Q2 2026: large caps plateau, the broad market climbs:

Q2 closed with 498 of 500 S&P 500 constituents reported. The median constituent grew GAAP net income 14.2% year over year on aggregate revenue growth of 14.4%, with 86% beating estimates and 69% of reporters improving on the prior year. S&P 500 Growth led every measure — a 27.1% median with 84% participation — while the Russell 3000's median of 17.2% sat above the S&P's on the weakest participation of the three at 64%.

Set against the last four quarters, the three universes are doing different things. The S&P 500 and Growth share a shape: a Q4 2025 trough, a Q1 2026 peak, and a modest easing in Q2. The S&P's median ran 9.2%, 5.9%, 15.9%, then 14.2%; Growth ran 21.0%, 22.2%, 28.2%, then 27.1%. Both are consolidating just below a high, and both finished Q2 within roughly a point of their own Q1 pace — a difference too small to read as directional.

The Russell 3000 is the exception and the quarter's most useful signal. Its median has risen from 10.8% to 17.2% since Q3 2025, and its participation has risen from 59% to 64%, improving again in Q2 while large caps eased — crossing above the S&P 500's median for the first time in the series. The broad market has not yet found its ceiling. But the 64% figure deserves emphasis: more than a third of Russell constituents still went backward. The recovery is real in the middle of the distribution and absent in the tail — a stronger middle carried by fewer companies.

The season's pattern: large caps continue to earn more growth than the broad market on an absolute basis, but the broad market is improving faster against its own base. The rally's earnings foundation is widening, not rotating.

Main Themes From the Earnings Season:

Over nine weeks of this earnings season, the same handful of forces kept appearing. I believe five themes ran through this season, and I have explained them below –

1. Physical Constraints on the AI Buildout

This appeared in eight of nine weeks, and each week it moved one layer deeper into the economy. Micron set the background in June: high-bandwidth memory sold out, with buyers signing three-to-five-year supply agreements, which is unusual for what has historically been a spot-market commodity. In July, Elon Musk thanked Micron by name on Tesla’s call for an allocation. In August, Amazon raised its capital budget by $20 billion and named memory cost as the reason; Caterpillar reported a $72 billion backlog with orders into 2030; and Arista said the industry will not clear the shortage until 2028. Cisco then put numbers on it: memory-driven price increases added about five points to its revenue growth while cutting 210 basis points from gross margin. By September, NVIDIA guided its own margin down from 75% to 74% on memory costs, and Oracle disclosed that customers renewing GPU contracts pay about a 20% premium. In short, the constraint is physical supply, not money, and physical shortages take years to relieve rather than quarters.
2. Distorted Headline Earnings

Almost every week produced a headline number that needed adjusting. Hyperscalers marked up their private AI-lab stakes through the income statement: Alphabet’s $9.11 came mostly from SpaceX, and Salesforce’s $5.90 included $2.53 from Anthropic. Merck and Gilead posted GAAP losses on acquisition charges while beating on an adjusted basis. Then tariff refunds arrived after the Supreme Court struck down the IEEPA tariffs: Walmart booked $2.9 billion, Target $994 million, Ross $253 million. What each retailer did with the money mattered more than the refund itself. Walmart spent its refund on more than 11,000 price cuts and fell 9%. Target let it reach the bottom line and rose 4%. Academy Sports excluded it from adjusted earnings entirely and rose 14%. I think investors are now clearly reading past the headline to find the operating number.

3. Uneven AI Impact on Software

In early August, the market was selling all software on the assumption that AI agents would replace it. Datadog beat and fell 17%. By late August, it had changed its mind selectively. Salesforce rose 23% and CrowdStrike 21% on AI-linked results. MongoDB reported that coding agents are signing up for its database directly, making them customers rather than replacements. Palo Alto’s AI security products passed $100 million in annual recurring revenue within four quarters. Meanwhile, Intuit, whose products automate routine consumer and small-business work, guided next year’s revenue $1.2 billion below expectations. In short, AI is separating software rather than destroying it: companies that own proprietary data are being paid by the machines, and companies that automate routine tasks are being replaced by them.
4. Widening Consumer Split by Income

The split itself is not new. What changed this season is that companies began disclosing which half of their customers stopped spending. Walmart’s US comparable sales grew 2.6%, its weakest in about six years, with traffic slowing to 1.5% from 3%. Academy Sports reported that traffic from households under $50,000 lagged behind households above $100,000. Casey’s private-label snacks grew 16% while comparable national brands fell 8%, and Kroger is expanding its private label from about 130 items toward 1,000. That last point has a second-order effect: when a grocer shifts shelf space to its own label, the margin moves from the packaged-food company to the retailer. If it continues, it shows up as a volume problem in consumer staples earnings over the next few quarters.
5. Fading Scrutiny of AI Capital Spending

This is the one theme that disappeared, and I think it is the most important open question going into next quarter. From mid-July to mid-August, the market was auditing AI spending closely: Alphabet fell 7% for raising its capital budget, and AMD fell 9% on a beat-and-raise. By late August, it stopped. Dell rose 16%, and Snowflake 17% on strong results, and Broadcom and Dell published targets for fiscal years that have not started. The visibility is genuinely better, with Amazon’s AI capacity contracted on five-year terms and Microsoft’s backlog at $627 billion. But NVIDIA’s free cash flow fell to $21.3 billion from $48.6 billion while it raised $24.9 billion of debt and arranged $500 billion of third-party financing so customers can borrow to buy compute. When a buildout moves onto credit, the risk changes from whether demand appears to whether the financing holds.
What I Am Watching Next –

1. Micron reports on 30 September. It set the background for this season in June and is the cleanest read on whether the memory shortage is still tightening.

2. The software cluster reports in late October. The split between data owners and routine-work automation should start showing up in guidance, not just in share prices.

3. Megacap capital spending guidance in late October will show whether scrutiny of AI spending returns. I would watch free cash flow and debt issuance more closely than the capex number itself.