COHU — Q2 2026 Financial Report Analysis

AI-chip test demand pulls Cohu back to break-even

Cohu makes the machines and parts that test computer chips before they ship. Its main products are "handlers", which move chips into test sockets and hold them at a set temperature, plus the contactors (the socket parts that touch each chip) and inspection equipment. In the quarter to June 27, 2026, net sales rose 38.4% to $149.0 million, from $107.7 million a year earlier and up 19.1% from $125.1 million in Q1. The 10-Q says the gain was "primarily driven by stronger demand across AI-based computing applications, industrial and automotive semiconductor markets," while mobile and consumer demand was flat. The extra sales took Cohu from a $16.9 million net loss a year ago to a $0.2 million net loss ($0.00 per share), close to break-even.

At a glance

• Systems sales up 78% ($70.7M vs $39.6M). Customers are buying whole machines again, not just parts and service, which is what happens when a chip-equipment cycle turns up.
• Operating margin 0.2% vs -16.0%. Operating margin is the share of sales left after running the business, before interest and tax. Cohu moved from losing 16 cents on every sales dollar a year ago to roughly breaking even, and from -8.9% in Q1.
• Q3 guided to $170M ± $7M, another 14% above Q2 at the midpoint. Management also raised its 2026 estimate for high-performance-computing (data-center chip) revenue to $100–110 million.

Results at a glance

"n/m" = not meaningful: a percentage change between two losses doesn't tell you anything useful. All figures from the 10-Q and the July 30 earnings release; first-half 2026 sales were $274.1M, up 34.1%.

What drove the quarter

Equipment, not parts, did the heavy lifting. Cohu splits its sales into "systems" (whole test handlers and inspection machines) and "non-systems" (contactors, spare parts, kits, service and software). Systems sales rose $31.0 million, three-quarters of the $41.3 million total increase. Non-systems sales, which track how many chips customers are testing rather than how many new machines they buy, rose a steadier 15.1%. That split is the clearest sign in the filing that customers have moved from "running old machines harder" to "buying new ones".

The buyer is mainly AI-server chips. Management's own framing is that AI and high-performance-computing spending is "more resilient than automotive, industrial, and consumer markets." The press release credits the Eclipse handler with T-Core, a handler that actively controls the temperature of high-power processors during testing, for winning data-center work. Cohu raised its estimate of 2026 high-performance-computing revenue to $100–110 million and says its pipeline of AI-driven compute opportunities is about $850 million. Those are management estimates, not booked orders.

Where it shipped. Shipments to Malaysia more than doubled to $26.5M (from $10.8M) and "Rest of World" rose 62% to $67.2M, while Taiwan fell 12% to $19.9M. Cohu reports sales by where products are shipped, which shows where chips are tested (often at contract test houses in Southeast Asia), not who the end customer is.

Utilization is the leading indicator. Cohu estimates that its customers' test cells (the tester-plus-handler setups on factory floors) were about 80% busy at the end of June, up from the prior quarter; the 10-Q describes this as "low 80%" driven by industrial chip demand. When existing capacity is that full, customers have to buy more machines to grow output, which is why this number tends to move before equipment orders do.

Margins and costs. Gross margin (sales minus the direct cost of making the products, here before amortization of acquired intangibles) rose to 45.4% from 43.7%, which the 10-Q puts down to "a more favorable mix of systems sold" and higher volume spreading factory overhead. Of the extra $41.3M in sales, about $20.5M (roughly 50%) dropped to gross profit. Research and development rose only 7.6% to $24.9M and SG&A (selling, general and administrative costs) rose 15.3% to $34.4M, so both fell sharply as a share of sales. Restructuring charges fell to $0.6M from $1.2M as the 2025 cost-cutting program winds down.

