AI Stocks Sold Off in July. These 6 Are Where the Buildout Data Still Says Buy

By AltIndex Research · 19 min read · July 25, 7:45 am

AI infrastructure stocks have had an extraordinary run, but the last month has taken a lot of it back. Every one of the six names below is trading below its 2026 high, several by more than 20%, even as the spending that drives their revenue keeps arriving. The AI 2027 scenario predicted this buildout would define the market. If that call is right, this pullback is an entry point rather than a top.

In April 2025, a group of forecasters led by former OpenAI researcher Daniel Kokotajlo published AI 2027, a scenario with a specific market call attached: by late 2026, the stock market would be up 30% on the year on the back of the AI buildout.

Six months still remain. The index half is well behind pace, with the S&P 500 entering July up roughly 9%. The spending half has arrived on schedule: the largest cloud infrastructure companies are on track for around $670 billion of capital expenditure in 2026.

That money flows from a handful of hyperscalers and AI labs into a much smaller group of suppliers: the companies selling the power, chips, cooling, networking and test equipment every new datacenter needs. For Microsoft or Amazon, AI capex is one line in an enormous business. For these suppliers it is close to the whole business, which is why it shows up in their hiring and headcount data first.

We screened over 3,000 companies for the AI breakout stories where hiring, headcount, employee outlook, etc are still accelerating. Six stocks cleared, all carrying a buy signal according to our alternative data analysis. If the AI 2027 story holds, each of them might still be a buying opportunity.

Six Names, Six Layers of the Buildout

Our AI Score runs from 0 to 100 and blends four families of signal: company fundamentals, customer activity, employment data and public attention. Anything above 60 reads as a buy signal, 40 to 59 is a hold, and below 40 is a sell. All six of these clear 60, and one of them is currently leading our Top Stocks list.

How Much Is Already Priced In

Five of the six are up in 2026 and the spread between them is enormous. Marvell has gained 125.3% since the end of 2025. Teradyne is up 76.1%, Vertiv 73.3%, TSMC 33.9%. Nvidia has managed 8.7%, barely ahead of the index. Constellation is the only one down, at 23.7% below where it started the year.

That spread is the risk map. Marvell has more than doubled and trades at 71.5 times earnings, which means the buildout has to keep running for years before that multiple makes sense. Vertiv at 75.8 times and Teradyne at 69.5 times carry the same problem in smaller size. Nvidia at 31.8 times has barely participated in 2026, which makes it the lowest-beta way into the theme. Constellation is the only one the market is treating as though the buildout will not happen.

Every one of the six is also below its 2026 high: Marvell by 35.6%, Vertiv by 21.6%, Teradyne by 20.7%, Constellation by 15.1%, TSMC by 10.6%, Nvidia by 8.1%. A pullback across an entire theme is either the entry or the first crack. The sections below examine each stock's alternative data to determine whether it still has room to grow.

Constellation Energy: Revenue Beat by 28%, Stock Down 24%

Constellation carries an AI Score of 85, and it got there while the stock fell. The stock is down 23.7% in 2026 and sits 15.1% below its high for the year.

The workforce went the other direction. LinkedIn headcount has climbed from 10,145 in August 2025 to 13,414 in July 2026, a 32% increase. Most of that arrived at once, between March and April 2026, when the count went from 10,675 to 13,061 in a single month. A jump that size reflects a corporate action rather than organic hiring, so the trend worth reading is the steady climb on either side of it. Job postings sit at 318 and rose 6.5% month over month.

The signal we weight most heavily here is employee business outlook, the share of employees who expect business conditions at their employer to improve. At Constellation it has risen in eleven of the last twelve months, from 75% to 87%. Employees rarely turn more optimistic about their employer's prospects while the business deteriorates. They see the contract pipeline before investors read about it.

AltIndex employee business outlook vs share price, August 2025 to July 2026

The fundamentals have already turned. Q1 2026, reported May 11, brought $11.12 billion of operating revenue against $6.79 billion a year earlier, a 64% jump, with net income to common shareholders of $1.59 billion versus $118 million. Adjusted operating EPS was $2.74 against a $2.59 consensus. Management affirmed full-year guidance of $11.00 to $12.00 and set a target of more than 20% annual earnings growth through 2029.

Both step changes trace to one deal. Constellation closed its $21.8 billion acquisition of Calpine on January 7, 2026, adding 23 gigawatts and taking the fleet to roughly 55 gigawatts. That is acquired scale, not organic growth, and it explains the revenue and the headcount jump together. What it does not explain is the share price. The stock trades at 23.9 times earnings, the cheapest multiple of the six, on a nuclear fleet running at a 92.3% capacity factor with roughly 5,000 MW submitted to PJM's interconnection queue, and it is down 23.7% on the year.

