at 20.1 as of July 22, down from 20.4 at the June 30 quarter end, and remains above both its five-year average of 19.9 and its ten-year average of 19.0, according to FactSet. The compression came almost entirely through the earnings side of the ratio. Since June 30, the price of the index has declined by less than 0.1% while the forward twelve-month earnings estimate has risen 1.6%. Analysts project earnings growth of 27.3% for the third quarter and 24.9% for the fourth, per FactSet, and those two quarters are the cleaner read on underlying trend. Multiple compression against rising estimates is a valuation reset, not a growth scare. Investors did not lower their expectations for corporate profits. They declined to pay more for them, holding the index flat as estimates climbed. That is what happens to a forecast when confidence in it slips, and it does not happen to a description.

The first wave compounded through a period in which the discount rate applied to distant cash flows fell almost continuously. The second is being asked to compound through the opposite.

The 30-year Treasury yield reached 5.21% in July, the highest since 2007, while the 10-year settled near 4.68%, according to Bloomberg. With the long bond near 5.2%, the curve is delivering the tightening the Federal Reserve has declined to deliver, and long-dated real yields, rather than the policy rate, now set the ceiling on equity multiples.

Capital intensity and the rate regime compound each other, and this is the part of the analogy that should concern investors most. A wave built on asset-light monetization faces a rising discount rate on the value of its distant cash flows. A wave built on capital deployment faces that same rising discount rate and a rising cost of the capital required to get there. The same yield presses on both ends.

The month drew a line through the second wave that the first never had. The platforms with disclosed cloud revenue held up, with Microsoft gaining 24.6% for the month on cloud growth and Amazon gaining 13.9% on its fastest quarterly cloud-unit growth since 2021. The semiconductor and AI infrastructure complex absorbed the damage, falling roughly 21% over the month before its 8% single-session rally.

The earnings aggregates carry the same warning. Alphabet reported earnings per share of $9.11 against a $2.88 estimate, but the result rested on a $98 billion accounting gain on its investment holdings rather than on operating profit, and that company alone accounted for 92% of the net dollar increase in index earnings during the week ended July 24, according to FactSet. Excluding Alphabet, blended second-quarter growth measures 25.9% rather than 37.9%. The cohort is no longer a single position, and it is no longer a single quality of earnings.

The clearest evidence that the second wave may have entered a different phase sits in the returns rather than the narrative. Through July 31 the Bloomberg Magnificent 7 Index returned 0.30% year to date. The S&P 500 returned 10.12%, the equal-weighted S&P 500 13.24%, and the Russell 3000 Value Index 20.77%. Seven months is a short window, but the cohort has not led this market.

What the cohort has not done is give up index weight. The top ten S&P 500 constituents still account for approximately 40.6% of the index, above the roughly 30% peak of the dot-com era. On a concentration-adjusted basis the index now behaves as though it held roughly 45 stocks rather than the 120 of a decade ago. Breadth has improved substantially, with roughly 52% of members outperforming the index year to date, the strongest reading since 2016 and well above the 27% that outperformed in 2023, and roughly two-thirds of members higher on the year. Breadth and concentration are distinct measures, and 2026 has improved only one of them.

The current rout does not by itself settle whether the AI narrative was overbuilt. What it does is shift the burden of proof from price action back to cash flow. In the first wave, the market paid for businesses that converted growth into free cash flow while spending less to do it. In the second, it has paid for businesses spending more each year to defend a position whose returns are still largely prospective.

Three conclusions follow for portfolios:

Own the economics, not the label. The label describes what has already worked, and July rewarded disclosed cash generation over promised conversion. Internal dispersion from 1,305% to 99.5% within a single cohort is the argument against owning the cohort as such.

Treat long bond yields as the governing variable. The structural case for broadening does not require artificial intelligence to disappoint. It requires only that a 30-year Treasury bond yield at 5.21% continue to cap the multiples available to long-duration assets while ten of eleven sectors grow earnings and the average stock participates.

Recognize what remains unresolved. A 21% monthly decline in semiconductors followed by an 8% single-session rally, alongside an 18% swing in the Kospi, is not the behavior of a market that has settled on a view, especially when the reversal that closed July rests almost entirely on two companies.

The next several quarters of monetization disclosure, not the next several weeks of price, will determine which reading was right. The first wave was paid for by the cash flows it produced. The second is being paid for in advance.

July ended almost exactly where it began, and that flat close conceals the most consequential repositioning of the year. Value led growth, Energy and Financials led the market, and Technology gave background even after a sharp two-session recovery. The rotation narrowed at month end but did not reverse.

The more durable development was in rates. A divided Fed, three dissents favoring a hike, and a 30-year yield at its highest since 2008 reset the market’s assumption about policy direction. Bonds offered no offset, and the divergence between long Treasuries and high yield identifies the month accurately: a rates event, not a growth scare. Renewed supply disruption in crude complicates a disinflation trend that had only begun to look durable.

Our outlook remains constructive on the broadening trade. Provided that long-dated yields cap the multiples available to the most expensive parts of the market while the large majority of sectors continue to grow earnings, we favor value, cyclicals, and the average stock over concentrated growth exposure, and we would treat duration risk in fixed income with caution until the long end stabilizes. Investors should be careful not to read a two-day rally on cloud revenue as a resolution of the rotation that preceded it.

Sincerely,

The James Research Team