What the headline numbers hide

• About $2.8M of the $17.5M operating swing is accounting, not operations. Amortization of acquired intangibles fell to $7.3M from $10.1M because some assets from past deals finished being written off in 2025. That is a real but non-recurring help to reported profit.
• Below the operating line, interest income is doing real work. Interest income rose to $3.9M from $1.4M because Cohu is investing the cash from the $287.5M of 1.50% convertible notes it sold in September 2025; interest expense rose to $1.6M. Net interest added about $2.2M, which is why pre-tax profit ($2.0M) was higher than operating profit ($0.3M).
• Taxes ate all of the pre-tax profit. The $2.1M tax charge was larger than the $2.0M pre-tax profit, which the 10-Q attributes to "a change in mix of income between tax jurisdictions": profitable foreign units pay tax while losses elsewhere don't offset it. That gap alone is the difference between a small profit and a small loss.
• The GAAP vs adjusted gap is wide. Non-GAAP EPS was $0.26 against GAAP $0.00. The $14.8M of pre-tax adjustments are mostly share-based compensation ($6.6M) and intangibles amortization ($7.3M), plus $0.6M restructuring. Stock-based pay is a real cost to shareholders even if it isn't cash. Non-GAAP EPS also uses 53.4M diluted shares versus 47.3M basic, mainly because the convertible notes would turn into shares if the stock stays high.
• The convertible notes are now in play. Because Cohu's share price traded above 130% of the $27.18 conversion price, holders could convert from July 28 to September 8, 2026. Cohu pays the $287.5M principal in cash and can settle anything above that in cash or shares, so a rising stock price means more potential dilution. Even so, cash and investments of $498.2M exceed total debt of about $296M (mostly the notes).
• Cash flow is positive but leaned on suppliers. First-half operating cash flow was $20.8M against a $12.2M net loss, helped by non-cash charges and a $21.3M increase in accounts payable (unpaid supplier bills). That was offset by inventory up $18.2M and receivables up $16.8M, both of which the 10-Q ties to higher volume. Capital spending was only $4.3M. Building inventory and receivables ahead of a ramp is normal, but some of this cash comes from paying suppliers later.
• Inventory write-downs doubled while sales grew. Charges for excess and obsolete inventory were $3.2M in Q2 versus $1.5M a year earlier ($5.5M vs $3.2M in the first half). In an upturn you'd expect these to shrink; it suggests parts of the product line serving weaker end markets (automotive, consumer) are still carrying stock that won't sell.
• No help from buybacks. Cohu repurchased no shares in the first half, and the basic share count rose to 47.3M from 46.7M, so none of the improvement came from shrinking the share count.
• No order backlog disclosed. Cohu doesn't report bookings or backlog in the 10-Q (only $4.8M of contracted revenue due beyond one year), so guidance and utilization are the only forward signals. One customer was 11% of Q2 sales, so concentration is a real but moderate risk.

Takeaway: Cohu's recovery is now an equipment-buying recovery: systems sales jumped 78% and roughly half of each extra sales dollar became gross profit. But at about $150M a quarter it still only just breaks even, so the $170M Q3 guide and the AI-handler ramp have to land before reported earnings, not just adjusted ones, turn clearly positive.

Outlook

Management guided Q3 2026 sales to $170 million ± $7 million, which would be the third straight quarter of sequential growth (Q1 $125.1M, Q2 $149.0M). It did not give a margin or EPS forecast. The things to watch:

1. Does operating profit scale with sales? If roughly half of extra revenue keeps flowing to gross profit and operating costs grow slowly, another $20M of sales would lift operating income meaningfully. A margin that stalls would suggest the AI handler mix or pricing is less favorable than the systems shift implies.
2. High-performance-computing revenue versus the $100–110M estimate. This is the part of the business management is leaning on; the rest (automotive, consumer) the 10-Q still calls weak or flat.
3. Inventory write-downs and receivables. If these keep rising faster than sales, the cash quality of the recovery weakens.
4. Convert dilution. If the share price stays above the conversion trigger, the extra shares behind non-GAAP EPS become a live issue for per-share earnings.

Our read: chip-test equipment is a cyclical business, and every forward signal in this filing (systems mix, utilization near 80%, a higher guide) points to an upturn that is broadening from AI into industrial. The filing doesn't give backlog to confirm it, and outside AI the 10-Q still describes customers as "cautious with capacity investments", so the recovery currently rests on one end market.