Teradyne

Phase A — Understand the business

Lens 1 · Company Overview

Teradyne (founded 1960, Massachusetts corp, ~6,600 employees as of 2025-12-31, ~600 contractors) is the #2 global player in automated test equipment (ATE) and a meaningful player in industrial robotics. It sells the machines that test chips and electronics after they're fabricated — the "did this $40,000 AI accelerator actually come out functional?" gate at the end of the semiconductor line.

Three reportable segments (restructured March 2025 — Product Test was carved out as a new segment by combining production board test, defense/aerospace, and wireless test):

• Semiconductor Test — FY2025 $2,523.7M, 79% of revenue. FLEX/UltraFLEXplus (SoC/compute), Magnum (memory/HBM/DRAM/flash), ETS/Eagle (analog, SiC/GaN power), plus system-level test (SLT) and HDD/SSD test. This is the AI-leveraged crown jewel.
• Product Test — FY2025 $358.0M, 11%. Circuit-board test, wireless test, PIC (silicon-photonics) test, and defense/aerospace instrumentation.
• Robotics — FY2025 $308.3M, 10%. Universal Robots (cobots — 110,000+ sold since 2008) + Mobile Industrial Robots (AMRs — 11,000+ sold).

Business model / payment terms: hardware sold at a point in time on contractually stated prices (ASC 606); ~83% products / ~17% services (extended warranty, PCS, parts). No customer returns/refunds. Standard 12-month warranty. The test business is capital-equipment, cyclical, and customer-concentrated — not recurring.

Customers / concentration (Lens 13 ammunition): In FY2025 the top 5 direct customers = 44% of revenue (up sharply from 36% in 2024, 32% in 2023). Two customers each specified >10% (12% and 10%); one direct customer purchased 19% of consolidated revenue. ~89% of revenue is outside the US (Taiwan 36%, China 14%, Korea 14%, US 11% in FY2025).

CEO: Gregory (Greg) Smith, CEO since 2023-02-01 (16-year Teradyne veteran, ex-President of the Semiconductor Test Division).

Lens 2 · Supply Chain

Teradyne sits in the middle of the chip value chain — it's a tool vendor to the people who make and test chips, and its own product is assembled by outsourced manufacturers.

Upstream (inputs into Teradyne):

• Electronic & mechanical components from "a wide range of suppliers." Majority multi-sourced, but certain items are sole-sourced — a named chokepoint risk in its own 10-K ("temporary adverse impact if any of our sole source suppliers delay or cease to deliver").
• Contract manufacturers for the test businesses, with "significant operations in Malaysia." Robotics is built in-house in Denmark and the US.
• 2025 bolt-ons added supply/IP nodes: AET (ex-Infineon ATE team, Regensburg Germany, $18.3M) into Semi Test; Quantifi Photonics (NZ-based PIC test, $127.2M) into Product Test.

Teradyne → end customer (who buys the testers):

• IDMs (e.g. Infineon — also a customer; the AET deal "strengthens the relationship"), fabless designers, foundries (TSMC etc.), and OSATs (outsourced assembly/test — ASE, Amkor, etc., who buy the broadest FLEX fleet).
• A "specifying customer" (an OEM/IDM/fabless that picks the platform) and a separate "purchasing customer" (who places the order) — meaning a hyperscaler/chip designer can drive Teradyne demand without being the named buyer. The 19% direct purchaser in 2025 "includes certain revenues specified by other customers."
• Geographic exposure: the devices Teradyne's tools test are "fabricated and tested by foundries and subcontractors in Taiwan, China, Korea." Taiwan jumped to 36% of FY2025 revenue (21% in 2024) and 41% in Q1-2026 — i.e. the AI build-out is routing through Taiwan/TSMC-adjacent test capacity.

Chokepoints / single-source dependencies named: sole-source components (admitted); Malaysia contract-manufacturing concentration; China export-control exposure (US Dept of Commerce restrictions "have limited our sales and likely will continue to limit sales to certain customers").

