Earnings Call Highlights: Teradyne (TER) | FY2026 Q3 — AI Is Over 60% of Revenue, Yet Compute Is Only a Fifth: Test Dollars Are Flowing In from Memory, Power and Copper
1. Executive Takeaway: Three Signals
First, AI demand at Teradyne has decentralized. CEO Greg Smith said AI-driven revenue is more than 60% of total revenue, and also that the portfolio outside compute delivered close to 80% of sales. Read those two lines together: Teradyne's biggest AI gains are not coming from GPU/ASIC test, where it is still catching up, but from HBM memory, data center power management, hard drive capacity, rack-level copper and assembly robots. That is a sturdier structure than winning a few hyperscalers.
Second, compute share is a 2H27 story, not a today story. The timeline across the four compute customers is clear: the first hyperscaler is in the mature dual-vendor phase, with the next major ramp in mid-2027; the merchant GPU customer is in the fast-follower phase; the second hyperscaler has passed qualification and is expected to place capacity buys in 2H27; the third is still in initial qualification. Management also conceded the ASIC programs will be "more back-half loaded."
Third, a test vendor itself is now naming the CPO bottleneck. Teradyne kept its CPO test equipment TAM at roughly $100M in 2026, growing to $300-700M by 2028 and potentially $1B+ longer term. But it said where the range lands depends on CPO economics, and that today's test time and test intensity are themselves limiting how broadly CPO gets deployed.
2. Revenue and Financials
Metric | FY2026 Q3 | YoY | QoQ |
Revenue | $1,306.0M | +69.8% | -1.7% |
Gross margin (GAAP) | 59.1% | +0.6pp | -0.7pp |
Operating income (GAAP) | $402.0M (30.8% operating margin) | +176.6% | -8.2% |
Net income (GAAP) | $342.4M | +186.4% | -8.6% |
Diluted EPS (GAAP) | $2.17 | +189.3% | -8.8% |
Non-GAAP EPS | $2.29 | +169.4% | -7.3% |
Both revenue and non-GAAP EPS beat the top of guidance (prior guide: revenue $1.20-1.30B, non-GAAP EPS $1.85-2.15). The 1.7% sequential dip is not weakening demand: management had flagged that compute orders would bottom in Q3, and memory, IST, Product Test and Robotics filled the gap.
Segment breakdown:
Segment | Revenue | Share of total | YoY |
Semiconductor Test | $1,072M | 82% | +77% |
of which SoC | $676M | 52% | — |
of which Memory | $301M (record) | 23% | — |
of which Integrated Systems Test (IST) | $95M (record) | 7% | ~+150% |
Product Test | $138M | 11% | +56% |
Robotics | $96M | 7% | +28% |
For the first nine months, revenue was $3.92B, up more than 85% YoY, and non-GAAP EPS was $7.31, up 235%. Q3 free cash flow was $227M ($806M year to date), and the quarter closed with $625M in cash and investments. Teradyne bought back $99M of stock and paid $20M in dividends in the quarter.
▲ Teradyne FY2026 Q3 core financials (revenue / gross margin / operating margin / net income / EPS / YoY / Q4 guidance / segment revenue) (Image source: Simple Tech Trend original; data from Teradyne's SEC 8-K earnings release and IR press release)
3. AI Is Over 60% of Revenue, Compute Is Only a Fifth: Where the Money Comes From
Turn "close to 80% outside compute" around and compute is roughly a fifth of revenue, about $260M, up nearly 50% YoY. The rest of the AI upside is spread across five lines, each with a concrete physical reason behind it.
3.1 Memory: Shorter HBM Stacks Are Good News for Teradyne
Memory revenue was $301M, the fourth straight quarter above $200M and another record, driven by HBM and DRAM, with NAND recovering as well.
The most interesting part of the call was HBM stack height. Smith said DRAM wafer supply has been so tight over the past two quarters that AI accelerators are moving to shorter HBM stacks (say from 8-high toward 4-high) to get more HBM units out of the same DRAM wafers. The same wafer volume yields twice as many 4-high stacks as 8-high stacks. Why it matters for Teradyne: its share in DRAM wafer sort is lower, while its share in HBM performance test is higher. Same wafers, more units to test, and demand shifts toward the step where Teradyne is strongest.
Management was direct about the 2027 ranking: HBM is the star, with DRAM and Flash also growing. Korean test vendors mainly serve wafer sort, so their head-on overlap with Teradyne's performance test is limited.
3.2 Auto and Industrial: Really Data Center Power
Auto and industrial within SoC grew nearly 90% YoY, but the main driver is not cars. It is power management chips for AI data centers. Product revenue in this space grew from under $400M in 2025 to over $600M in 2026, and management expects 800V DC power distribution and integrated voltage regulators (IVR) to accelerate the trend.