Three congressional purchases landed in the last four months: John Boozman bought on April 2 and again on June 12, and Mark Warner bought on April 13. Repeat buying by the same member is the pattern worth noticing, not the dollar size. Earnings land August 6.

TSMC: Job Postings Just Hit a 12-Month High

Whoever wins the AI race, the chips get made in Taiwan. TSMC scores 79, and its alternative data is still accelerating.

Job postings reached 613 in July, up 16.7% month over month and the highest reading in twelve months, against 402 a year ago. Headcount grew from 23,309 to 26,128 over the same window, up 12%. The share of employees with a positive business outlook moved from 61% to 72%.

The financials are the strongest of the group. Q2 2026, reported July 16, brought $40.20 billion of revenue, up 33.7% year over year in dollar terms, with net income of NT$706.56 billion, up 77.4%, and a 67.7% gross margin. Both lines beat consensus. TSMC guided Q3 to $44.6 to $45.8 billion and lifted full-year revenue growth expectations to slightly above 40%.

The number that matters most for everything else on this list is the capex line. TSMC raised its own 2026 capital expenditure range to $60 to $64 billion from $52 to $56 billion, and committed a further $100 billion to Arizona, taking its total there to $265 billion. A foundry does not raise capex by $8 billion mid-year on a demand signal it expects to fade.

Two caveats sit in the data. The employee rating of 66 is the lowest of the six companies here, a reminder that a fab is a demanding place to work even when the order book is full. And at 37.1 times earnings the stock is no longer cheap, up 33.9% in 2026 and roughly 70% over twelve months.

Vertiv: Hiring Nearly Doubled in Twelve Months

Vertiv sells the power distribution, cooling and thermal management that turns a warehouse into a datacenter. It scores 79, and its hiring data is the most aggressive in the group.

Job postings went from 926 in July 2025 to 1,820 in July 2026, a 97% increase, peaking at 1,942 in June. Headcount rose 17%, from 15,866 to 18,563. The share of employees with a positive business outlook improved from 70% to 77%. Stock sentiment sits at 91.1 (bullish) and rose 5.6% month over month.

AltIndex job postings data, monthly average, August 2025 to July 2026

Q1 2026, reported April 22, brought $2.65 billion of net sales, up 30% year over year, though only 23 points of that were organic, with 4 from acquisitions and 3 from currency. Adjusted EPS of $1.17 beat the $1.01 estimate and rose 83%. The Americas drove it, growing 44% organically. Vertiv raised full-year adjusted EPS guidance to $6.30 to $6.40 and guided Q2 to $3.25 to $3.45 billion.

At 75.8 times earnings this is the most expensive name on the list, and it fell 4.5% in the session. Earnings on July 29 are the near-term test of whether the hiring converts.

Teradyne: The Least Crowded Story, The Biggest Revenue Jump

Teradyne makes the test equipment every accelerator and memory die passes through before it ships. It is a volume play that does not require picking an architecture winner, and it scores 77.

Job postings tripled, from 118 a year ago to 396 today, a 235% increase with a May peak of 479. Headcount grew 12% to 6,210. Web traffic rose from 144,147 monthly visits to 200,065, up 39%, having peaked at 243,299 in March. The share of employees with a positive business outlook improved from 75% to 79%, and stock sentiment reads 96.6 (very bullish), the highest of the six.

Q1 2026, reported April 28, was a company record. Revenue of $1.282 billion against $686 million a year earlier, an 87% increase, with net income of $398.9 million, non-GAAP EPS of $2.56 against roughly $2.11 expected, and gross margin of 60.9%. Semiconductor test passed $1 billion for the first time.

Management put AI-related demand at nearly 70% of revenue, up from about 60% the prior quarter. That is the cleanest read on this list of how fast the buildout is concentrating. Teradyne guided the full year to roughly $6 billion of revenue and $9.50 to $11.00 of non-GAAP EPS. The stock fell sharply after that print regardless, which is what a high bar looks like.

Teradyne reports again on July 28, the first of these six to test whether the buildout is still accelerating.

Marvell: Raised Guidance Twice, Stock Down 36% From Its High

Marvell designs custom silicon and high-speed interconnect for data centers, the chips that move data between accelerators rather than doing the computing themselves. With an AI score of 78, it's the name on this list where the reported numbers and the share price have come apart most violently.

The positives are real. Job postings nearly doubled year over year, from 143 to 283. Web traffic rose 48%, from 394,194 monthly visits to 582,078. The share of employees with a positive business outlook climbed from 84% to 91%, second only to Nvidia among the six, and the employee rating sits at 88.

The market signals point the other way. Stock sentiment has slid from 96.3 (very bullish) a year ago to 70.3 (close to neutral) today, the weakest of the six. The stock is 35.6% below its 2026 high, the deepest drawdown of the six, and fell 7.2% in the latest session.

