A handler presses each chip into a socket, holds it at a set temperature while it is tested, and sorts it. Cohu also sells the contacts, testers and inspection machines.
The company · 2 · What exactly does it sell?
Parts and service are half of sales
Each $1 of sales, second quarter of 2026
53%47%
Parts, contacts and serviceMachines
Machines swing with the chip cycle; parts sales are steadier.
The company · 3 · How does it make money?
Machines in good years, parts every year
Over 25,000installed machines
About 16%fall in parts sales in a downturn
About 57%fall in machine sales then
Every installed machine keeps buying parts, but even those fell in the last downturn.
Is it a good business? · 4 · What gives it an edge?
Every installed machine keeps buying parts
Over 25,000
installed machines
60%
of sales from parts that wear out and are replaced
-16%
parts sales in the 2024 downturn, while machine sales fell 57%
Recurring parts sales are steadier than machines, but not immune to the cycle.
Cohu's chief executive calls Hon essentially the only rival in AI handlers, and Hon holds most of those sockets.
Is it a good business? · 6 · What does its record show?
Sales more than halved, then began recovering
Sales, $M
887.2
812.8
636.3
401.8
453.0
569.3
Dec 2021Dec 2022Dec 2023Dec 2024Dec 2025last 12 mo
Downturn$887M → $402Mrevenue, 2021 to 2024
Recovery$569Mrevenue in the past year
Profit23% → −2.4%operating profit per $1
Fixed costs turned 2024's revenue fall into a loss, and now lift recovering sales.
Is it a good business? · 7 · How much room is there to grow?
Test spending rises, mostly on testers
$20.8Bchip test equipment spend by 2028, per SEMI
About $3Bopen to Cohu, it says
About $850Ma year in AI chips and memory
Most of the spending is on testers Cohu does not sell.
Is it a good business? · 8 · Who is in charge?
A long-serving team steering toward AI
Since 2014Luis Müller has run Cohu
About 3%of the shares held by insiders
22 Mayboth leaders sold outside pre-set plans
The investor day set a long-range target but no count of customers that have approved its AI handler.
The story right now · 9 · Why do AI chips need special handlers?
AI chips run hot and must be tested hot
About $850Ma year in AI chip handling
About 3 timesas many AI handlers it can ship in 2027 as in 2026
+265.2%the shares in the year to 5 October
AI processors draw so much power that holding them at a steady temperature during testing is hard. Cohu's new handler does it and is winning its first customers.
The story right now · 10 · Why do sales swing so hard?
Chipmakers buy machines only when they expand
Chip demand risesFactories add capacity.
Orders for machines jumpUp 126%.
Demand coolsOrders stop.
Sales halveAs in 2022 to 2024.
The price assumes the AI ramp works and the cycle stays near its best year.
The story right now · 11 · What changed lately?
Orders more than doubled and the outlook rose
$173.0Msales, up 37%
126%rise in orders
7.4 centsoperating profit per sales $1
About 82%of test floors busy
Solid quarter
It will build to ship about three times as many AI handlers in 2027.
The story right now · 12 · What is everyone arguing about?
Is the price already paying for the ramp?
Many investors think
An AI test winner early in an upturn
We think
The price pays for the ramp and a record peak
The AI handler is about a sixth of sales. We would be wrong if that line passes $200M a year.
The price and our view · 13 · What is a share worth?
The price sits near our sunny future
$19.79what we think a share is worth · the last close is of that
$19.79our value
Attractive below $15.83
Neutral
Unattractive above $23.75
$12.78gloomy: growth halves and less kept per $1
$19.79our estimate: experts' path, then slowing
$32.19sunny: growth beats forecasts by a quarter
Value per share in three futures. None is a prediction of the share price.
We estimate what a share is worth from the cash the business should earn over many years. Our rule: buy only when the price is at least a fifth below our estimate, as a cushion for being wrong.
The price and our view · 14 · How could we be wrong?
A stronger rival and a turning cycle
A stronger rivalBiggest
Hon holds the AI sockets.
The cycle turnsWatch
Sales halved last time.
Loans turning into sharesShowing now
Bonds may convert.
Hon's share of socketsUnknown
Nobody publishes it.
Supply could also limit how fast the ramp runs.
The price and our view · 15 · How will we know?
February counts AI handler customers
More customers approving the AI handlerGood sign
Orders falling backBad sign
AI handlers reaching $200M a year in 2027We turn positive
FEB 202711
Fourth-quarter 2026 results, 11 February. Companies report every three months; each report is a test of the story.
Our view also changes on price alone: if the shares fall to within a fifth of our estimate, we move to Neutral.
The price and our view · 16 · So, what do we think?
An AI ramp and a record year, priced as normal
$19.79our value
Attractive below $15.83
Neutral
Unattractive above $23.75
UnattractiveHigh conviction
Cohu's handler for hot AI chips is winning its first customers. But a Taiwanese rival holds most of those sockets and keeps far more per $1, and sales more than halved in the last downturn. Unattractive.
