$TER: A 70% Growth Quarter the Market Read as a Peak

The 60 second version

• Teradyne builds the machines that test chips before they ship, and AI data center silicon is now its biggest source of growth.
• The shift: buyers of AI accelerators and high bandwidth memory are moving from one test supplier to two, because one vendor’s capacity has been holding up their ramps.
• Its AI compute and memory test business sits in Flight on our curve. The easing constraint is test capacity plus a qualified second source, and Teradyne has roughly quadrupled UltraFLEXplus capacity in about 18 months.
• The evidence: Q3 2026 revenue of $1,306 million, up 70%, record memory test sales of $301 million, a second hyperscaler past qualification, $806 million of free cash flow in nine months, and no debt.
• The next test: Q4 results, Q1 2027 guidance and a new target earnings model, expected around January 27, 2027.
• What breaks it: a Q1 2027 guide midpoint below $1.20 billion, 2027 framed as flat or down, or the second hyperscaler’s capacity buys slipping out of 2027.

On October 27, Teradyne reported third quarter revenue of $1,306 million. That was 70% more than a year earlier and above the top of its own guidance range. Its fourth quarter guide also came in above consensus. Within a week the shares had fallen about 17%, and over the month they lost about 39%.

The market’s objection is easy to see. Revenue has now sat at about $1.3 billion for three quarters running, and gross margin is guided lower. For a capital equipment company in a cyclical industry, a flat line at the end of a steep climb looks like a summit.

We think that reading gets the timing wrong. The flat stretch is a gap between programs. The first hyperscaler’s next big ramp lands in mid 2027, the second has qualified and is expected to buy capacity in the second half of 2027, and a merchant GPU customer has several devices in development. Our thesis in one sentence: Teradyne is winning the second source position in AI chip test, the 2027 orders that would prove it have not been placed yet, and at about 21 times 2027 consensus earnings the price only asks for modest growth that the pipeline already points toward.

The risk is real. A digestion pause in AI spending could arrive before those programs do, and we set out that case in full below.

The shift

For most of the last two decades, semiconductor test was a line item that chipmakers tried to shrink. AI changed that. A modern accelerator is a huge piece of silicon wrapped in stacks of high bandwidth memory, and every one of those devices has to be tested at wafer, at package and often again as a full system before it reaches a data center rack. Complexity goes up, so test time per device goes up. Test capacity has become a bottleneck on the ramp.

The binding constraint for buyers was supply. One dominant supplier held most of the advanced compute test market, and its capacity and lead times were setting the pace of device ramps for some of the largest customers in the world. A hyperscaler spending billions on a custom chip does not want its launch date set by one vendor’s factory. So the big buyers began moving to dual vendor test, and that is the opening Teradyne has been walking through.

What eased the constraint was a combination of qualification and capacity. Over roughly 18 months to the third quarter of 2026, Teradyne has about quadrupled its UltraFLEXplus manufacturing capacity, and it has cleared qualification at a second hyperscaler. Memory adds a second push. Shortages of HBM are nudging designs toward lower stack heights, which means more units to performance test, and management says its share in that segment is higher than its share elsewhere.

We would place semiconductor test as a sector in the middle of its AI curve. The demand is visible and funded. What is still open is how the share settles between the two suppliers once the second source programs reach volume.

The company

Teradyne sells automated test equipment. Its UltraFLEXplus testers handle system on chip and compute devices, its Magnum line tests HBM, DRAM and NAND memory, and its Integrated Systems Test unit sells system level test. The customers are chipmakers, outsourced assembly and test houses, hyperscaler custom chip programs, GPU vendors and memory makers. It also runs a Product Test business, which includes hard disk drive test, and a Robotics arm built from Universal Robots and MiR.

In the third quarter, Semi Test brought in $1,072 million, up 77%. Within that, SoC test was $676 million, with compute up about 50% from a year earlier. Memory set a record at $301 million, its fourth straight quarter above $200 million, and Integrated Systems Test set its own record at $95 million.