The fundamentals have not cracked at all. Q1 fiscal 2027, reported May 27, delivered record revenue of $2.418 billion, up 28% year over year, with data center at $1.833 billion and 76% of the total. Non-GAAP EPS of $0.80 beat the $0.75 estimate, and Marvell raised its revenue outlook for both fiscal 2027 and fiscal 2028, guiding Q2 to $2.7 billion.

So the price fell 35.6% from its June high while the company was raising two years of guidance. Marvell also closed two acquisitions in February, Celestial AI and XConn, which helps explain the hiring surge and complicates reading the July slowdown as a demand signal. This is the widest gap on the list between what the company reported and what the market did with it. The company reports earnings in late August.

Nvidia: Revenue Up 85%, the Cheapest Multiple of the Six Growth Names

Nvidia has an AltIndex AI score of 66, a buy signal, and it is the value entry of this group. At 31.8 times earnings it trades cheaper than Vertiv, Teradyne and Marvell, on a business that in the quarter ended April 26, 2026 produced record revenue of $81.6 billion, up 85% year over year, with GAAP net income of $58.3 billion and a 75.0% gross margin. Non-GAAP EPS of $1.87 beat the $1.77 estimate. Management guided the current quarter to $91 billion, raised the dividend from a penny to 25 cents, and authorised another $80 billion of buybacks, all while assuming no data center revenue from China.

The internal data is the best on this list. The share of employees with a positive business outlook stands at 95%, and headcount (according to LinkedIn data) grew 21% over twelve months.

What keeps the score below the others is the forward-looking hiring. Job postings peaked at 3,032 in November 2025 and have fallen to 2,456, down 10.9% month over month. A company adding headcount while cutting open roles is completing a hiring cycle rather than starting one, and we logged an expansion or downsize event on July 15.

AltIndex headcount vs job postings, August 2025 to July 2026

The demand-side signals are flat alongside it. Stock sentiment reads 74.96 (close to neutral), the lowest of the six and down from 79.3 a year ago. Web traffic has held between 28.7 million and 39.4 million monthly visits since August 2025 with no direction, most recently 37.4 million, while headcount grew 21% over the same window.

The business is plainly not the constraint. The score reflects a stock where the alternative data has flattened while the reported numbers accelerate, which is a different setup from the names above it. Read one way that is a warning the hiring cycle has topped. Read another, it is the largest company in the buildout growing 85% and trading at half the multiple of its suppliers. Next earnings land in late August.

What Could Go Wrong

The AI Score measures signal direction, not valuation. Three of these six trade above 65 times earnings, and alternative data confirming that a buildout is happening tells you nothing about whether it is correctly priced.

The deeper risk is the capex itself. That $670 billion of 2026 cloud infrastructure spending is equivalent to more than 90% of those companies' expected cash flows, and every business on this list sits downstream of it. AI 2027's spending forecast has tracked well through the first half of the year, and TSMC's mid-year capex raise says the suppliers believe it, but the scenario also assumes those datacenters eventually earn a return. If 2027 disappoints on that question, the spending stops at the source and all six reprice together, which is roughly what their synchronised pullback from 2026 highs is already testing.

Marvell's headcount series on our platform is unreliable at present and we excluded it from the analysis above rather than estimate around it.

The Bottom Line

These six do not carry the same risk, and the difference is what each one has already done this year. Marvell up 125.3%, Teradyne up 76.1% and Vertiv up 73.3% have priced in a great deal of the buildout already. At 71.5, 69.5 and 75.8 times earnings, those three carry execution risk: they have to keep beating quarter after quarter, and Teradyne showed what happens when they do and the market wants more anyway.

Constellation down 23.7% on the year and Nvidia up just 8.7% have not had that run. Both trade at multiples the others cannot justify, 23.9 and 31.8 times, on businesses sitting at the same chokepoints. If you believe the buildout continues, those two are the cheaper way in.

This distinction only matters if the buildout slows. Should AI 2027 keep tracking on the spending side through 2027, all six are positioned to run, and the ones that have already tripled may keep going. The pullback across the whole group is the market asking that question, and none of these six has answered it yet.

Three of them report inside two weeks: Teradyne on July 28, Vertiv on July 29, Constellation on August 6.

Sources: AltIndex job postings data, AltIndex LinkedIn employee data, AltIndex employee business outlook data, AltIndex web traffic data, AltIndex social sentiment data, AltIndex congressional trading data, company quarterly earnings reports and SEC filings. All financial figures were verified against company press releases and SEC filings rather than taken from AltIndex alone. Market data as of July 24, 2026. Year-to-date price changes are measured from the final 2025 weekly close in AltIndex data; drawdowns are measured from each stock's highest 2026 weekly close. Stock prices are historical and for informational purposes only. This article does not constitute investment advice.

Disclosure: This article is for informational purposes only and does not constitute investment advice. AltIndex aggregates publicly available alternative data signals. Past signal performance does not guarantee future results. Always do your own due diligence before making investment decisions.