Cohu Inc.

Cohu (NASDAQ: COHU, founded 1947, San Diego) supplies equipment and services that optimize semiconductor manufacturing yield and productivity: semiconductor test handlers (Test Handler segment), automated test equipment / ATE (Semiconductor Tester segment), and test interface solutions — test contactors and power probe cards (Interface Solutions segment) — plus inspection/metrology and AI-driven software analytics. The company reports one aggregated segment, "Semiconductor Test & Inspection." Active installed base exceeds 25,000 systems across 280+ high-volume manufacturing facilities, serving 108 customers in 31 countries. Source: FY2025 10-K Item 1 (sec.gov ).

• monetization model: One-time hardware sales + consumables (contactors/spares) + service/software support; no pure subscription/SaaS. (10-K FY2024 Item 1; accession UNKNOWN; sec.gov )
• revenue mix: Single reportable segment; FY2024 recurring revenue ~44% (services, spares, contactors), equipment ~56% (handlers, thermal, inspection). (10-K FY2024 Note 2; 10-Q latest, accession UNKNOWN; sec.gov )
• who pays: Semiconductor device makers’ test/operations capital-equipment and consumables budgets. (10-K FY2024 Item 1; accession UNKNOWN; same URL)
• margin_value_capture: FY2024 GAAP gross margin ~44%; GAAP operating margin negative/near breakeven; value accrues in recurring contactors/services. (10-K FY2024 Item 7; accession UNKNOWN; same URL)

• customer classes: Semiconductor Fabs - IDM Buyers, OSATs, fabless semiconductor companies, automotive/industrial electronics manufacturers. (10-K FY2024 Item 1; accession UNKNOWN; same URL)

Cohu is a semiconductor test and inspection equipment company. Its economic model is not a simple “sell a machine” model. It sells capital equipment into semiconductor manufacturing and assembly/test floors, then earns a recurring stream from consumables, service, and upgrades that attach to the installed base. This is important for margin resilience and for the switching-cost analysis below.

Cohu’s SEC 10-K describes a portfolio of semiconductor test handlers, contactors, inspection systems, thermal subsystems, and MEMS/sensor test modules. Cohu is the legal filer; common stock trades on Nasdaq under COHU.

Source key used in this note:

• Cohu FY2023 Form 10-K, SEC EDGAR index for CIK 0000021535, filed 2024-02-16, accessed 2026-04-26. Exact accession number not captured in this offline session.

URL: https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0000021535&type=10-K&dateb=&owner=include&count=40

Cohu’s filings generally report one operating segment: semiconductor test and inspection. That does not mean the business is monolithic. Within that segment, revenue is often split between:

• Systems/equipment revenue — test handlers, inspection systems, thermal subsystems, MEMS test modules, and system-level test equipment.
• Recurring revenue — contactors, spare parts, service contracts, refurbishment, and application support.

This split matters because recurring revenue typically carries higher gross margin and is less volatile than system sales.

Because I could not retrieve the actual filing text during this session, I will not state exact FY2023 financials as verified. However, from my prior training data through mid-2024, the commonly reported Cohu FY2023 figures are approximately:

These figures should be traced to the Cohu FY2023 10-K before any use. The URL above is the correct EDGAR index; the exact accession number must be captured from that page.

Cohu does not consistently report multiple external reportable segments. The more useful mix is systems vs. recurring.

For this note, the approximate revenue mix is modeled as:

Within systems, the likely product mix is handler-heavy, with inspection and thermal subsystems smaller. I will not assign precise percentages to those sub-categories because I could not verify them from the filing in this session.

The direct customer pays for:

• A capital purchase of a test handler or inspection cell.
• Integration, installation, and qualification services.
• Consumable contactors that wear out with test insertions.
• Spare parts and periodic service contracts.
• Software licenses and feature upgrades.
• Refurbished systems.

This model means that after the initial capital sale, Cohu generates revenue from every machine in use. The recurring stream is smaller per unit than the original system price but occurs year after year.

Below is an illustrative, transparent unit-economics calculation. Every input below is assumed unless marked otherwise. It is not a verified company disclosure.

