Semiconductor automated test equipment (ATE). Nasdaq TER / TSE 6857. Run 17 September 2026. Built from Teradyne 10-Ks FY2022 to FY2025 and earnings call transcripts through Q2 2026 (research folder sources/), and Advantest tanshin FY2023 to Q1 FY2026, results slides, notes and Q&A, and the 84th securities report (advantest.com). Events swept through 17 Sep 2026; most recent events checked: Teradyne India office and Goldman Sachs conference remarks (9 to 14 Sep 2026), Advantest completion of its 150bn yen buyback (20 Aug 2026). Figures in USD; Advantest converted at its own disclosed average USD/JPY for each year (143 FY2023, 153 FY2024, 150 FY2025). Advantest's fiscal year ends in March, one quarter after Teradyne's December year. Not a valuation and not a recommendation.
Three cells carry the next five years. The first is compute SoC shipped to Taiwan and other Asian OSATs: AI accelerators, custom ASICs, server CPUs and networking. The second is DRAM and HBM memory test in Korea. The third is optical and system-level test for data centers. Advantest grows with the first cell almost mechanically: it holds about 66% SoC share, and it is building capacity toward about 10,000 SoC systems a year (FY2025 and Q1 FY2026 Q&A). Teradyne's growth has a second driver on top of the market, which is share. Compute customers are qualifying a second test vendor for supply security; Teradyne expects "about 30%" of merchant compute over 3 to 5 years (Goldman Sachs conference, 9 Sep 2026, secondary summary; approx., unverified). It is also gaining share in HBM and DRAM final test, where Advantest itself expects to lose some share (TER 2025 10-K; Advantest Q1 FY2026 Q&A). Guidance points the same way for 2026. Teradyne's first half of $2.61bn is 50 to 52% of the year, which implies about $5.0 to 5.2bn, or +57 to 64% (inferred from Q2 2026 call). Advantest guides 1,714bn yen, +52% (Q1 FY2026 slides).
Advantest turns incremental SoC volume into profit through mix and scale. FY2025 sales rose 349bn yen on only 34bn yen more SG&A and 7bn yen more R&D, so EBIT margin went from 29.3% to 44.2% and gross margin from 57.1% to 64.3%. Q1 FY2026 gross margin reached 69.5% (tanshin; Q1 FY2026 slides). Teradyne's Semiconductor Test segment grew 19% in 2025, but its pre-tax margin moved only from 26.3% to 27.8%. Group gross margin has sat inside a 57 to 59% band for four years, because Teradyne is adding R&D and go-to-market spending to win second-source sockets, and its CFO cites "pricing elements" and memory mix as margin strains (2025 10-K; Q2 2026 call). That leaves an uncomfortable combination. Teradyne's growth edge comes from share won as the challenger, and challengers price to win; Advantest's margin edge comes from being the incumbent those customers are diversifying away from. Teradyne's own target model tops out at a 30 to 34% operating margin on about $6bn of revenue. Advantest guides 49.4% for FY2026.
Advantest's exposure is volume leverage concentrated in one end market. In FY2025, 95% of SoC tester sales were computing and communications and 50% of group sales shipped to Taiwan, and its margin moves with volume. In the last downturn, FY2023, sales fell 13%, gross margin dropped from 57.0% to 50.6%, and EBIT margin fell from 29.9% to 16.8%. Over the same stretch Teradyne's gross margin held at 57.4% (tanshin; TER 2023 10-K). Teradyne's exposure is the specifier-and-purchaser structure. Its top five direct customers rose to 44% of 2025 revenue from 32% in 2023; one direct purchaser took 19%, and two chip designers each specified 10 to 12%. Losing one hyperscaler program would cost more than a cyclical dip (TER 2025 10-K). On the evidence Advantest breaks first in a broad AI capex digestion. Teradyne breaks first if one named program moves, and its losing Robotics unit (-$99m pre-tax in 2025) is a cushion that costs money rather than one that pays.
