The Best Half Since 2021, And Almost None Of It Came From Big Tech

On June 30, the S&P 500 closed the first half of 2026 up 9.6%, the Nasdaq up roughly 12.8%, and the Dow up 8.9% at a record 52,317.^[1] The second quarter alone was the best three months for the S&P and the Nasdaq since the spring of 2020.^[1] Every recap you read this week will call it an AI rally and move on.

Here is the number those recaps leave out. The Magnificent Seven (Apple, Microsoft, Alphabet, Amazon, Nvidia, Meta, Tesla), which together are about a third of the S&P 500 by weight, returned 2.6% over the same six months.^[2] Microsoft was down 19.3%. Meta was down 11.7%. Tesla was down 12.5%.^[2] The seven companies everyone assumes are the market went sideways to down, and the market had its best first half since 2021 anyway.

That is not a footnote. That is the story. The rally broadened, and the index barely let you feel it, because the index is built to feel whatever the biggest seven names feel. Understanding why is the difference between reading H2 correctly and getting run over by it.

The rally was real, and it came from the bottom of the market

The Russell 2000, the small-cap index (companies roughly ranked 1,001 to 3,000 by size in the US market), gained about 22% in the first half. That is its best first half since 1991.^[3] Chip-related names accounted for 16 of its 50 best performers, and several of them, Aehr Test Systems, Ichor, MaxLinear, more than quintupled.^[3] Consensus 2026 earnings growth for the Russell 2000 climbed to 38%, up from about 23% at the start of the year.^[3]

So the AI trade did drive the rally. It just stopped paying the landlords and started paying the contractors. The money went to memory, test equipment, power, cooling, networking, the boring physical layer that gets bought when someone else has already committed to spending $725 billion. Which somebody has.

Amazon, Microsoft, Alphabet, and Meta are guiding to roughly $725 billion of capital expenditure (capex, the money a company spends on physical assets like data centers and chips) in 2026, against about $410 billion in 2025.^[4] A 77% increase. Amazon alone is near $200 billion.^[4]Every one of those dollars is somebody else's revenue. That is the actual mechanism behind small-cap outperformance, and it is a lot more durable than a sentiment story.

The mega-caps got repriced, not rewarded

Owning Microsoft in H1 2026 meant losing a fifth of your money while the index made you nearly ten percent. Read that twice, because it inverts the entire mental model most people have carried since 2023.

The reason is not that AI stopped working. It is that the market finally started pricing capex like capex. When you spend $150 billion on assets that depreciate over five to six years, you book roughly $17 to $20 billion a year in depreciation before you have earned a dollar back.^[4] Free cash flow compresses even when revenue is growing. The gap between what the hyperscalers are spending on AI infrastructure and what the AI ecosystem is actually selling runs to roughly $600 billion a year, a capex-to-revenue divergence that already exceeds the 2001 telecom bubble on the same measure.^[5] Meta fell 9.25% on the day it raised capex guidance this spring.^[5] That is a market saying “we do not believe the return on that yet,” out loud, with real money.

I do not read this as the AI trade breaking. I read it as the AI trade maturing. The first phase rewarded anybody who announced a number. The second phase asks who converts the number into cash. Nvidia, up only 4.5% on the half, is the tell.^[2] The single most obvious beneficiary of a $725 billion buildout barely moved, because at a $5.2 trillion market cap and about 8% of the S&P 500 by itself, the good news was already inside the price.^[6]

What the Fed actually did for you

The rate story in H1 was noisier than the index return suggests. The year started with an energy shock out of the Iran conflict, which fed straight into headline inflation and had traders pricing hikes, not cuts. Then the war moved toward resolution and the shock unwound.

Fed Chair Kevin Warsh, speaking at the ECB's Sintra forum on July 1, said that “expectations of inflation over the first four weeks of this period have come down, inflation risks have come down,” crediting energy prices that had fallen “quite substantially” since the US-Iran memorandum of understanding.^[7] The June CPI print backed him up: headline down 0.4% month over month, the biggest drop in six years, with year-over-year headline at 3.5% against 4.2% in May, and core at 2.6%.^[8]

This matters more for small caps than for anything else you own. Small caps carry more floating-rate debt, so they are the most rate-sensitive thing in the market. Bank of America estimates every additional 25 basis points of hike knocks about 2% off Russell 2000 operating earnings.^[3] The best-performing index of the half is also the one with the most to lose if Warsh turns out to be patient rather than dovish.