New supply node forming: MultiLane Test Products JV (announced 2026-01-29, ~$157M for 75%, expected close H1-2026) — to build test for high-speed I/O in AI data-center equipment.

Lens 3 · Competitive Advantages (moats)

The structure is a duopoly, and Teradyne is the #2.

• SoC/compute test: Teradyne historically the share leader on the FLEX/UltraFLEXplus platform; it claims ~50% share of compute "VIP" (vertically-integrated producer) testing — i.e. the custom-ASIC accelerators hyperscalers design.
• Memory test: Teradyne is the laggard here — Advantest holds an estimated 60–70% of memory test (the T5500/T5800 lines dominate DDR5/HBM); Teradyne's Magnum is gaining share off a smaller base ("share gains in HBM and DRAM final test" per its own 10-K) but does not lead.
• SoC overall: Advantest ~35–45%, so the two split the SoC market with Teradyne ahead in compute but behind in some mobile/RF.

Durable moats (real but narrow):

1. Switching costs / installed base. A chipmaker's test programs (IG-XL software, multi-site recipes) are written to a platform; re-validating on a rival tester is costly and slow. The FLEX architecture is "widely used at OSATs," who leverage capital across device types. This is the genuine moat — sticky software + qualified hardware.
2. Duopoly scale / R&D intensity. FY2025 E&D was $504.6M (15.8% of revenue) — a fixed-cost barrier a sub-scale entrant can't clear.
3. Co-development proximity to chip designers. The AET deal (Infineon) and the specifying-customer model embed Teradyne in customer roadmaps.

Bargaining power — the weak spot. With the top 5 at 44% and one customer at 19%, the customers hold the power, not Teradyne. Customer consolidation is explicitly flagged as a risk that could "exert increased pressure on our prices." Teradyne is a price-taker to a handful of mega-buyers.

Robotics has no demonstrated moat — it competes against KUKA, ABB, FANUC, Yaskawa, Omron, and a swarm of low-cost Asian cobot makers (Techman, Doosan, Jaka, AUBO), and it lost money (see Lens 4).

Lens 4 · Segments (the single most important lens for this name)

Revenue by segment, FY2025 vs FY2024:

Segment income (loss) BEFORE taxes — FY2025:

This is the whole story in one table: Teradyne's consolidated 20.4% operating margin is a blend of a ~28%-margin Semi Test business carrying a Robotics unit that lost ~$99M in 2025 (worse than the $56.5M of revenue it shed). Strip Robotics out and the "real" Teradyne is a high-margin AI-test pure-play; keep it in and ~$100M/yr of profit is being burned chasing a cobot market that won't inflect.

The Q1-2026 inflection:

Two things changed at once: (1) Semi Test roughly doubled on AI compute, throwing off ~42% incremental PBT margin; (2) Robotics' loss collapsed from $(37.2)M to $(1.0)M as the 2025 restructuring (~400 heads, $24.3M severance) + revenue recovery finally bit. If Robotics breaks even and Semi Test holds, the consolidated margin re-rates structurally higher. Geography Q1'26: Taiwan 41%, Korea 19%, China 11% (down from 19% — export controls biting), US 7%.

---

Phase B — Measure performance

Lens 5 · Earnings Result (latest print = Q1-2026, three months ended 2026-03-29)

The numbers:

• Revenue $1,282.5M, +87.0% YoY (vs $685.7M). Largest single quarter in company history.
• Products $1,143.0M; Services $139.5M.
• Gross profit $780.9M, GM 60.9% (vs 60.6% — up despite the volume surge; mix + Semi Test leverage).
• Operating expenses essentially flat in dollars ($307.9M vs $294.5M) while revenue nearly doubled → income from operations $473.0M, 36.9% operating margin (vs 17.6%). This is the operating-leverage story made explicit: S&A fell to 13.0% of revenue (from 22.9%), E&D to 10.6% (from 17.2%).
• Net income $398.9M; GAAP diluted EPS $2.53 (vs $0.61, +315%); non-GAAP EPS $2.56.
• Effective tax rate 13.3% (vs 12.2%; higher on Pillar Two + lower credits).

