Semiconductors
PrivateBest-positioned AI-infrastructure "picks-and-shovels" test name with a real moat and a re-accelerating order book — but priced for perfection at ~54x trailing / ~32x forward after a +100% year, with reported numbers flattered by a one-time IEEPA tariff refund and Spirent inorganic lift. Quality compounder, wrong entry; WATCH for a cyclical/multiple reset.
Research
The verdict
Best-positioned AI-infrastructure "picks-and-shovels" test name with a real moat and a re-accelerating order book — but priced for perfection at ~54x trailing / ~32x forward after a +100% year, with reported numbers flattered by a one-time IEEPA tariff refund and Spirent inorganic lift. Quality compounder, wrong entry; WATCH for a cyclical/multiple reset.
Primary sources
Source documents — open to read in full
What it is. Keysight is the world's largest pure-play electronic design-and-test company — the instruments, software and services that engineers use to design, validate and manufacture electronic systems. It is the direct descendant of the original Hewlett-Packard test-and-measurement business: HP → Agilent (1999 spin from HP) → Keysight (spun out of Agilent November 1, 2014). HQ Santa Rosa, CA; ~16,800 employees as of Oct 31, 2025.
How it makes money. Three revenue forms, sold mostly as integrated solutions:
Customers. Highly diversified across the value chain — semiconductor & chipset makers, network-equipment manufacturers, hyperscalers/cloud operators, telecom carriers, aerospace/defense primes and government research agencies, automotive OEMs + Tier 1/2 suppliers, and universities. No customer concentration is disclosed — the 10-K names no 10%+ customer, consistent with a long-tail base of R&D buyers. customers.csv is empty on the shelf.
Contract structure. Largely transactional capital-equipment sales (recognized on shipment/delivery), plus deferred-revenue software subscriptions and support contracts (deferred revenue $652M current + $232M long-term at Oct 31, 2025). Backlog ~$2,697M at FY2025 year-end (vs $2,375M prior year), the majority expected to convert within six months — short-cycle, not a multi-year take-or-pay book.
Sales model. >80% of business runs through a direct sales force of sales/solution engineers, complemented by 800+ channel-partner agreements. The direct model is itself a moat input — it puts application engineers inside customer R&D labs.
Two reportable segments: Communications Solutions Group (CSG) and Electronic Industrial Solutions Group (EISG) — see Lens 4.
The commercial-layer file kb/hardware/wiki/supply-chain.md is the pointer but holds generic hardware-chain content; the filing is the authoritative source here.
Upstream (inputs → Keysight). Keysight buys semiconductors, electromechanical components/assemblies, and raw materials (plastic resins, sheet metal) from a global supplier base. Critical nuance: the highest-value differentiation is vertically integrated. Keysight operates five specialized in-house technology centers — Santa Rosa CA, Colorado Springs CO, and Böblingen, Germany — that fabricate the parts not commercially available: microwave monolithic integrated circuits (MMICs), thick/thin-film circuits, optical components, high-speed probes, precision-machined mechanical parts, and advanced multi-chip / system-in-package modules. These proprietary components are the performance edge; they are not bought, they are made.
Manufacturing (the company). A hybrid in-house + contract-manufacturer model. The centralized manufacturing hub is Penang, Malaysia — the largest facility, doing final assembly, tuning, calibration and test of advanced instruments. Contract manufacturers handle assembly and PCB fabrication; in-house holds the complex, differentiated work. Build-to-order: products configured on receipt of firm customer orders.
Downstream (Keysight → end customer). Direct sales force (>80%) → into customer R&D labs and manufacturing lines; plus 800+ distributors/resellers/reps for lower-touch and reach extension.
Named stakeholders across the chain (who actually sits where):
Chokepoints / single-source. The 10-K flags that some custom-design parts are not readily available from alternate suppliers due to unique design or long design lead-times; Keysight mitigates via multi-sourcing and design-for-alternative-components. Penang is a single dominant assembly node — a geographic concentration risk (Malaysia) that doubles as a tariff/geopolitical exposure (see Lens 10/12). Non-cancellable supplier/CM purchase commitments ~$450M, majority <1 year.
The moat is real and multi-layered — this is the highest-quality lens in the file.
Switching costs via measurement correlation + installed base. In high-frequency/high-speed test, a customer's design data, automation scripts, and pass/fail criteria are calibrated to a specific vendor's instruments. Re-validating against a competitor's box is costly and risky. Keysight notes "continuous bandwidth upgrades keep legacy customers on multi-year refresh cycles" — the install base self-perpetuates ``. Engineers trained on Keysight stay on Keysight.
Technology leadership at the bleeding edge. Keysight is consistently first-to-market at each speed transition (it had 800G/1.6T and now 3.2T/system-emulation tooling out ahead of volume). In test, being first matters enormously: the tool must exist before the customer can develop the next-gen device, so the test vendor that ships first captures the design-in.