Robotics did $96 million, up 28%, and made a profit. We will admit a soft spot here. For years the cobot business was the part of Teradyne that investors forgave rather than valued, a bet on factory automation that kept promising more than it delivered. A profitable quarter does not make it a growth engine, but it does mean the AI story no longer carries a drag behind it. Back to test.

The detail we keep returning to sits in the 10 Q. Over the first nine months of 2026, Teradyne bought back $173.5 million of its own stock at an average price of $349.62 a share. Today the stock trades at $249.32. Management paid about 40% more for its shares than the market now asks, and the board kept buying while the company was building inventory and repaying debt. Insiders can be wrong about price. It is still the clearest statement of what the people with the order book think the business is worth.

Where it sits on our curve

We put Teradyne’s AI data center test business in Flight. Our reason is the product’s own evidence, separate from the sector’s: the compute platform is in volume at one hyperscaler, qualified at a second, and engaged with a merchant GPU vendor, while memory test has delivered four consecutive quarters above $200 million. That is past Liftoff. It falls short of Orbit, because the second wave of compute customers has not yet turned qualification into capacity orders.

The buyer’s threshold is specific. A hyperscaler or GPU vendor needs a second tester that matches the incumbent on throughput and cost of test, passes qualification on its device, and then earns a capacity buy. The first hyperscaler is through all three steps. The second is through two, with capacity buys expected in the second half of 2027. The merchant GPU customer is in what management calls the fast follower phase, with several devices in development.

On our evidence sequence, the furthest step is expansion. Compute revenue grew about 50%, memory sales spread across HBM, DRAM and NAND customers, and the hard disk drive test unit added a third major customer. The commitment behind it is cash. Teradyne collected $806 million of free cash flow in nine months and repaid its entire revolver, so these were paid shipments rather than demonstrations or pilots.

Is it repeatable? We think the dual sourcing shift is structural, because the buyers have decided they will not be held to one supplier’s lead times again. The memory strength is less certain. Some of it rides the HBM shortage, and shortages end.

The numbers and what the street expects

Revenue was $1,306 million in the third quarter, up from $769 million a year earlier and slightly below the second quarter’s $1,329 million, according to the October 27 press release. The fourth quarter guide is $1.25 billion to $1.35 billion, roughly 20% growth on last year. That puts 2026 at about $5.2 billion against $3.19 billion in 2025, growth of about 64%.

Gross margin was 59.1%, up 60 basis points from a year earlier and down 70 basis points from the second quarter on mix. Management guided the fourth quarter to 57.5% to 58.5%, and in the prepared remarks it flagged supply cost inflation heading into 2027. That is the number we watch most closely. If margin keeps slipping while revenue holds flat, the peak reading gains weight.

The balance sheet is clean. Cash and investments were $625 million at September 27, up 21% from the prior quarter, and short term revolver borrowings fell to zero from $200 million at the end of 2025. Free cash flow was $227 million in the quarter and $806 million for nine months, up nearly 250%. Diluted shares fell 1.0% to 157.5 million.

Working capital is the line that does not fit the story cleanly. Since December, receivables have risen to $1.11 billion from $787 million, inventory to $575 million from $380 million, and prepayments to $614 million from $428 million. That is about $700 million tied up in the business. Some of it is a company building ahead of demand it expects. Some of it could become a problem if that demand slips.

Now the forward picture. Yahoo’s aggregate has fourth quarter consensus at $1.31 billion of revenue and $2.12 of EPS, against management’s non GAAP EPS guide of $1.95 to $2.25. For the first quarter of 2027, consensus sits at $1.40 billion and $2.35, and that EPS figure was reported at $2.50 a week earlier. Full year 2026 consensus is $5.23 billion and $9.44, and 2027 is $6.52 billion and $11.99, which is about 25% growth.

The direction of revisions is up, which surprised us given the share price. Over the last seven to thirty days, full year 2026 EPS moved from $9.10 to $9.44 and 2027 from $11.37 to $11.99, and Zacks shows current year consensus up 2.8% in 30 days. Analysts trimmed price targets after the print while raising earnings estimates, according to TIKR’s summary of broker notes. We could not confirm the analyst count or the high to low spread from a primary source.