Context that matters: one quarter of $2.53 GAAP EPS compares to full-year FY2025 GAAP EPS of $3.47 — Q1'26 alone did ~73% of all of last year's earnings. AI-related demand was ~70% of Q1'26 revenue (up from ~60% in Q4'25); SoC $882M, memory $203M (HBM/DRAM).

Guidance — the catalyst for the selloff:

• Q2-2026 revenue $1.15–1.25B; non-GAAP EPS $1.86–2.15. Both are sequentially DOWN from Q1's $1.282B / $2.56. Management framed it as digestion/lumpiness in AI test orders, not a demand break — but the tape read it as a near-term peak.

Balance-sheet flags:

• The one real flag is in FY2025 working capital: accounts receivable jumped +$305.6M (to $786.9M) — revenue was heavily back-end-loaded, so FY2025 reported revenue ran ahead of cash collection. By Q1-2026 the revolver was fully repaid ($200M → $0) and cash+securities sat at $394.0M, so the AR build is converting. Worth watching that DSO doesn't keep climbing into a peak.
• Inventory provision $25.8M in FY2025 (E&O, "downward revisions to previously forecasted demand" — $17.5M of it in Semi Test) — a small but real reminder this is still a cyclical capital-equipment business even amid the AI boom.

Market reaction: stock fell ~14% on the print (from ~$380 prior close to ~$327), and ~17% on the week — a record quarter sold off hard because the forward guide and a ~54x P/E left no room.

Lens 6 · Earnings Calls (sentiment trend)

No transcripts on disk (transcripts/ empty) — sentiment reconstructed from `` call coverage, FY2025-Q1 through Q1-2026:

• Tone trajectory: cautious-recovery (2023 trough) → "AI is real" (2024) → outright bullish on compute/VIP (H2-2025) → record-but-guarded (Q1-2026).
• Phrases that recur and escalate: "AI-related demand," "compute," "VIP customers," "HBM/DRAM share gains," "operating leverage." AI went from a sub-theme to "the bulk of our revenues."
• Things they stopped saying / softened: the Robotics growth story has been demoted from a headline ("invest to drive high long-term growth in Robotics," 2023) to a restructuring/break-even story by 2025–26. The narrative center of gravity moved entirely to Semi Test.
• New on Q1-2026: product launches — Photon 100 (silicon-photonics test) and Omnyx (server-board test) — management leaning into AI-data-center adjacencies beyond the core SoC tester.

Lens 7 · Comps

Peer set = the ATE duopoly + adjacent test/instrument names. Multiples are `` with the read date or n/a. No multiple is fabricated.

Read: The whole ATE/test complex is priced for the AI test cycle — Cohu at ~58x EV/EBITDA, TER re-rated to ~54x P/E, Advantest ~31x EV/EBITDA. TER is not obviously cheap or expensive vs the group; it's expensive vs its own history (P/E ~2.5x its early-2025 level). The P/E is also TTM and therefore understates the forward multiple's collapse if 2026 EPS doubles — on a forward basis TER is cheaper than the trailing 54x implies. The honest statement: valuation depends entirely on whether 2026's run-rate is a new plateau or a cycle peak. TTM P/E says "priced for perfection"; forward P/E on annualized Q1'26 says "reasonable for hypergrowth." That gap is the debate.