Vertical integration of proprietary components. The five in-house tech centers (MMICs, optics, high-speed probes — Lens 2) deliver "performance levels not commercially available". Rivals can't simply buy the same parts.
Scale + breadth. Keysight's own framing: "none of our competitors offer the equivalent range of product and services at the scale we do or serve all the same markets in the aggregate or by segment". ~16% global T&M share, #1 in the industry ``. Breadth lets it sell end-to-end solutions (design → simulate → validate → manufacture) that point-product rivals can't.
Bargaining power. Over customers: strong in leading-edge segments (few alternatives at the frontier), weaker in commoditized general-electronics where regional specialists compete. Over suppliers: strong for commodity inputs (multi-sourced), structurally low dependence because it makes its own critical parts. Gross margin ~64% consolidated and ~67% in CSG evidences pricing power.
Moat durability: high. The threat is not a single competitor copying it but (a) cyclicality compressing the whole sector and (b) a speed-transition where Keysight is not first. Today it is first. Brand: the "Keysight mark" is the only IP the company itself calls material — tells you the moat is institutional (people, correlation, breadth), not a patent wall.
Two reportable segments (segment profitability excludes SBC, acquisition amortization, acquisition/integration, restructuring, and interest — i.e., these are clean operating margins). segments.csv empty; all figures below sourced to the 10-K segment tables.
Communications Solutions Group (CSG) — ~69% of FY2025 revenue
| CSG | FY2025 | FY2024 | FY2023 |
|---|---|---|---|
| Revenue | $3,726M | $3,420M | $3,685M |
| YoY | +9% | (7)% | — |
| Gross margin | 66.9% | 67.6% | 67.7% |
| Operating income | $986M | $921M | $1,068M |
| Operating margin | 26.5% | 26.9% | 29.0% |
. Two end markets: Commercial communications (~67% of CSG, +10% in FY25) — driven by AI data-center interconnect (400G/800G Ethernet, terabit solutions, 800G transceiver manufacturing capacity), plus 5G/6G/O-RAN/NTN/quantum R&D. Aerospace, Defense & Government (~33% of CSG, +8% in FY25) — space/satellite, radar, spectrum operations, defense modernization. Spirent (closed Oct 2025) adds wireless network test/assurance + positioning into CSG. Trend: re-accelerating, AI is the engine.
Electronic Industrial Solutions Group (EISG) — ~31% of FY2025 revenue
| EISG | FY2025 | FY2024 | FY2023 |
|---|---|---|---|
| Revenue | $1,649M | $1,559M | $1,779M |
| YoY | +6% | (12)% | — |
| Gross margin | 59.6% | 59.9% | 61.9% |
| Operating income | $407M | $357M | $581M |
| Operating margin | 24.7% | 22.9% | 32.7% |
. End markets: semiconductor (up — AI-chip R&D, fab capacity; Synopsys OSG + PowerArtist additions), general electronics (up), automotive & energy (down in FY25 — the soft spot; EV/SDV demand digesting). EISG is structurally lower-margin than CSG (more general-electronics commoditization) but margin recovered +2 ppts in FY25 on operating leverage. Trend: mixed but improving; semis offsetting auto weakness.
Geography (consolidated revenue change, FY2025 over FY2024):
The cycle in one read: FY2023 was peak ($5,464M rev, 24.8% consolidated op margin), FY2024 the trough ($4,979M, depressed margins on a post-COVID test downcycle), FY2025 the recovery ($5,375M), and H1-FY2026 the breakout (see Lens 5). Both segments now growing; AI is pulling CSG and the semiconductor sub-market of EISG.
The latest filing on the shelf is the Q2-FY2026 10-Q. It is a blowout — and the single most important quarter in the file.
Headline (GAAP, research-layer):
| Metric | Q2 FY26 | Q2 FY25 | YoY |
|---|---|---|---|
| Revenue | $1,717M | $1,306M | +31% |
| Gross margin | 68.6% | 62.4% | +6 ppts |
| Operating income | $407M | $207M | +96% |
| Operating margin | 23.7% | 15.9% | +8 ppts |
| Net income | $349M | $257M | +36% |
| Diluted EPS (GAAP) | $2.02 | $1.49 | +36% |
Non-GAAP beat ``: non-GAAP EPS $2.87 vs $2.32 consensus (a ~24% beat), +69% YoY; revenue $1.72B vs ~$1.71B est. Record FCF $472M in the quarter.
Orders are the tell. Q2 orders $2,051M, +56% YoY; H1 orders $3,696M, +43%. Book-to-bill ≈ 1.19x ($2,051M orders / $1,717M revenue) — orders running well ahead of shipments, i.e., the recovery has momentum into H2.