Five cells carry most of the two companies' combined test economics; each score is argued in the tabs below. Neither company discloses product-by-region revenue, so the cells pair each product line with the region its sales ship to (inferred).
| Cell (product x region) | Teradyne | Advantest | Why (one clause, sourced) |
|---|---|---|---|
| Compute SoC x Taiwan and Asian OSATs | 3 | 5 | Advantest holds 66% SoC share, up from 56%, and ships 51% of sales to Taiwan (FY2025 notes); Teradyne's compute share is "stabilizing and starting to inflect", mostly from 2027 (Q2 2026 call). |
| Memory test (DRAM, HBM) x Korea | 4 | 3 | Teradyne is gaining HBM and DRAM final-test share with memory book-to-bill above 2 (10-K; Q2 2026 call); Advantest expects to lose memory share from about 60% (Q1 FY2026 Q&A). |
| China mainstream SoC and memory | 2 | 3 | China is 19% of Advantest sales and growing, though local vendors weigh on the low end; Teradyne says trade restrictions limit its competitiveness and it moved manufacturing out of China (10-K; FY2024 and Q1 FY2026 notes). |
| Non-compute SoC (auto, industrial, mobile) x US, Europe, Japan IDMs | 3 | 2 | This is about half of Teradyne's SoC sales and its historical stronghold, where it keeps share (Q4 2025 call); about 5% of Advantest SoC sales (inferred from 95% compute). |
| Data-center system, optical and storage test x US and Asia | 4 | 2 | Teradyne's IST grew 2.5x quarter on quarter in Q2 2026, and it added Quantifi, the MultiLane JV and TestInsight; Advantest's other systems fell 5% in FY2025, after a 21.4bn yen Essai impairment in FY2024. |
Scores are anchored to the exhibits in the three tabs; a score with no exhibit behind it does not ship. Scores are per cell and are not summed. The cells differ in size by a factor of ten and overlap where China crosses product lines.
Normalization. Teradyne reports three segments (Semiconductor Test, Product Test, Robotics) and splits Semiconductor Test into SoC, Memory and IST (2025 10-K Note E). Advantest merged Mechatronics into a Test System segment in FY2025, which it splits into SoC, Memory and Other systems, plus Services and Others (FY2025 tanshin; Q1 FY2026 slides). We map Advantest's "Other systems" (handlers, device interfaces) against Teradyne's IST as the nearest equivalent; the two are not the same product set. Neither company splits compute from non-compute SoC in its filings. We use Teradyne's call remark that 2025 was "nearly 50% compute" and Advantest's slide showing 95% of SoC sales in computing and communications, which overstates pure compute because it includes phone chips.
The headline contrast is where revenue sits. About 65% of Advantest's revenue is compute-oriented SoC test (inferred). Teradyne's compute SoC is about 28% of its revenue (inferred), and about 21% sits in Robotics and Product Test, which Advantest does not have at all.
| Cell | TER rev (% total) | TER 2y CAGR | ADV rev (% total) | ADV 2y CAGR | Margin signal / leader |
|---|---|---|---|---|---|
| SoC test (all) | $1,890m (59%) | +14.9% | $5,116m (68%) | +76.8% | Advantest; SoC drives its 50.9% segment margin |
| Memory test | $505m (16%) | +14.4% | $1,143m (15%) | +41.3% | Advantest by size, Teradyne on share momentum; both call memory margin-dilutive (Q2 2026 call; Q1 FY2026 Q&A) |
| System-level / other systems | $129m (4%) | -3.4% | $537m (7%) | ND (restated) | Mixed; Teradyne IST inflecting on HDD in 2026 |
| Services and others | ND | ND | $728m (10%) | +12.7% 1y | Advantest segment profit only 8.8bn yen, low margin |
| Robotics + Product Test | $666m (21%) | -3.7% | none | none | Teradyne only; Robotics pre-tax -$99.4m, Product Test +$60.7m |
| Ship-to region, latest FY | Teradyne | YoY | Advantest | YoY (yen) |
|---|---|---|---|---|
| Taiwan | $1,155m (36%) | +92% | $3,797m (51%) | +74% |
| China | $451m (14%) | +20% | $1,422m (19%) | +22% |
| Korea | $446m (14%) | -36% | $1,202m (16%, label inferred) | +15% |
| US / Americas | $361m US (11%) | -4% | $297m Americas (4%) | -6% |
| Europe | $215m (7%) | -14% | $154m (2%) | +16% |
| Japan | $66m (2%) | -59% | $167m (2%) | +59% |
Compute SoC. Advantest's driver is volume on an incumbent platform: rising test intensity from 3D packaging and longer final-test times, plus capacity going from 5,000 to about 10,000 systems a year, a plan it is pulling forward (FY2025 and Q1 FY2026 Q&A). Teradyne's driver is customers adding a second vendor. It has majority share at its first hyperscaler, is qualified as second source at a second, and shipped its first merchant GPU order in Q2 2026 (Q2 2026 call; Goldman conference, secondary). Advantest's driver lasts as long as AI capex does. Teradyne's lasts as long as customers keep valuing supply security enough to pay for correlating a second platform. Both are organic; Teradyne's small 2025 to 2026 acquisitions sit mostly outside SoC (Quantifi $127.2m, the Infineon ATE team $18.3m, 75% of the MultiLane JV for $157m, TestInsight undisclosed).