Lens 8 · Stock-Price Catalysts (what moves TER >5%)

Pattern over the last ~5 years — TER trades on the semiconductor test CYCLE and on AI-demand signals, with guidance (not the print) as the trigger:

• 2021: rode the post-COVID peak to ~$3.70B revenue / $5.98 non-GAAP EPS — multi-year high.
• 2022–2023 downcycle: Test revenue −14% YoY in Q3'22; 3nm ramp delays; FY2023 revenue −15% to $2.68B. Multiple compression — the classic ATE bust.
• 2024 recovery: AI memory + networking + Robotics drove Q1'24 above guidance — the AI-test narrative is born.
• 2026-04 (the latest big move): record Q1 beat, but −14% on the day / −17% on the week because Q2 guidance stepped down sequentially.
• 2026-06: +7.19% on Jun 20–21; added to the Nasdaq-100 effective Jun 22, 2026 (index-buying catalyst). Stock ~$361–380 range.

Lesson for the name: the market reacts to forward guidance and AI-demand commentary far more than to the reported quarter, and to the direction of the cycle. A great print with a soft guide sells off; a mediocre print with a "return to growth" guide rallies. High beta (single-day moves of ±7–17% are routine).

---

Phase C — Judge people & books

Lens 9 · Management

• CEO Gregory Smith (since 2023-02-01). Track record: 16-year insider who ran the Semiconductor Test division — i.e. he was the architect of the exact business now winning the AI cycle. He inherited the company at the 2023 cyclical trough and has presided over the recovery to a record Q1-2026. Archetype: professional operator / insider promote, not a founder. Skin in the game is salary+equity (PRSUs tied to relative TSR vs NYSE Composite and to 3-yr cumulative non-GAAP PBIT-margin) — i.e. comp is genuinely performance-linked, which is a positive governance signal.
• Capital-allocation history — disciplined and shareholder-friendly, arguably too aggressive into the peak:
• Buybacks $702.1M in FY2025 (vs $198.6M FY24, $397.2M FY23) + dividends $76.3M = ~$778M returned, which exceeded FY2025 FCF (~$450M) and was partly funded by drawing the revolver and running cash down from $553M to $294M. Share count fell to 156.1M (from 161.7M). Buying back stock with borrowed money near a cyclical/AI peak is the one capital-allocation decision a bear can fairly question — though by Q1-2026 the revolver was repaid in full, so it was effectively a bridge.
• M&A: bolt-on and adjacency-focused, not empire-building — AET ($18.3M), Quantifi ($127.2M), the MultiLane JV ($157M/75%), and a 10% strategic stake in Technoprobe (equity-method investment $537.1M on the balance sheet; contributed $19.9M of equity earnings in FY2025). These extend the test franchise (PIC, high-speed I/O, probe cards) rather than diversify away from it. Reasonable.
• Red flags: none material. Comp is performance-tied; M&A is on-strategy; the only critique is pro-cyclical buyback aggressiveness. The Robotics money-pit is a strategic/capital-allocation question (why keep funding a ~$100M/yr loss?) but management is visibly restructuring it toward break-even, which Q1-2026 validated.

Lens 10 · Forensic Red Flags

Forensic pass across income statement, balance sheet, cash flow. Grounded in the filings; every figure labeled.