The concentration problem nobody wants to talk about

Here is the part that should actually change how you are positioned.

The Magnificent Seven are about 33 to 34% of the S&P 500 and carry a combined market cap north of $22 trillion.^[6] Add JPMorgan, Broadcom, and Berkshire and the top ten holdings are close to 40% of the index.^[6] Historically the top ten averaged around 24% of the S&P, and the prior modern peak was 28% back in 1970.^[6] Nvidia alone, at roughly $5.2 trillion, is about 8% of the whole thing.^[6]

Most people hear “concentration risk” and picture a crash. I think that framing is lazy. H1 2026 just showed you the real version, and it is quieter and more insidious. Concentration did not blow up your portfolio. It taxed it. A third of your index went up 2.6% while the other two-thirds did the actual work. The equal-weight S&P beat the cap-weighted S&P by several percentage points on the year for exactly that reason.^[10]

The asymmetry is what gets you. When the mega-caps lead, the index captures it and the equal-weight lags. When breadth leads, the index only partially captures it, because a third of the weight is sitting out. And if the mega-caps actually break, on a capex writedown, a depreciation reset, an antitrust ruling, breadth cannot save you, because a 33% block falling 25% is an 8-point hole in the index that the other 493 have to dig you out of.

What is actually fragile going into H2

Three things, in the order I worry about them.

1. The capex-to-revenue gap. Roughly $600 billion a year separates AI infrastructure spending from AI ecosystem revenue, and the divergence is already wider than the 2001 telecom peak on the same ratio.^[5]An MIT study found 95% of enterprise generative-AI pilots showed no measurable P&L impact.^[5] The buildout does not need to be wrong for this to hurt. It only needs to be early, because depreciation starts on schedule whether the revenue does or not.
2. Small caps are a rates bet in disguise. A 38% consensus earnings growth forecast is a very high bar.^[3] If the Fed stays put, or the energy-led disinflation reverses, the most crowded winner of H1 has the most floating-rate debt and the least margin for error.^[3]
3. The multiple.The S&P's forward P/E of about 20.5 sits above its five-year average of 19.9 and its ten-year average of 19.0.^[11] Not a bubble. Not cheap either. It means H2 returns have to come from earnings, because there is no room left for the multiple to do the lifting.

How I would actually be positioned

I am not going to pretend to know what the next six months do. Nobody does. But the structural read is clean enough to act on.

Pair the cap-weighted index with something equal-weighted. Not because equal-weight is magic, it lagged badly for three straight years, but because it is the cheapest available hedge against the specific risk you are carrying without having chosen to. Cap-weighted trades near a 25 to 27x P/E on some measures. Equal-weight trades closer to 22x.^[10] You are paying up for the concentration.

Own the picks-and-shovels layer, but own it knowing what it is. Memory, test, power, and networking companies are pure derivatives of hyperscaler capex. If the $725 billion becomes $500 billion, those names do not fall 30%, they fall 60%. Position size accordingly. A 400% move in a small-cap chip supplier is not a valuation, it is a bet on a spending line item staying up.^[3]

And be honest about the mega-caps. A 19% drawdown in Microsoft while it grows AI revenue at a 123% run rate is not obviously a broken business.^[5] It is the market demanding proof of return. Sometimes the market is right. Sometimes it hands you the best entry it has offered in three years. The way I think about it, if you believe the capex is productive, the mega-caps just got cheaper while the thing they are building got more expensive. That is an odd place to be selling.

The best half since 2021 was not the market telling you AI won. It was the market telling you it has started grading on cash instead of narrative. That is a healthier market. It is also a harder one, because the index will no longer do your thinking for you. It never really did. It was just easier to believe when seven stocks went up every year.