What drove it. CSG + EISG both up; CSG 72% / EISG 28% of Q2 revenue. The engine is AI: AI-related revenue (predominantly wireline) was $500-600M in H1 FY2026 — "almost in line with what we did the whole of last year," with AI customers doubling ``. Acquisitions added ~7 ppts and FX ~1 ppt to the +31% — so organic growth ≈ +23%.
The asterisk (important — reported numbers are flattered). A one-time IEEPA tariff refund following US Supreme Court / Court of International Trade rulings: Keysight recorded a ~$100M receivable (refunds + statutory interest); excluding a $40M tariff revenue adjustment, Q2 revenue was $1.76B (+35%) but the refund also added ~$57M to operating income for the quarter ``. A chunk of the margin and EPS upside is non-recurring. Underlying organic strength is still excellent, but the optical "68.6% gross margin / 23.7% op margin" overstates the run-rate.
Guidance — raised ``: Q3 FY26 revenue $1.73-1.75B (+29% YoY midpoint), non-GAAP EPS $2.43-2.49 (+43% midpoint); full-year revenue growth raised to "high-20s percent." Management tone shifted clearly more confident vs the FY24-trough commentary.
Balance-sheet flags (FY2025 year-end, research-layer):
Market reaction: muted on the day (+0.1% aftermarket) `` — because the stock had already run ~100% into the print; the beat was largely priced in. That is itself a signal about expectations (Lens 12).
transcripts/ is empty on the shelf — this lens is ``, drawn from the public Q1/Q2 FY2026 call coverage.
Trajectory of tone (FY2024 trough → H1 FY2026):
Recurring phrases now (the things management keeps saying): "first-to-market," "early innings," "AI clusters / speed transitions to 1.6T/3.2T," "silicon photonics and co-packaged optics (CPO)," "system-level AI workload emulation," "defense modernization," "secular growth trends." The four AI demand pillars are the new narrative spine: (1) scaling of AI compute clusters, (2) speed transitions (800G→1.6T→3.2T R&D), (3) optical interconnects incl. CPO/OCS, (4) system-level emulation ``.
Things they stopped saying: the FY24 vocabulary of "softness," "cautious customer spending," "inventory digestion," "headcount reduction." The restructuring drumbeat has gone quiet. Net sentiment shift: clearly positive, and specifically AI-anchored. Caveat for calibration: management is talking its book at a euphoric moment; "early innings" is exactly what you'd say to justify a high multiple.
Peer set: the test-and-measurement + automated-test-equipment (ATE) complex.
| Company | Ticker | Mkt cap (USD) | Trailing P/E | Forward P/E | EV/EBITDA | Notes |
|---|---|---|---|---|---|---|
| Keysight | KEYS | ~$56.9B `` | ~54x `` | ~32x `` | n/a | #1 T&M, ~16% share |
| Advantest | 6857.T / ATEYY | ~$140B `` | n/a | n/a | n/a | AI/HBM ATE; ~14% share; biggest re-rate |
| Teradyne | TER | n/a | ~54x `` | ~48x `` | ~60x `` | ATE (AI/HBM, robotics); expensive |
| Rohde & Schwarz | private | n/a — private | n/a | n/a | n/a | #2-3 T&M, not listed |
| Tektronix (Fortive) | FTV | n/a (parent) | n/a | n/a | n/a | T&M inside Fortive |
| NI / Emerson | EMR | n/a (parent) | n/a | n/a | n/a | NI acquired by Emerson 2023 |
Read. The entire frontier-test complex has been AI-re-rated: Teradyne ~48x forward / ~60x EV/EBITDA, Advantest a ~$140B name. Against that, Keysight's ~32x forward looks relatively less stretched than the ATE pure-plays — but on its own history it is extreme: trailing ~54x vs a 5-year median ~30.8x (≈ +76%) . GF Value pegs "fair" at ~$198 vs ~$329 spot — a model-driven ~40-66% overvaluation flag (treat as one mechanical estimate, not gospel). The honest conclusion: KEYS is the highest-quality, most-diversified name in the group and is cheaper than the ATE darlings on forward P/E — but it is priced two standard deviations above its own norm, and the cushion depends on the AI-test cycle not rolling over.
Mostly ``; the structural moves:
The 2024→2026 AI re-rating (the dominant move). From a 52-week low of $152.85 to a high of $374.96, ~+100-113% over the trailing year ``. This is the market repricing KEYS from "cyclical GDP+ test vendor" to "AI-infrastructure picks-and-shovels." The single biggest driver of the share price is the AI-data-center capex narrative, transmitted through CSG's wireline/interconnect business.
Earnings inflections. The FY2024 downcycle (orders/revenue declines) compressed the multiple; the Q1→Q2 FY2026 beats-and-raises (orders +56%, FY guide to high-20s%) re-rated it up. Earnings + orders are the proximate catalysts.