• Revenue recognition: clean and conservative — point-in-time on shipment/delivery under ASC 606, "acceptance is typically a formality," no returns/refunds allowed. Services deferred and recognized over the contract. Low aggressiveness risk.
• Receivables vs revenue (the one thing to watch): AR grew +$305.6M in FY2025, far outpacing the +13.1% revenue growth — a back-end-loaded year. Not fraud-flavored (it's lumpy capital equipment), but it means FY2025 GAAP revenue ran ahead of cash. OCF $674.4M still slightly exceeded net income $554.0M, so earnings quality at the cash level is fine; the AR build is the item to keep watching into a potential peak.
• Inventory: net inventory +$81M to $379.6M; E&O provision $25.8M booked (up from $18.9M); $151.8M of cumulative inventory reserves carried with "no pre-determined timeline to scrap." On a doubling-revenue ramp, rising inventory is expected, but the reserve build is a quiet cyclical tell.
• SBC vs non-GAAP: stock-based comp $64.0M in FY2025 (~2% of revenue, ~12% of net income) — modest for a tech name; non-GAAP EPS ($2.56 in Q1'26) sits only ~$0.03 above GAAP ($2.53), so the non-GAAP "flattering" is small and honest. Good.
• Goodwill/intangibles: Goodwill $521.0M, acquired intangibles $51.3M — both rose on 2025 M&A but are small vs $4.18B total assets and $2.80B equity. Impairment risk low. The $537.1M Technoprobe equity-method stake is the largest single "off-P&L" asset and carries mark-to-thesis risk if probe-card demand turns.
• Leverage: minimal. $200M revolver draw at FY2025 year-end was fully repaid by Q1-2026; the company is effectively net cash (cash+securities $394.0M, ~$0 debt as of 2026-03-29).
• Tax: low effective rate (FY2025 ~12.1%; Q1'26 13.3%) leans on a Singapore tax holiday (renewed Dec 2025 through 2035; saved $21.6M / $0.14 per share in FY2025) — a real but disclosed and now-extended dependency; Pillar Two is a creeping headwind.

Regulatory findings (required sub-section).

• SEC Litigation Releases: none naming Teradyne (2021-06-24 → 2026-06-24). Verified via EDGAR EFTS.
• SEC AAERs (accounting/auditing enforcement): none. Verified via EDGAR EFTS.
• Item 3 / Note O Legal Proceedings (company's own disclosure): "Teradyne is subject to various legal proceedings and claims which have arisen in the ordinary course of business such as... patent, employment, commercial and environmental matters. Teradyne believes that it has meritorious defenses... and that the potential losses... are unlikely to have a material adverse effect." — i.e. boilerplate, nothing material.
• Non-SEC (FTC/DOJ/FDA/export): no material enforcement actions surfaced; the live regulatory exposure is US export controls on China (a commercial drag — "has limited our sales and likely will continue to" — not an enforcement finding), plus DFARS compliance for the defense/aerospace business.
• Conclusion: No material regulatory or legal findings — verified via SEC EDGAR EFTS (LR, AAER), the FY2025 10-K Note O, and web search as of 2026-06-24.

---

Phase D — Project & stress-test

Lens 11 · Forward Projection (FY2026E / FY2027E / FY2028E EPS)

Built bottom-up from FY2025 actuals + Q1-2026 actual + Q2-2026 guidance. Every input labeled; outputs are `` with the arithmetic shown. Fiscal year = calendar year.

Anchors: FY2025 GAAP EPS $3.47. Q1'26 GAAP EPS $2.53. Q2'26 guide: rev $1.15–1.25B, non-GAAP EPS $1.86–2.15 → GAAP midpoint ≈ ~$1.95.

• FY2026E — Base ≈ $8.40 GAAP EPS.
• Q1 actual $2.53 + Q2 guide-mid ~$1.95 = $4.48 first half. H2 assumed roughly flat-to-Q2 run-rate as AI test digests (no further sequential ramp baked): ~$1.95 + ~$2.00 = ~$3.95. Total ≈ $8.43. Round to $8.40.
• Revenue implied ≈ $1.28B + $1.20B + ~$1.20B + ~$1.25B ≈ $4.9–5.0B (+~55% YoY).
• FY2026E — Bull ≈ $9.50. H2 re-accelerates (AI orders resume sequential growth, Robotics turns positive); ~$2.40 avg H2 quarters.
• FY2026E — Bear ≈ $7.20. Q2 marks the peak; H2 quarters fade to ~$1.60 on a sharper AI digestion / China export hit.
• FY2027E — Base ≈ $9.00 (+7% on FY26 base): assumes the AI test plateau holds and modest growth resumes; Bull ~$11.50 (custom-ASIC TAM keeps compounding, memory-test share gains compound); Bear ~$6.00 (a genuine ATE downcycle as 2026 over-ships). The cyclicality cuts both ways — note FY2021's $5.98 non-GAAP peak was followed by a multi-year decline.
• FY2028E — Base ≈ $10.00; Bull ~$14; Bear ~$5.50. Widening cone reflects how cycle-dependent the out-years are.