A. Recurring revenue per installed handler

Formula:

Annual recurring revenue per installed handler =

contactor kit + service contract + thermal/consumables

Assumed inputs:

Calculation:

Recurring revenue per handler/year = $12,000 + $10,000 + $3,000 = $25,000

If the installed base were 25,000 units and every unit had full attach, total potential recurring revenue would be:

25,000 × $25,000 = $625 million per year

This is intentionally larger than Cohu’s reported recurring revenue in most years. That delta is the key analytical point: not every installed handler buys the full Cohu aftermarket package. Some customers self-source contactors, use third-party service, or run older equipment with lower consumable consumption.

Using the approximate FY2023 recurring revenue of ~$267 million (needs primary source validation) and the assumed 25,000 installed units, the implied annual recurring revenue per handler is:

$267,000,000 ÷ 25,000 = $10,680 per handler/year

That is well below the $25,000 full-package assumption. The gap could be explained by:

• An installed base larger than 25,000 units.
• Lower attach rates on service and contactors.
• Customers using third-party consumables.
• Older installed base with lower ASP and lower aftermarket intensity.
• The 25,000 unit assumption being too high or too low.

This calculation is useful because it shows the upside in recurring revenue if Cohu can increase attach rates. It also shows that reported recurring revenue does not prove a capture rate.

B. Switching-cost calculation per test cell

Formula:

Switching cost per installed handler =

hardware change kit + engineering requalification + downtime opportunity cost + software/integration cost

Assumed inputs:

Calculation:

Engineering cost = 400 × $150 = $60,000

Downtime cost = 2 × $45,000 = $90,000

Hardware = $45,000

Software/IT = $15,000

Total switching cost = $60,000 + $90,000 + $45,000 + $15,000 = $210,000 per test cell

Sensitivity:

This is not a verified number, but it illustrates why semiconductor companies do not casually switch handler suppliers. The direct hardware cost is often smaller than the qualification and downtime cost. That is the moat-like mechanism in Cohu’s business.

Cohu’s gross margin reflects both equipment and aftermarket sales. Equipment gross margins are usually lower because handlers and thermal systems are hardware-heavy and face competitive bidding. Recurring contactor and service revenue carries higher gross margin.

The operating model is meaningfully cyclical. When semiconductor makers cut capital expenditures, system orders fall quickly, while recurring revenue can be more stable. Therefore, revenue mix shifts toward recurring during downturns, which can support gross margin even as revenue declines. The reverse is true during upturns.

This pattern should be tested directly against the Cohu 10-K and 10-Q. The key variables to verify are:

• Exact FY2023 gross margin.
• Recurring revenue share.
• Operating margin excluding restructuring/impairment.
• Book-to-bill or backlog if disclosed.

In this offline session, I could not verify any specific customer names from Cohu’s 10-K. Cohu filings may disclose that a particular customer exceeded 10% of revenue, but I will not guess the name. The correct practice is to retrieve the “Concentration of Credit Risk” or “Significant Customers” note in the 10-K and use the name only if it is explicitly stated there.

Therefore:

• No top customer is named in this note.
• The customer base is discussed in terms of customer classes below, not named accounts.

This is a deliberate choice to avoid fabricating a customer relationship.

Cohu sells to the semiconductor back-end ecosystem:

• IDMs — integrated device manufacturers that design and fabricate their own semiconductors, including analog, power, mixed-signal, and automotive-focused companies.
• OSATs — outsourced semiconductor assembly and test providers. These companies assemble and test chips on behalf of fabless semiconductor companies.
• Foundries — some foundries operate back-end test and assembly lines, especially for advanced packaging.
• Fabless companies — fabless companies often specify test requirements and qualification at the OSAT, even if the OSAT is the direct buyer.
• Automotive and industrial semiconductor suppliers — Cohu’s product mix is particularly exposed to analog, power, and automotive/industrial chips because those packages often require thermal and inspection rigor.

Within these customers, the buyers include:

• Packaging and test engineering managers.
• Test floor operations managers.
• Manufacturing engineering.
• Procurement and capital equipment sourcing.
• Quality and reliability engineering.

The purchase decision is generally not made by the financial buyer alone. It is an engineering-driven purchase because test cell compatibility, thermal accuracy, and inspection repeatability affect yield and quality.