Memory. Teradyne's Magnum testers can also run logic tests, which lets memory makers switch the same tool between uses (Q2 2026 call). Teradyne expects the 2026 memory market to be 40% larger than 2025. Advantest's memory sales were 90% DRAM in FY2025, and it expects to lose share, especially in NAND.
Both sell direct. Teradyne sells test systems through its own sales force; only Robotics goes through distributors (2025 10-K). Advantest sells through its own regional subsidiaries; Advantest America and Advantest Taiwan each exceed 10% of consolidated sales (84th securities report). The difference is who holds the decision. At Teradyne a "specifying customer" (the chip designer) picks the platform and an OSAT or foundry buys it (10-K), so the pricing conversation is with the most powerful chip companies in the world. Advantest calls the V93000 the de facto standard in the foundry and OSAT space (FY2025 Q&A), so the purchaser's existing test floor defaults to it. Neither discloses a price and volume split.
| Teradyne | Advantest | |
|---|---|---|
| Manufacturing | Contract manufacturers Flex, Plexus and SAM Meerkat in Malaysia and Thailand, plus an internal FLEX line in Cebu (10-K) | Gunma factory plus outsourced partners; a contract manufacturer is ramping a second Southeast Asia plant (84th securities report; Q1 FY2026 Q&A) |
| Single-source risk | "Certain items are obtained from sole sources" (10-K) | Materials availability during expansion flagged; specific parts ND |
| Lead time / capacity | 12 to 16 weeks held while UltraFLEXplus shipments more than doubled (Q1 2026 call) | Target of about 10,000 SoC systems a year, being pulled in |
| Input cost pressure | Memory mix a margin "strain" into 2027 | DRAM inside testers is the "main headwind"; pass-through under evaluation |
Advantest estimates its CY2025 share at 66% of SoC and about 60% of memory testers, about 65% overall of a $9.0bn market (FY2025 notes). An analyst on Teradyne's Q2 2026 call put Teradyne near 37% of a test market defined to include burn-in, and management did not dispute the figure (approx., unverified; the two companies' market definitions do not reconcile). Teradyne's 10-K names Advantest, SPEA and Cohu as competitors. Chinese domestic vendors such as Hangzhou Changchuan are growing fast at home (36Kr, Aug 2026; approx., unverified), and Advantest said in April 2025 that they had weighed on its low-end SoC share.
Compute SoC: the mechanism is customers dual-sourcing. Advantest is more exposed because it holds the share being redistributed; a move toward 30% second-vendor share at merchant GPU would take roughly a fifth to a third of the socket volume away from the incumbent (inferred). Memory: DRAM price inflation raises tester input costs while HBM programs are lumpy; Advantest names DRAM as its main margin headwind, and Teradyne's memory growth dilutes its margin. China: export licensing and local substitution. Advantest is more exposed in dollars (19% of sales), and Teradyne has already lost ground there. Adjacencies: Robotics has lost money in each of the last three years (2023 -$54.3m, 2024 -$77.6m, 2025 -$99.4m pre-tax), which is Teradyne's self-inflicted drag.
1. VIP compute position on UltraFLEXplus. It held about 50% of vertically integrated compute in 2025 (Q4 2025 call) and has majority share at its first hyperscaler, having moved from qualification in 2023 (Goldman conference, secondary). Durability: Med. It is programme-specific and re-competed each chip generation.
2. Magnum memory test. It is taking HBM and DRAM final-test share (2025 10-K), and the logic-capable tool lets memory makers move it between uses. Durability: Med.
3. Data-center breadth across photonics, copper interconnect and HDD system-level test. It was largely bought (Quantifi, MultiLane, TestInsight), so durability is Low; money can copy it.