As AI rack power climbs, the count and specs of power chips climb with it, and test demand follows. This line has nothing to do with GPU share.
3.3 IST: Hard Drive Test Time Scales with Capacity
Integrated Systems Test (IST) revenue was $95M, up about 150% YoY and a record. The key mechanism: HDD test time scales with capacity, not unit count. AI-driven nearline storage pushed 2026 exabyte growth for HDD above 20%, and Teradyne added a third major customer, so all three major drive makers are buying.
3.4 Product Test: Copper Turns from Commodity into an Engineered, Tested Component
Product Test revenue was $138M, up 56% YoY and nearly 30% QoQ, with a 1.4x book-to-bill. Smith's explanation: scale-up networking is still mostly copper, and backplanes and cables now run at unprecedented data rates, so copper has turned from a cheap commodity into a system element that must be rigorously tested, driving demand for TDR and BERT test gear from the MultiLane joint venture (MLTP). It is the same trend we broke down in OCP Global Summit 5/6 | Copper Retreats into the Package, 1-3 Meters Goes to Active Copper First: copper is not leaving, but every meter is getting harder to build and needs more measurement.
The CFO also noted that Product Test is accretive to company gross margin, so this growth helps margins.
3.5 Robotics: Profitable for the First Time, but the Real Ramp Waits Until 2028
Robotics revenue was $96M, up 28% YoY, a third-quarter record, and the group turned profitable in the quarter. Electronics and semiconductor is now robotics' largest vertical, with ODMs and CMs using cobots for AI server tray assembly and test. Revenue from the largest customer tripled from 2025 to 2026.
But Smith was cautious on timing: partnerships with physical AI labs need a gestation period. 2027 should bring growth in transactions with the labs and the start of end-market growth, while the real ramp is "more of a 2028 thing."
4. The Compute Dual-Vendor Timeline
Teradyne's main rival in AI compute test is Advantest. In Earnings Highlights: Advantest (6857.T) | FY2026 Q1 we broke down its toll-booth position across GPU, ASIC and CPU test. Teradyne's way in is customers' dual-vendor strategy: they want a second test supplier to secure capacity, spread supply chain risk and make sure test capability arrives in time for new devices.
Customer | Current phase | Next milestone |
Hyperscaler #1 | Mature dual sourcing | Next major ramp in mid-2027 |
Merchant GPU customer | Fast follower, multiple devices in development | Additional revenue in 2027 |
Hyperscaler #2 | Passed qualification, now fast follower | Capacity buys in 2H27 |
Hyperscaler #3 | Initial qualification | No timeline given |
Smith was also candid about an awkward piece of math: Teradyne made significant share gains in compute in 2026, but the compute market grew so fast that it diluted overall share. In other words, it is chasing a target that keeps getting bigger. Management listed in-house silicon efforts such as Terafab, OpenAI and Anthropic as additional X factors, and is approaching them the same way as the first three hyperscalers.
On capacity, the flagship SoC platform UltraFLEXplus is at about 4x the capacity of 18 months ago, with the next step-up being planned. Teradyne qualified a second contract manufacturer (CM) in 2026 and is adding capacity for its memory test platform and Eagle power test line. Management says it is largely holding lead times, but unforecast spot demand will see longer lead times than a year ago.
Teradyne also launched three new UltraFLEXplus instruments aimed at next-generation AI devices arriving in 2027 that no current ATE can test:
UltraPin5000-EM: up to 80x deeper memory than the current instrument, for device characterization
UltraVS64-HP: tests accelerators drawing up to 15,000 amps
UltraPort-PCIe6: the industry's first ATE solution for PCIe Gen 6
5. CPO and Optical Test: The TAM Is Unchanged, the Bottleneck Is Test Time
An analyst asked whether Teradyne was de-emphasizing CPO. Smith's answer is worth reading line by line.
The numbers didn't change: the CPO test equipment TAM is about $100M in 2026 and $300-700M by 2028, a range management still considers right, with $1B+ possible longer term. One initial customer is driving a significant 2027 ramp, but from a very small base.
High end or low end depends on economics: CPO test time and test intensity are high today, which limits how broadly it is deployed, and the assembly flow has other challenges to work out. That matches the OSAT view we covered in 2026 OCP APAC Summit | ASE | Nicole Tien | After Opening the Package: CPO Is Stuck Not on Light but on Test. The difference is that this time a test equipment vendor is saying it.