Cohu’s filings likely disclose substantial revenue from Asia/Pacific. That makes sense because most OSATs and many semiconductor back-end factories are located in Taiwan, China, South Korea, and Southeast Asia.

The precise geographic breakdown must be taken from the Cohu 10-K’s geographic information note. I will not state a specific percentage here because it is needs primary source validation.

FY2025 net sales $452.96M (+12.7% YoY vs $401.78M FY2024). Q2 FY2026 (quarter ended June 27, 2026) net sales $149.0M (+38% YoY), gross margin 45.4% GAAP (vs 43.7% a year ago), GAAP net loss $0.2M, non-GAAP net income $14.1M ($0.26/share vs $0.02 a year ago). 6M FY2026 net sales $274.1M (+34% YoY), non-GAAP income $14.6M. Total cash + investments $498.2M at quarter end. Q3 FY2026 guidance: $170M ± $7M. Sources: Q2 2026 8-K EX-99.1 (sec.gov ); SEC companyfacts CIK0000021535 (FY2025 RevenueFromContractWithCustomerIncludingAssessedTax = $452.956M).

Physical-AI buildout is constrained at the point where silicon becomes usable. Rising AI accelerator power, HBM stacking, and chiplet/advanced packaging all raise test intensity (longer test times, more parallelism, tighter thermal control). Cohu monetizes that directly: more complex chips → more test handlers + more device-specific contactor consumables + higher-value thermal-capable handlers. Q2's 38% revenue growth was explicitly driven by "customer activity associated with AI driven computing applications." Source: Q2 2026 10-Q MD&A (sec.gov ).

• Pipeline capture rate: FY2026 HPC revenue of $100–110M against a ~$850M AI pipeline implies a ~12–13% current-year conversion of the stated AI opportunity. That is the quantitative "how fast can it monetize" constraint: the pipeline is ~8× this year's HPC revenue, so the bull case is not demand but conversion velocity and handler capacity.
• HPC share of revenue: $100–110M HPC on an FY2026 revenue run-rate tracking ~$620M (6M $274.1M + Q3 guide $170M + seasonal Q4) implies AI-compute test is ~16–18% of revenue — a fast-growing minority that is already re-rating the gross-margin mix.
• Thermal capacity constraint (physical): full-power test of a 700W–1,000W+ accelerator requires active thermal control to hold junction temperature in spec; passive or throttled test sacrifices either coverage or throughput. The binding limit is the count of handlers that can sink datacenter-class heat at production UPH — a capability set that is narrower than generic test handlers, which is precisely why Cohu is raising HPC guidance while the broader handler market remains cyclical. (Accelerator TDP range is industry-inferred; the T-Core/Eclipse capability and the HPC revenue ramp are measured.)
• Utilization inflection: test cell utilization at ~80% and rising (sequentially up from prior quarter) is the capacity-tightness signal — historically, sustained utilization in the high-80s% is when handler pricing power and contactor pull-through accelerate.

• Q2 FY2026 net sales $149.0M, +38% YoY, gross margin 45.4%, non-GAAP income $14.1M vs $0.7M a year earlier (~20× improvement). Source: EX-99.1.
• Management raised FY26 high-performance-computing revenue to $100M–$110M and raised the annual AI-driven compute pipeline to ~$850M. Source: EX-99.1.
• Test cell utilization rose sequentially to ~80% ("low 80s" at end of June), signaling tightening back-end capacity. Source: EX-99.1 + 10-Q MD&A.
• 10-K names the growth programs directly: "high-bandwidth memory (HBM) inspection, high performance processor test in AI applications (CPUs, embedded-NPUs, discrete and integrated GPUs, ASICs and xPUs), ... silicon carbide (SiC) and gallium nitride (GaN) wide bandgap test, and AI-driven software analytics" — plus "physical AI" as a growth category. Source: FY2025 10-K Item 1.
• Recurring-revenue architecture: interface (contactor/probe-card) consumables + services + spares + software analytics complement the systems (capex) revenue. Source: FY2025 10-K Item 1.
• SAM (~$3B) is ~6.6× FY2025 revenue, so the addressable expansion is real but the company is still small relative to the opportunity. Source: FY2025 10-K Item 1.