1. V93000 installed base at foundries and OSATs. It calls the platform the "de facto standard" (FY2025 Q&A), and its estimated SoC share rose from 56% to 66% in CY2025 (company estimate). Teradyne's own description of the 9 to 12 month dual-vendor correlation process corroborates the switching cost. Durability: High.
2. Capacity scale. It plans about 10,000 SoC systems a year and is weighing 15,000+ (Q1 FY2026 Q&A). Durability: Med, since capacity can be bought but takes years.
3. Memory incumbency at about 60% share, 90% of it DRAM. Durability: Med, and eroding by its own guidance.
Teradyne: in 2025 its top five direct customers were 44% of revenue (36% in 2024, 32% in 2023). One direct purchaser took 19%, two specifiers drove 12% and 10%, and two customers held 22% and 20% of receivables (2025 10-K). Advantest: no end customer above 10% was found in the 84th securities report or the tanshin, and its investor guide calls the base well-diversified. Its own Q&A, though, says US fabless SoC business is concentrated in Taiwan, so the true concentration is probably hidden by the OSAT layer (inferred). Evidence of pricing power: Advantest's gross margin went from 57.1% to 69.5% in five quarters with no price-driven share loss. Teradyne's CFO lists "pricing elements" among the margin variables (Q2 2026 call).
Teradyne runs an asset-light model on several contract manufacturers, with an in-house backup line in Cebu. Its gross margin stayed between 57.4% and 59.2% through the 2022 to 2023 component inflation and downturn (10-Ks), a sign of Partial-to-Strong pass-through. Advantest mixes in-house Gunma production with outsourcing and makes more of its memory testers in-house. Its gross margin fell 6.4 points in FY2023, which we read as volume absorption rather than input cost (inferred), and it now names DRAM costs as the main headwind, with pass-through only "under evaluation". Teradyne exited Chinese manufacturing in 2023 (about $1bn of equipment moved; Asia Financial, Jan 2024; approx., unverified), so it has already paid for geopolitical supply insurance.
| Cell | Route control | Pocket price | Continuity | Outcome (share / margin) | Confirming KPI |
|---|---|---|---|---|---|
| Compute SoC x Taiwan | Advantest default at OSAT; Teradyne via specifier | Advantest above Teradyne (inferred from 69.5% vs 59.8% gross margin) | Both outsourced in SE Asia | Advantest share flat to down, margin up; Teradyne share up, margin flat | Advantest CY2026 SoC share estimate, Apr 2027 |
| Memory x Korea | Direct, both; memory makers specify and buy | ND | DRAM input costs hit both | Teradyne share up, margin down; Advantest share down, margin down | Teradyne 2026 memory revenue above $800m (vs $505m in 2025) |
| China mainstream | Advantest direct; Teradyne constrained by trade rules | Parity to below vs local vendors | Licensing risk for both | Advantest share flat, margin flat; Teradyne share down | Advantest China share of sales staying at 18 to 20% |
The tension. Teradyne grows fastest in memory, where it has the least pricing power; both companies call memory margin-dilutive. It also wins compute share by being the second vendor, a role that customers value precisely because it caps the incumbent's pricing. Teradyne's growth therefore comes packaged with weaker net price by construction.
1. Installed base in compute SoC. Advantest holds about 66% share; Teradyne has one mature dual-vendor compute account. Impact: Major. Teradyne needs 3 to 5 years of socket-by-socket correlation, and that time cannot be bought. 2. Portfolio drag. Robotics lost $99.4m pre-tax in 2025, against Advantest's zero. Impact: Moderate, about 3 points of group margin, and fixable in 18 to 24 months by restructuring or exit, so it is not a moat either way. 3. Buyer structure. Teradyne's specifier model concentrates negotiating power; impact Moderate, and structural as long as AI silicon is designed by a handful of firms.