↑ ↓ to move · C for contents
What it does
More about the company
Who it is
Before a chip ships, it is tested. A handler picks each packaged chip, presses it into a socket wired to a tester, holds it at a set temperature while the test runs, and sorts it good or bad. A processor tested too hot is scrapped though good; too cool, it may fail later.
Cohu sells those handlers, the contacts in the socket that wear out, its own testers for simpler chips, inspection machines and software. Teradyne, a much larger tester maker, names Cohu as a competitor.
How it makes money
Two ways. Machines are sold when chip testers add capacity, and stop selling when they don't: total sales fell from $887.2M in 2021 to $401.8M in 2024. Steadier money comes from the contacts, kits, spares and service the installed machines keep needing, which were 60% of sales in 2025 and about half of the June quarter's.
Growth and profit
What growth costs
Most of Cohu's spending on research, selling and running the company stays about the same whatever it sells, so as sales recover, much of each extra dollar becomes profit: management expects about 40% of the next quarter's added sales to. The same fixed costs turned the 2024 slump into a $71.7M operating loss.
Its record
The numbers
Is it growing?Fast
Sales grew 34% over the twelve months to Sep 2026, faster than the 13% in the year to Dec 2025.
Is it profitable?Losing
Operating margin is −2.4%, down from 6.8% in the year to Dec 2023.
How steady is it?Swingy
Yearly sales growth has ranged from −37% to +39% over the last 5 years, with sales falling in 3 of them.
Does it earn good returns?Poor
Shareholders' money lost 2.8% a year, down from 24% in the year to Dec 2021.
Is it making cash?Solid
It turned 9.8% of sales into free cash, $55.9M over the last twelve months.
Is the balance sheet safe?Net cash
It holds $508M of cash against $302M of debt, $206M more cash than it owes.
Are shares being diluted?Creeping
The share count rose 12% in 4.8 years, and pay in shares costs 4.0% of sales.
Why it wins
The evidence
Each new chip design needs its own set of socket contacts, and they wear out. So every one of Cohu's more than 25,000 installed machines keeps buying parts as long as it runs, and a test floor that runs well does not swap machine makers mid-programme.
The evidence: from 2023 to 2024 machine sales fell about 57%, but these recurring sales fell only about 16%. Steadier, but not immune.
Who could beat it
The rivals in full
The AI handler sockets Cohu wants already have an owner. Its chief executive names Hon Precision as essentially the only rival there, the long-standing supplier to the firms that test AI chips. Hon more than doubled its 2025 sales, earned about 50% operating margins on them, and grew another 91.6% in the first half of 2026. Advantest, the biggest tester maker, sells handlers too.
Each rival
Hon Precision thermal test handlers7769.TW
The Eclipse handler itself, at the test subcontractors that package and test AI processors: the buyer keeps the incumbent's handler and thermal control rather than qualifying Cohu's.
Advantest test handlers and testers6857.T
Both the handler and the tester on a test floor, sold as one package by the market leader in testers.
Cohu's own testers for power, radio and mixed-signal chips, the line where Cohu is the newcomer.
How big it could get
The market and who has signed
The market
SEMI, the industry's trade body, expects $15.3B of spending on chip test equipment in 2026, up 31.0%, and $20.8B by 2028. Most of that is testers Cohu does not sell. Cohu says about $3B is open to it, and counts about $850M a year of AI processor and memory spending at the customers it is chasing. It sold $453.0M in 2025.
Who has signed
Little is on file. One unnamed customer ordered $26M of AI handlers in July, mostly for delivery this quarter. The last reported orders on hand were $165.1M at December 2025, before the AI ramp, and many orders can be cancelled with little or no penalty.
Who runs it
Leaders, pay and the cash
Who runs it
Luis Müller, a former Teradyne executive, has run Cohu since 2014 and steered it toward AI handlers; finance chief Jeffrey Jones has been there since 2007. Insiders own about 3%. Both sold shares near $46 in May outside any pre-set plan, and no insider bought in the year. The bonus pays for sales growth and profit; most long-term shares pay only if the stock beats small companies.
What they did with the money
It bought Xcerra in 2018, a small software firm in 2025, and repurchased little stock. In 2025 it borrowed $287.5M through bonds that turn into shares, which at today's price means about a tenth more shares later.
The last report
The latest report
Cohu sold $173.0M, raised its computing outlook to $120M and guided up
What it did to the thesis Solid quarter
The quarter and the call leaned against our doubt: a bigger qualified pipeline, triple the capacity and a 2027 computing line above $200M. But our routes also need a far higher operating margin and more qualified customers, and neither is shown, so the view stays Unattractive.
The report
The September quarter brought $173.0M of sales, up 37%, against a guide of about $170M. Orders rose 126%, with systems up 318%, and Cohu raised this year's computing revenue estimate to about $120M from up to $110M.
It guides the fourth quarter up to $178M and puts its pipeline of computing sales at about $1.0B, of which $230M a year is qualified. On the call it said it will build to ship about three times as many AI handlers in 2027. Even so, only 7.4 cents of each sales dollar was operating profit on the accounting basis.
What's happening now
The background
The shares have nearly quadrupled in a year, and most of the rise came from investors paying more for each dollar of sales rather than from the sales themselves. That is the market pricing in the AI ramp.