The honest framing: at ~$361–380, FY2026E base ~$8.40 → forward P/E ~43–45x, falling to ~40x on FY2027E base ~$9.00. That is not the ~54x trailing headline — but it still requires the AI test plateau to hold. The single swing factor is whether 2026 is a plateau (multiples are defensible) or a peak (a cyclical EPS decline + multiple compression compounds downward, as it did in 2022–23).

(Brier forecast NOT logged — --watchlist unattended; the projection is committed in-dossier but not promoted to a tracked forecast, per skill rules.)

Lens 12 · Bull vs Bear

Bull case. Teradyne is the cleanest public pure-play on the physical test of AI silicon — every custom ASIC and HBM stack a hyperscaler designs has to be tested, and Teradyne holds ~50% of compute VIP test. The model just demonstrated extreme operating leverage (Q1'26: revenue +87%, opex roughly flat, operating margin 17.6%→36.9%), so incremental AI revenue drops to the bottom line at ~40%+ margins. Robotics — a ~$100M/yr drag — just went to break-even, which alone re-rates consolidated margins. Capital allocation is disciplined (net cash, ~$778M returned in FY2025, on-strategy bolt-ons). The custom-ASIC wave (shipments growing ~45% in 2026 vs ~16% for GPUs) structurally favors Teradyne's compute-test stronghold over the GPU-test relationship Advantest monopolizes. New adjacencies (silicon-photonics test via Quantifi/Photon 100, high-speed I/O via MultiLane, server-board test via Omnyx) extend the TAM.

Bear case (risks that could permanently impair or de-rate):

1. It's still cyclical, and 2026 may be the peak. Sequential Q2 guidance is down. ATE is a capital-equipment cycle — TER did $5.98 non-GAAP EPS in 2021 and then fell for two years. A ~54x trailing / ~43x forward multiple on possibly-peak earnings is the textbook setup for a de-rate.
2. Customer concentration is extreme and worsening — top 5 = 44% (was 32% two years ago), one customer 19%. The AI boom has made Teradyne more hostage to a handful of hyperscaler ASIC programs; if one in-house chip slips or moves test in-house/to Advantest, a double-digit revenue chunk evaporates.
3. Advantest, not Teradyne, owns the structural lead in memory test (60–70%) and the exclusive GPU-test relationship. If the AI test dollars concentrate in HBM/GPU rather than custom-ASIC SoC, Teradyne under-participates in its own bull market.

Pre-mortem (18 months out, thesis broke): It's late 2027. Hyperscaler ASIC orders digested through 2026, Q2'26 was the peak, and 2027 revenue is down ~20% as the test fleet over-built. The stock round-tripped from ~$380 to ~$200 on a cyclical EPS decline ($8.40 → ~$6) and multiple compression (43x → ~25x). The buybacks done at $150–180 look smart; nothing done above $350 does. Robotics is still sub-scale. The lesson: it was a cycle, priced as a secular.

Contrarian view (what the market may be refusing to see): the Q2-driven ~14% selloff may be over-reading normal AI-order lumpiness as a cycle top. If custom-ASIC test is genuinely a multi-year secular build (not a 2021-style inventory spike), then on FY2027 base EPS ~$9 the stock at ~40x is not the bubble the trailing 54x implies — and Robotics break-even is a free call option the bears are ignoring. The bull/bear hinges on one un-knowable: secular plateau vs cyclical peak.

Lens 13 · Devil's Advocate (short-seller)

Dismantling the bull case.