Management has said three things about 2027. It expects sequential growth going into the year, it expects “healthy growth” across the year, and it will publish an updated target earnings model on the fourth quarter call. None of that is a number. The fourth quarter guide midpoint of $1.30 billion sits right at consensus, so the gap is in 2027, where the street has penciled in 25% growth that management has not quantified and that leans on custom chip programs weighted to the second half.

Why it wins

Follow the money. The customer pays for tester systems as capital expenditure, priced per test cell at several million dollars each, and then pays recurring service and software fees on the installed base. Each new unit brings bill of materials costs, supply costs that management expects to rise in 2027, factory capacity, and the working capital that sits in receivables and inventory while the cell is built and shipped.

Today Teradyne keeps most of the benefit. Capacity is tight, test times rise with every generation of device, and gross margin holds at 58% to 59%. The longer view is less comfortable. The same dual sourcing that lets Teradyne in also gives buyers a second quote to wave at both suppliers, so pricing power should erode as capacity catches up. We expect that pressure in 2028 and later, not next quarter.

Would a much better, much cheaper AI model hurt? We think it helps. Cheaper inference means more chips deployed, more chips means more devices to test, and more complex chips mean longer test times per device. Teradyne sells shovels to the people building the models, and better models increase the demand for shovels.

The moat, tested honestly, is narrow and deep. Teradyne is one of only two suppliers with qualified, high volume platforms for advanced compute and HBM test. Its IG XL software and the installed UltraFLEX base are sticky, because test programs are written for a platform and rewriting them costs engineering months. Its open approach to probing, handling and interface partners lets customers keep control of their own supply chains, which matters to a hyperscaler trying to escape single supplier dependence. And it has already passed hyperscaler qualification twice. That last one is hard to buy.

What could go wrong

The strongest bear case runs like this. 2026 was a one time AI capacity build stacked on top of a memory cycle peak, and some of Teradyne’s gains were overflow orders from a competitor that simply could not ship fast enough. On that reading the share gains are partly borrowed, and they get handed back once the incumbent’s factories catch up. Three flat quarters and a fourth quarter EPS guide that does not step up fit this story well.

We take it seriously. We have no evidence for the overflow piece, and nothing in the filings supports it, but nothing rules it out either. What separates the two readings is observable. A peak would show compute and memory falling together and lead times shortening. Our reading needs the first quarter guide to show sequential growth and the second hyperscaler and GPU programs to become capacity orders.

The timing risk is the one that worries us most. Compute share gains are weighted to the second half of 2027. If AI capital spending pauses for digestion in the first half, or memory capex rolls over, Teradyne could post two or three soft quarters before the new programs arrive. Estimates would come down, and a stock that has already lost 39% in a month could lose more.

Concentration is a blind spot. We could not find a 10% customer disclosure in the 10 Q, and the company does not name its hyperscalers or its GPU customer. A delay at any one of them would show up quickly in a business this lumpy.

Financing is not a worry. With $625 million of cash and investments, no debt and $806 million of nine month free cash flow, Teradyne funds itself to the January test and well beyond it. The cash question is working capital, and whether that $700 million build turns back into cash or into writedowns.

The investment thesis

Teradyne is becoming the qualified second source for AI compute and HBM test, and the market is pricing a pause between programs as the end of the cycle.

The chain holds together. The constraint is easing because buyers have chosen dual vendor test and Teradyne has the capacity to meet them. Adoption is responding: one hyperscaler in volume, a second qualified, a GPU vendor engaged, memory at a record. The company is keeping the value, with gross margin near 59% and free cash flow up nearly 250%. And the financing leaves the full result for shareholders, with no debt, a falling share count and no need to raise money.

Now reverse the price. At $249.32, with an enterprise value of about $38.7 billion, the stock trades at about 20.8 times 2027 consensus EPS of $11.99 and about 26 times 2026’s roughly $9.44. At a through cycle multiple of 22 to 23 times, today’s price needs about $11 of 2027 EPS. That means roughly $5.9 billion to $6.0 billion of 2027 revenue, growth of 13% to 15%.

