2026-06-08 — views $CAMT · Camtek · Hawk inspection systems · Advanced-packaging & HBM inspection/metrology

Camtek's $105M inspection order, all 2027 delivery: how AI packaging and HBM bookings now run a year ahead of the fab

Camtek booked over $105M in inspection orders on June 2, 2026 — $55M from a tier-1 OSAT and $50M+ of Hawk systems for HBM — all slated for 2027 delivery. A look at why back-end metrology bookings now lead the cycle.

What shipped

On June 2, 2026, Camtek — the Israeli inspection-and-metrology specialist that lives almost entirely in the back end of the fab — said it had received more than $105 million in multi-system orders. The release splits cleanly into two buckets: a $55 million multi-system order from a tier-1 OSAT (outsourced assembly and test provider) for 2.5D and 3D AI-device inspection, and over $50 million — all of it for Hawk systems — from “a leading HBM player” for AI-related applications.

The detail that matters most is buried in one sentence: “All these orders are expected to be delivered in 2027.” This is not a backlog Camtek will recognize this year. It is revenue the company is booking roughly a year and a half before it ships. That timing is the real story, and it tells you something about how the inspection layer of the supply chain is now being scheduled.

Order	Customer type	Stated value	Systems	Delivery
AI device inspection	Tier-1 OSAT	$55M	Multi-system (2.5D/3D)	2027
HBM inspection	”Leading HBM player”	Over $50M	All Hawk	2027
Total	—	>$105M	—	2027

For scale: Camtek reported $121.7 million in Q1 2026 revenue (up ~2.5% year over year) and has guided to over 25% revenue growth in the second half of 2026 versus the first half. A single press release announcing $105M+ of orders is therefore close to a full quarter of revenue landing in one day — and it lands in next year’s bucket. CEO Rafi Amit framed it as “the exciting momentum that we have experienced since the beginning of the year.”

Why two customers, two different jobs

It is easy to lump these together as “AI orders,” but the two halves point at different process problems.

The OSAT $55M is about heterogeneous integration: 2.5D interposers and 3D stacks where dozens of dies, bumps, and through-silicon connections have to be inspected at multiple steps. As more of the value of an “AI chip” moves out of the transistor and into the package, the OSAT becomes the place where yield is won or lost — and inspection coverage there scales with the number of bonding and stacking steps, not with wafer starts.

The HBM $50M+, all-Hawk order is narrower and arguably more telling. HBM stacks (8-Hi, 12-Hi, and the 16-Hi parts memory makers are now racing toward) are punishingly sensitive to particle and bump defects because a single bad die can scrap an entire expensive stack. Concentrating the whole order on one product line — Hawk — suggests the customer has standardized on a specific inspection recipe and is buying capacity, not evaluating tools. That is what a ramp looks like, versus a pilot.

The macro backdrop: back-end is outgrowing the front end

Camtek’s order doesn’t sit in isolation. In its December 16, 2025 forecast, SEMI projected total semiconductor equipment sales of about $145 billion in 2026 and a record $156 billion in 2027. Inside that, the segment growth rates are the interesting part:

Segment	2026 growth	2027 growth
Wafer fab equipment (WFE)	+9.0% (to ~$126.2B)	+7.3% (to ~$135.2B)
Test	+12.0%	+7.1%
Assembly & packaging	+9.2%	+6.9%

Test and assembly-and-packaging are forecast to grow faster than core WFE in 2026. The big front-end names — the lithography, etch, and deposition giants — still own the dollars, but the rate of change is tilting toward the back end, exactly where Camtek, and the broader inspection/metrology cohort, operate. A $105M back-end inspection order with 2027 delivery is a concrete, dated data point underneath that forecast.

The skeptical read

A few cautions are worth stating plainly, because order press releases are a genre that rewards optimism:

“Orders” are not revenue, and 2027 is far away. Anything booked 18 months out carries cancellation and push-out risk. The 2027 delivery framing is a feature for backlog-watchers but also means none of this de-risks 2026.
No named customers. “Tier-1 OSAT” and “leading HBM player” are unverifiable by design. The dollar figures are specific; the counterparties are not.
Single-day announcements can be lumpy. Camtek choosing to issue a release implies these are unusually large bookings — which cuts both ways: great signal, but also a reminder that not every quarter produces a $105M headline.

None of that contradicts the direction. It just means the right unit of analysis is the pattern of back-end bookings across vendors, not any one release.

Practitioner note

If I were tracking the fab-equipment cycle as an operator or analyst, I would stop treating front-end WFE billings as the only leading indicator and start logging back-end inspection/metrology bookings with their stated delivery year. The Camtek release is useful precisely because it timestamps demand: $105M of inspection capacity committed in mid-2026 for 2027 HBM and 2.5D/3D output is a forward read on how much advanced-packaging volume the supply chain expects to be qualifying next year. I’d build a simple running tally — vendor, dollar value, “OSAT vs HBM vs logic,” and delivery quarter — and watch whether the 2027 delivery bucket keeps filling. A back end that is pre-selling capacity a year out is a back end that believes the AI-packaging ramp is real; the day those releases start carrying 2026 delivery dates instead of 2027, the cycle has caught up to itself.

Under-considered angle

The framing everyone reaches for is “AI demand is huge.” The under-discussed point is where the inspection burden is migrating. In a classic logic-shrink cycle, inspection scales with wafer starts and litho layers. In an HBM/advanced-packaging cycle, it scales with bonding and stacking steps and stack height — and a 16-Hi HBM stack carries far more inspectable interfaces than a 2D die ever did, with brutal yield-loss math because one bad die kills the whole stack. That decouples inspection-tool demand from raw wafer volume: you can have flat wafer starts and still see inspection bookings climb, simply because each finished AI part now passes through far more “look at it before you bond it” gates. Camtek concentrating an entire $50M+ order on one inspection product line for HBM is the visible edge of that shift — the back-end inspection layer is quietly becoming a per-stack tax on the AI build-out, not a per-wafer one.