Mapping. Both present costs by function. Teradyne (US GAAP) reports cost of revenues excluding acquired intangible amortization, selling and administrative, and engineering and development, with restructuring and intangible amortization on separate lines. Advantest (IFRS) reports cost of sales and SG&A and no R&D line; R&D is disclosed separately (65.5bn, 71.4bn and 78.1bn yen). We assume R&D sits inside SG&A and subtract it (inferred; the placement was not confirmed in the securities report). If part of R&D sits in cost of sales, Advantest's true gross margin is higher still. EBIT is reported operating income: Teradyne 2024 includes a $57.1m gain on sale of a business, and Advantest FY2024 includes a 21.4bn yen Essai impairment. Fiscal years are offset by one quarter.
| % of sales, 3y avg | Teradyne CY23-25 | Advantest FY23-25 | Gap | What drives it |
|---|---|---|---|---|
| COGS | 42.0 | 42.7 | -0.7 | Averages hide Advantest's swing from 49.4 to 35.7 |
| R&D | 15.9 | 9.9 | +6.0 | Teradyne funds three segments and second-source compute efforts |
| SG&A (ex R&D) | 21.3 | 16.2 (inferred) | +5.1 | Robotics selling cost ($135m in 2025) and a smaller revenue base |
| Gross margin | 58.0 | 57.3 | +0.7 | Teradyne is stable; Advantest is volume-levered |
| EBIT margin | 20.1 | 30.1 | -10.0 | Opex intensity, then scale |
| Year | TER COGS | TER R&D | TER S&A | TER GM | TER EBIT | ADV COGS | ADV R&D | ADV SG&A | ADV GM | ADV EBIT |
|---|---|---|---|---|---|---|---|---|---|---|
| 2023 | 42.6 | 15.6 | 21.6 | 57.4 | 18.7 | 49.4 | 13.5 | 19.2 | 50.6 | 16.8 |
| 2024 | 41.5 | 16.3 | 21.9 | 58.5 | 21.1 | 42.9 | 9.2 | 15.9 | 57.1 | 29.3 |
| 2025 | 41.8 | 15.8 | 20.3 | 58.2 | 20.4 | 35.7 | 6.9 | 13.4 | 64.3 | 44.2 |
Raw values, Teradyne $m, 2023/2024/2025: revenue 2,676.3 / 2,819.9 / 3,190.0; cost of revenues 1,139.6 / 1,171.0 / 1,332.7; E&D 418.1 / 460.9 / 504.6; S&A 577.3 / 617.0 / 648.9; operating income 501.1 / 593.8 / 650.1. Advantest bn yen, FY2023/2024/2025: sales 486.5 / 779.7 / 1,128.6; cost of sales 240.5 / 334.6 / 402.5; SG&A 159.0 / 195.4 / 229.6; R&D 65.5 / 71.4 / 78.1; operating income 81.6 / 228.2 / 499.1.
The most persistent gap is operating expense intensity. R&D plus SG&A was 37.2%, 38.2% and 36.1% of sales at Teradyne against 32.7%, 25.1% and 20.3% at Advantest, so Teradyne is heavier in every year. The gap is not only Robotics. Teradyne's Semiconductor Test segment alone spent 26.6% of sales on E&D, selling and G&A in 2025 and earned 27.8% pre-tax, against 50.9% for Advantest's Test System segment. The mechanism is scale on a shared platform. Advantest spends 78bn yen of R&D (about $520m) across $7.5bn of sales, while Teradyne spends $505m across $3.2bn and three segments. Gross margin does not show a persistent gap: Teradyne was ahead in 2023, Advantest in 2025. This partly contradicts the power map, since Advantest's pricing power appears only at high volume, and it confirms Answer 3: Advantest's margin is operating leverage, which runs both ways. Teradyne's gap is part structural (subscale SoC) and part self-inflicted (Robotics), and a management focused on the second part could recover about 3 points.
| Metric | Threshold | By when | If it hits, it favors | Where published |
|---|---|---|---|---|
| Advantest SoC tester share, CY2026 estimate | 66% or more / below 62% | Apr 2027 | Advantest / Teradyne | Advantest FY2026 results notes |
| Teradyne compute customers at mature dual-vendor stage | 3 (from 1) | Q4 2027 call | Teradyne | Teradyne earnings calls |
| Advantest quarterly gross margin under DRAM cost pressure | 65% or more each quarter | Q4 FY2026 (Apr 2027) | Advantest | Advantest quarterly slides |
| Teradyne top-five customer concentration | above 50% | FY2026 10-K, Feb 2027 | Advantest (raises Teradyne fragility) | Teradyne 10-K, customer note |
| Teradyne memory test revenue | above $800m for 2026 | Feb 2027 | Teradyne | 10-K Note E |
| Teradyne Robotics pre-tax result | loss under $30m | FY2026 10-K, Feb 2027 | Teradyne | 10-K segment note |