On 28 October sales came in at $173.0M, the computing outlook rose to about $120M and the call gave a 2027 view of 30% to 40% growth. The 10 November investor day set a target three to five years out but no count of qualified customers. Test floors are about 82% busy, and upturns usually last about six quarters.
Where we disagree
Both sides in full
The market
Cohu is an AI test winner early in an upturn, which is why the shares nearly quadrupled in a year.
We believe
The AI handler demand is real, but the price already pays for it and for the upturn peaking at a record.
The argument
Analysts and the company lead with the AI handler, and the ramp is genuine. But it is about a sixth of this year's sales, the rest is an ordinary upturn, and the rival in that socket is bigger and far more profitable. Analysts already expect sales nearly to double by 2028.
We would be wrong if the AI line passes $200M a year at a healthy company profit margin while the rest of the business holds.
What it's worth
How we value it
Valued on the analysts' revenue path for three years, each later year closing 30% of the gap to 4%, on official operating profit with stock pay a cost and tax paid in full; the thesis's own evidence is the check on that path.
The three cases
Bear case$12.78
Growth runs at half the analysts' path, 18% and 11%, the margin settles a fifth lower at 8.6%, and new capital earns only the 8% rate: the downturn test.
Our value$19.79
Revenue grows 37% and 21% over three years, the analysts' path (C59), then fades toward 4%, closing 30% of the gap each year; the margin moves to 10.7%, the analysts' year-three figure after stock pay and on the filed basis (2.5% less).
Bull case$32.19
Growth runs a quarter above the analysts' path, 46% and 27%, the margin reaches 11.8%, and new capital earns 11%.
The bet
Seen as a bet
This bet is on one business: Cohu's handler for AI processors. The rest of the company is counted separately at what the market has paid for it across a cycle, and covered well under half of the price this bet was struck at.
Even if the AI line reaches the yearly sales management sized its doubled factory for, the company is worth little more than it costs, short of the doubling a fair bet needs. If it stalls, shareholders keep an ordinary test-equipment business with spare cash.
What could go wrong
Every risk
The risks
A stronger rival
Hon Precision already supplies most AI test floors at about 50% operating margins, so Cohu may win only a few sockets.
The cycle turns
Sales fell from $887.2M in 2021 to $401.8M in 2024 when customers stopped adding capacity, and the same fixed costs produced losses.
Shares from the bonds
The convertible bonds hand about a tenth more shares to bondholders at today's price, so shareholders keep less of any success.
Supply limits the ramp
Lead times and parts costs are rising, and management says suppliers, not its factory, are what take most of its attention.
What we still do not know
Whether any bondholders turned their bonds into shares and cash in the July to September window. The September-quarter filing will say.
How many AI test sockets Hon Precision holds and whether Cohu is taking any. Nobody publishes the split.
Who placed the $26M order, and whether it was a test house or a chip designer buying directly.
What changes our view
Every route to a new view
Every route to a new view
→ Neutral
The share price falls to about $23.75, back within 20% of what we think the business is worth.
→ Attractive
The share price falls to about $15.83, 80% of what we think the business is worth.
→ Attractive
The AI handler line reaches $200M of sales a year in 2027. Cohu's profit margin passes 15% of sales for a full quarter, under ordinary accounting.
Research reopens
Cohu reports more than eight AI customers qualified, or names a test house moving from Hon Precision to its handler.
The next test
It supports the case if
Qualified computing customers still unstated or at most eight, no subcontractor named, and a fourth-quarter GAAP operating margin below 15% with a 2027 guide that does not reach it.
It pressures the case if
More than eight qualified computing customers or a subcontractor named moving from Hon Precision, or a GAAP operating margin above 15% in the fourth quarter or guided for a full quarter of 2027.
We would also be wrong if
Sales stay above $170M every quarter through 2027 while car-chip demand is still weak.
The verdict
Scores and how we weighed it
The scores
Growth
3 of 4
Business quality
2 of 4
Financial position
4 of 4
Execution record
3 of 4
Per share capture
2 of 4
Potential
2 of 4
Risk profile
3 of 4
Valuation
1 of 4
Evidence quality
3 of 4
How we weighed it
ValueNegative
The price sits above our value and above the bull case.
BusinessMixed
A mixed business: a promising new handler, but thin margins.
GrowthPositive
The computing ramp runs ahead of plan, with capacity tripled.
RisksNegative
A stronger rival holds most AI sockets, the cycle can turn and convertible bonds may add shares.
RivalsMixed
No filed rival compares directly; against its own normal it is far dearer.
The groupMixed
Chip-equipment names move together on news about chipmakers' spending.
Latest newsPositive
The last report leaned against our doubt with a bigger qualified pipeline.
In plain words
Our estimate of value
We forecast how much cash the business can earn over the next ten years and beyond, then ask what that stream is worth today to an investor who wants an 8% yearly return. We do it three times: a gloomy future, an expected one and a sunny one. We call a share Attractive only when its price is at least 20% below the expected estimate, and Unattractive when it is 20% above.