Camtek: Selling the Gates on the AI Packaging Toll Road

Camtek: Selling the Gates on the AI Packaging Toll Road

The prevailing narrative on Camtek is the toll-road story. The logic is seductive: as AI hardware shifts more of its value into advanced packaging — the processes that stack memory dies and stitch multiple chips into one package — every one of those structures must be inspected for defects before it can be trusted in a data center, and Camtek's inspection and metrology systems catch them. Complexity keeps rising, so inspection content keeps rising, so revenue compounds. A secular toll on the packaging highway. No cycle attached.

The direction is right. The mechanism is wrong, and the difference determines whether this stock holds a 40-times forward multiple. Camtek does not meter packages as they roll off the line. It sells the tools that chipmakers, memory suppliers, and OSATs — the outsourced assembly-and-test houses — purchase when they add packaging capacity. That is a fundamental distinction: a toll road collects every year on whatever traffic shows up, while Camtek books revenue once, when a customer buys another gate for a new line. The company's revenue is a function of packaging capacity added, not packaged volume shipped. That makes it a capacity-cycle franchise with a genuinely rising toll rate — not a metered annuity.

The Q2 Shape Is the Proof

The second-quarter print was a record, and at the same time a demonstration of exactly that distinction. Camtek reported $133.2 million of revenue, up 10% sequentially and 8% year over year, beat its own guidance, and told investors to expect $158–160 million in the third quarter and more than 30% growth in the second half versus the first.

Now look at the shape of 2026, not the level. The first half grew roughly 5% year over year, and the March quarter barely grew at all. Management described the opening months as a lag: the business "lagged at the beginning of the cycle" before order flow "came in to the full degree." A toll road does not lag. A tool franchise waiting on the timing of customers' capacity schedules does. When the orders landed, they landed in a batch. That is the lumpiness the equipment business keeps rediscovering, and it is the reason the "secular" label is doing more work than it should.

The numbers tell the whole story. CoWoS capacity, TSMC's 2.5D packaging line that mounts memory and logic chips side by side on a silicon base, is fully booked with lead times of 52 to 78 weeks even as the company raises monthly output toward 120,000–130,000 wafers by the end of 2026. Annual CoWoS demand has nearly tripled in two years, and HBM3E memory is sold out for the year. One estimate puts TSMC's 2026 capital spending at $52–56 billion, with 10% to 20% earmarked for advanced packaging.

Every one of those capacity increments is a new batch of inspection gates. The composition of the order book shows where the money is coming from: roughly 80% of year-to-date orders are for advanced packaging, more than half from OSATs and over a fifth from HBM makers gearing up HBM4. One leading OSAT alone placed a $31 million multi-system order in January for CoWoS-like packaging, part of more than $90 million of first-quarter OSAT bookings.

The Toll Rate Is the Structural Story

This is where the metaphor earns its keep, once corrected. The structural uplift is not in the number of gates; it is in the value of each gate, because the inspection content per unit of packaging capacity is rising. Every HBM stack generation packs in more layers, package sizes have grown, and hybrid bonding — the chip-to-chip interconnect method moving into advanced packaging — creates measurement steps that never existed on prior-generation tools. Silicon photonics, roughly 5% of year-to-date orders and expected to exceed that by 2027, is a new toll lane entirely.

The evidence is inside the product mix. The two newest platform families, Eagle G5 and Hawk, already account for about half of system revenue, with Hawk aimed at the highest-volume HBM applications. Higher-value tools replacing the old base is precisely the mechanism that lets a packaging-capacity business outgrow the wafer-fab equipment market, and management explicitly expects 2026–2027 performance in line with or better than overall wafer-fab equipment spend. Published market research sizes the advanced-packaging metrology and inspection equipment market at roughly $4.5 billion at last count, growing at a double-digit clip — a small pond by equipment standards, and the reason the toll-rate story matters more than the market-size story.

Two Markets, Not One

Any serious read has to separate the two markets inside the revenue line. Roughly 75% of second-quarter revenue came from advanced packaging tied to AI — the market where the constraint, the pricing power, and the new toll lanes all live. The balance is mainstream packaging, including China, which is expected to represent about 45% of 2026 revenue.

That China concentration cuts both ways. It is stable, slow-growing revenue that played no part in the AI re-acceleration thesis, and under the current export-control regime it is a policy variable no equipment supplier controls. It is also why the back half of 2026 can grow more than 30% while the full year lands near $586 million — only about 18% above 2025. The AI toll lanes are the growth; China is the ballast on the other end of the scale.

The Margin Gave Back the Beat

Then there is the reason the stock fell on a beat-and-raise. Camtek's shares, per Ainvest market data, sat at about $147 after the print — up 38% year to date and roughly 79% over the trailing twelve months, but already down about 14% over the prior four months and roughly a third off a 52-week high near $216, and still slipping, down roughly 11% over the past five sessions. The intraday reversal said investors were demanding more than accelerating sales.

The income statement explains the mood. Gross margin held steady near 51%, but operating leverage went into reverse: non-GAAP operating margin fell to 27.0% from 30.3% a year earlier, and on a GAAP basis it dropped to 20.4% from 25.9%, with GAAP net income down about a third as spending — chiefly research and development tied to the Visual Layer acquisition and AI investments — grew faster than the top line. Management targets 30–32% operating margin by year-end with revenue re-accelerating into 2027. At roughly 13 times trailing sales, a price-to-sales multiple similar to metrology rival Nova's, the stock's roughly 40–45 times expected 2026 non-GAAP earnings assumes that margin recapture actually arrives.

Investor Takeaway

The correct description of Camtek is a gate-seller at the single most constrained point of the AI supply chain, not a toll collector on a perpetual highway. Both parts of that sentence carry weight. Being the dominant inspection franchise where CoWoS and HBM capacity are sold out is a strong structural position, and the rising toll rate — more inspection value per capacity increment, plus new lanes in hybrid bonding and photonics — is real. But the revenue line is capex-driven and lumpy by construction, and 2026 is a reconcentration year: half the growth lands in the second half, margins are compressed at the start of the climb, and 45% of the book sits in China.

The key question is not whether AI packaging demand stays healthy. It is, and it is booked. The more important question is whether the suppliers hold the line — whether CoWoS and HBM capacity stay sold out through 2026 so the order backlog converts into 2027 revenue, whether operating margin actually returns toward 30% by the fourth quarter, and whether the new toll lanes grow from 5% of orders into something that moves the income statement. If those conditions hold, the multiple has a floor. If the margin line stalls, no amount of packaging complexity will support a gate-seller priced like a toll booth.