The strategy is a hedge: whether connections end up using CPO, NPO, pluggable optics or copper, clusters get bigger and connection counts rise. Teradyne serves pluggables and copper through Product Test, switch silicon through Semi Test, and pursues CPO through an open, non-exclusive ecosystem.
The connections have to be made; which technology makes them is still open. The best strategy for a test vendor is to get paid on every option.
The newest piece of that ecosystem is the optical test partnership with FormFactor announced this quarter, on top of existing ties with Technoprobe (probe cards) and ficonTEC (optical alignment and assembly). FormFactor is itself a key player in CPO wafer-level test; for background see Earnings Call Highlights: FormFactor (FORM) | FY2026 Q2. The CFO added that the FormFactor investment is expected to close in Q4, subject to regulatory approval.
Another launch that matters for optical readers is Iris 100: an optical test solution for microLED and VCSEL emitters, serving both mobile and optical networking. Teradyne's optical test bet is not CPO-only; it also covers short-reach sources like VCSELs and potential microLED optical interconnect demand.
6. Management Outlook
Q4 FY2026 guidance:
Item | Guidance |
Revenue | $1,250M – $1,350M (midpoint ~+20% YoY) |
GAAP EPS | $1.88 – $2.18 |
Non-GAAP EPS | $1.95 – $2.25 |
Gross margin | 57.5% – 58.5% |
Operating expenses | 28.5% – 29.5% of revenue |
Non-GAAP operating margin | 28% – 30% |
At the $1.3B midpoint, full-year revenue would be about $5.2B, up more than 60%, and Q4 would mark the fourth straight quarter above $1.25B. The gross margin guide is about one point below Q3, which management attributes to product mix and new product launches.
Reading the language:
"We would expect Q1 to be some level of growth": the CFO chose "expect," explaining that visibility into Q1 is good but intra-quarter movement continues, so no number yet. Visibility has stretched from about 13 weeks at the start of the year to "a couple quarters."
"We expect each quarter next year to grow": the strongest line of the call, reconfirmed under follow-up that Q4 to Q1 should also grow. Against Q1 declining by double digits in 2022-2025, Smith stressed that the pre-2024 mobile-cycle pattern is "over"; the rhythm is now set by data center build schedules and fab capacity.
"ASIC programs will be back-half loaded": the only caveat, and it affects the quarterly shape of 2027, not the full-year direction.
New target model pushed to January: the previous model assumed a $12-14B ATE TAM with Semi Test at 80% of revenue. The TAM has grown too fast and its mix has shifted, so management will rebuild it against a much larger ATE TAM.
7. Supply Chain and Customer Clues
Test's share of semi capex is rising: from 6.4% in 2025 to 8.1% in the first eight months of 2026. The more complex the chip and the higher the yield bar, the larger the slice of each capex dollar that goes to test.
Inventory is a bet on 2027: inventory rose about $200M over six months (from $380M at year-end to $575M on the balance sheet). The CFO said roughly half is true demand and half is buffer against longer component lead times; DIO is 189 days versus a normal level of around 200.
Component lead times are stretching: parts used in testers, including the DDR4 an analyst mentioned, have longer lead times. The memory shortage is feeding back into the test equipment supply chain itself.
Partner list: probing and optics: FormFactor, Technoprobe, ficonTEC; interconnect: MultiLane (MLTP JV); physical AI: NVIDIA, Analog Devices, Scale AI, Generalist, Bright Machines, Sereact.
CPO back-end capacity is scaling: the "initial customer driving a 2027 ramp" Teradyne mentioned lines up with the back-end capacity and prepayment moves we tracked in CPO Capacity Up 15x — 2026 W43 Industry Weekly.
8. Conclusion
This quarter Teradyne proved one thing: AI-driven test demand is not limited to the GPU die. HBM unit counts, data center power chips, HDD capacity, backplane copper and server assembly are all growing, and Teradyne happens to have products at every step. Low compute share actually makes this structure more resilient; the real compute ramp waits until 2H27.
Three metrics to track next quarter:
The ATE TAM assumption in the January 2027 target model: the last model used $12-14B. A big upward revision would mean management sees this test capex cycle as structural expansion.
Capacity orders from hyperscaler #2: slated for 2H27; if they show up early in Q4 or Q1 orders, dual-vendor conversion is moving faster than expected.
How fast CPO test time comes down: whether the TAM lands at $300M or $700M depends on how far test time per optical engine can be cut. That is a shared challenge for the optical industry and test vendors.
The one-line takeaway: Teradyne isn't winning AI by grabbing GPU test; AI is pushing test demand across the entire supply chain, and Teradyne happens to stand at every step.
This article is for technology and industry trend analysis only and does not constitute investment advice.
Comments