Our downside case has 2027 revenue flat at $5.2 billion, EPS near $9, and a 20 times multiple, which implies around $180 a share. Our central case is $6.0 billion of revenue and about $11 of EPS at 23 times, which implies about $253, close to where the stock trades. The upside case is consensus, $6.5 billion and about $12.50 of EPS at 28 times, which implies about $350, roughly where the company was buying back stock this year.

The gap is this. The evidence points to growth in 2027 that starts in the second half, and the street’s 25% assumes more than management has said. The price sits in between and asks for 15%. We think 15% is achievable on the programs already qualified, without the GPU customer contributing much, and that is why our central case supports today’s price.

The next test is the fourth quarter report, expected around January 27, 2027. The thesis strengthens if fourth quarter revenue comes in at $1.31 billion or better, the first quarter guide midpoint is $1.38 billion or higher, the new target model implies at least $6.0 billion of 2027 revenue, and memory stays above $250 million a quarter. It breaks if the first quarter guide midpoint falls below $1.20 billion, if 2027 is framed as flat or down against about $5.2 billion, or if the second hyperscaler’s capacity buys move out of 2027.

What we are watching

• Around January 27, 2027, the Q4 2026 report: revenue of $1.31 billion or more strengthens the case, while a print near the $1.25 billion bottom of guidance with soft commentary weakens it.
• Same date, Q1 2027 guidance: a midpoint of $1.38 billion or higher confirms sequential growth, and a midpoint below $1.20 billion breaks the thesis.
• Same date, the updated target earnings model: implied 2027 revenue of $6.0 billion or more supports our central case, and anything framing 2027 as flat or down against about $5.2 billion breaks it.
• Same call, memory and margin: memory above $250 million a quarter and fourth quarter gross margin inside the 57.5% to 58.5% guide are what we need, while memory falling alongside compute would point to a cycle peak.
• Late April 2027 (estimated), the Q1 2027 report: we want the second hyperscaler’s capacity buys still on track for the second half of 2027 and receivables and inventory starting to convert to cash, and a delay in those buys or a further working capital build would count against us.

The bottom line

The shift in AI chip test is that the biggest buyers no longer accept one supplier. Teradyne has spent 18 months building the factory and passing the qualifications to be the second, and the cash flow says customers are paying for it. Three flat quarters reflect the calendar of programs, and the next set lands in 2027.

The market has priced that pause as a peak, which leaves the stock needing only mid teens growth next year. We think the pipeline supports that, and the company’s own buybacks at nearly $350 a share suggest management does too.

What would change our mind is a first quarter guide that steps down instead of up, with compute and memory falling in the same quarter. If that happens in January, the bears read the curve better than we did.

Not financial advice.

References

1. Teradyne, Inc. Teradyne Reports Third Quarter 2026 Results. Teradyne Investor Relations. Published October 27, 2026. Accessed November 4, 2026. https://investors.teradyne.com/news-events/press-releases/detail/455/teradyne-reports-third-quarter-2026-results
2. Teradyne, Inc. Q3’26 Earnings Prepared Remarks. Teradyne Investor Relations. Published October 28, 2026. Accessed November 4, 2026. https://d1io3yog0oux5.cloudfront.net/_72d0b4ca9e4a1a38aea2af8ee1b50286/teradyne/db/938/9658/webcast_transcript/Teradyne_Q3%2726_Earnings_Prepared_Remarks.pdf
3. Teradyne, Inc. Form 10 Q for the quarterly period ended September 27, 2026. US Securities and Exchange Commission. Published October 2026. Accessed November 4, 2026. https://investors.teradyne.com/sec-filings/all-sec-filings/content/0001193125-26-441091/ter-20260927.htm
4. TIKR. Teradyne Lost 17% in a Week. UBS Still Sees 61% Upside. TIKR.com. Published November 3, 2026. Accessed November 4, 2026. https://www.tikr.com/blog/teradyne-lost-17-in-a-week-ubs-still-sees-61-upside
5. Zacks Equity Research. Down 38.9% in 4 Weeks, Here’s Why You Should You Buy the Dip in Teradyne (TER). Yahoo Finance. Published November 2, 2026. Accessed November 4, 2026. https://finance.yahoo.com/markets/stocks/articles/down-38-9-4-weeks-133501708.html