• Where revenue is concentrated and what breaks it: 79% of revenue and ~106% of profit (after Robotics' loss) is Semiconductor Test, and within it AI compute is ~70% of the company, sourced from a handful of VIP/hyperscaler ASIC customers, one of which is 19% of total revenue. The "test is mandatory" bull thesis is true but doesn't protect Teradyne specifically — the test dollars could go to Advantest. A single hyperscaler pausing or insourcing a flagship ASIC test program is a >10% revenue air-pocket.
• Why the moat is weaker than bulls think: the durable moat (test-program switching costs) is real for legacy sockets, but every new AI chip is a fresh socket up for grabs — and Advantest is winning the biggest one (GPU). Teradyne is defending share in mobile/legacy while fighting uphill in memory.
• Most dangerous competitor bulls underestimate: Advantest — already the share leader in memory and GPU test, the "ASML of test," and trading at a lower P/E (~21.6x vs ~54x). If you're bullish AI test, the better risk/reward may be the leader, not the #2.
• Worst capital-allocation move: $702M of buybacks in FY2025 funded partly by revolver draws, executed into a rising/AI-peak tape — buying high with borrowed money. Defensible (repaid by Q1'26) but pro-cyclical.
• Assumptions that must hold for today's price: AI test is secular not cyclical; 2026 is a plateau not a peak; the 19% customer stays; custom-ASIC keeps beating GPU; Robotics holds break-even. That's five things, and at least two (cycle direction, customer retention) are outside management's control.
• Valuation if growth disappoints 20–30%: FY2026 base EPS ~$8.40 → bear ~$6.0–6.5; pair with a cyclical de-rate to ~25–30x and you get ~$160–195, roughly −45% to −55% from ~$370. The downside is large precisely because both EPS and the multiple compress together in an ATE downcycle.
• Single scenario that permanently impairs: hyperscalers standardize AI-chip test on Advantest (or insource it), permanently capping Teradyne's compute-test share — turning a 50%-share leader into a #2 also-ran in the one market that matters. Plausibility: moderate-low (switching costs + co-development proximity argue against it), but it's the scenario that would break the thesis for good.

Lens 14 · Management Questions (ordered by information value)

1. Of Q1-2026's ~70% AI revenue, how much is concentrated in your single largest (19%) customer, and how durable are those VIP/ASIC test programs across the next two product generations?
2. Is the sequential Q2 step-down digestion/lumpiness or the start of a cyclical roll-over — what in your order book distinguishes the two?
3. What is your realistic share trajectory in memory/HBM test vs Advantest, and can Magnum/EPIC actually take share, or is that market structurally Advantest's?
4. As custom ASICs out-grow GPUs, does your compute-VIP share advantage widen — and conversely, how exposed are you if AI test dollars shift back toward GPU (Advantest's stronghold)?
5. What does normalized Robotics look like — sustained break-even, a path to segment margins that justify the capital, or a candidate for divestiture?
6. Why $702M of buybacks (partly revolver-funded) into a rising tape in FY2025 — what is the valuation framework that governs repurchase pace at these multiples?
7. How should we think about peak-to-trough amplitude this cycle vs the 2021→2023 decline, given how concentrated demand is in a few AI programs?
8. What are the gross-margin ceiling and the incremental (drop-through) margin you can sustain at $5B+ revenue?
9. How much FY2026 revenue is exposed to further US–China export-control tightening, and what's the realistic floor on the China business (14% → ?)?
10. What is the strategic logic and expected return profile of the MultiLane JV and the Technoprobe stake — and could either become a larger consolidation move?
11. How does the specifying-vs-purchasing-customer dynamic affect your pricing power as the top-5 concentration climbs toward 50%?
12. What is the silicon-photonics (Quantifi/Photon 100) test TAM and timeline — real revenue line or option?
13. How dependent is the FY2026 tax rate on the renewed Singapore holiday, and what's the Pillar Two drag trajectory?
14. What capex and capacity commitments has the Semi Test ramp locked in (you drew the revolver to fund it) — and what's the downside if AI orders soften?
15. What would have to be true for you to not be the test partner of choice for the next wave of hyperscaler custom silicon?

---