The Spirent M&A saga (2024-2025). Viavi agreed to buy Spirent (~$1.3B, Mar 2024); the deal lapsed (May 23, 2024) when Keysight outbid at ~$1.5B equity / ~$2.4B EV; DOJ forced divestiture of ~40% of Spirent (high-speed Ethernet, network-security, RF channel emulation) to Viavi; closed Oct 15, 2025 ``. A strategically significant but financially digestible bolt-on.
The June 2026 ~8.7% single-day drop. Shares fell ~8.7% to ~$328.66 around June 26, 2026 `` — a valuation/profit-taking wobble near the all-time high, not a fundamental break.
Tariff headlines (2025-2026). The IEEPA tariff regime and subsequent Supreme Court / CIT rulings (the $100M refund) are now a swing factor — both a cost risk and a one-time gain ``.
What the pattern reveals: the market reacts to (a) AI-capex sentiment above all, then (b) orders/guidance, then (c) macro/tariff/valuation. It is not a single-customer or dividend story (KEYS pays no dividend). The stock has become a high-beta AI-infrastructure proxy with a quality-compounder chassis underneath.
CEO — Satish Dhanasekaran (53). CEO since May 2022; President & COO Oct 2020-May 2022; before that ran CSG (the crown-jewel segment) 2017-2020, and rose through wireless/mobile-broadband GM roles from the 2014 Agilent spin onward. Archetype: internally-promoted operator-engineer, not a parachuted-in financial CEO. He owns the AI-test narrative and has, so far, delivered on it (orders +56%, FY guide raised on his watch). Track record: led CSG through the 5G build, then steered the whole company out of the FY24 trough into the FY26 breakout.
CFO — Neil Dougherty (56). CFO since the 2014 spin (SVP & CFO Dec 2013-May 2022, then EVP & CFO) — a decade-plus of continuity, ex-Agilent Treasurer/Corporate Development. Deep institutional knowledge; the capital-allocation architecture (buybacks, IG balance sheet, disciplined M&A) is his.
Bench: Ingrid Estrada (Chief Supply Chain & Operations Officer, ex-Agilent Global Sourcing GM) and a broadly Agilent/Keysight-lineage team. Low churn at the top.
Capital-allocation history — disciplined and shareholder-friendly:
Red flags — minimal. No related-party deals, no promotional behavior, no unusual comp structure flagged in the filing; LTP awards tied to TSR + operating margin (sensible alignment). The only "watch" item is goodwill + intangibles = $4,728M, ~42% of the $11.3B balance sheet post-acquisition-spree (Lens 10) — an acquisitive company carries integration and impairment risk by construction. Verdict: a high-quality, aligned, long-tenured operator-led team with a clean capital-allocation record.
Acting as a forensic analyst. This is a clean set of books — the forensic interest is in acquisition accounting and the tariff one-timer, not in any sign of manipulation.
Income statement. Revenue recognition is standard ASC 606 (point-in-time on hardware shipment/delivery; over-time for support/subscription). Non-GAAP vs GAAP gap is wide but explained: Q2 FY26 non-GAAP EPS $2.87 vs GAAP $2.02 — a ~$0.85 bridge driven by acquisition amortization, SBC, and acquisition/integration costs ``. SBC $162M in FY25 (up from $137M) — material but not egregious (~3% of revenue). Watch: with three FY25 acquisitions, acquisition-amortization add-backs will inflate non-GAAP for several years; judge the company on GAAP + organic where possible.
The tariff one-timer (the single most important forensic flag). The ~$100M IEEPA refund receivable and ~$57M operating-income benefit in Q2 FY26 are non-recurring ``. Reported Q2 gross/operating margins (68.6% / 23.7%) and EPS are flattered by it. The clean run-rate is lower. Do not extrapolate the reported Q2 margin.
Balance sheet. Goodwill $3,424M (from $2,388M) + other intangibles $1,304M (from $607M) = $4,728M, ~42% of total assets $11,301M — the post-Spirent/OSG step-up. Annual goodwill impairment test (qualitative) passed; no impairments in FY2023-2025. PwC flagged the Spirent & OSG developed-technology / customer-relationship valuations as Critical Audit Matters — i.e., the most judgmental estimates in the audit (royalty rates, obsolescence, attrition). This is normal for an acquisitive company but is exactly where a future write-down would originate if the acquired tech underperforms. Receivables ($939M) and inventory ($1,050M) both grew slower than revenue — no channel-stuffing or inventory-build signal.
Cash flow vs earnings. Cash flow is high quality — FY25 operating cash flow $1,409M exceeded net income $850M; FCF ~$1,282M (capex only $127M, asset-light). Deferred-tax swings (a $384M deferred-tax benefit reversal in FY25 vs a one-time Singapore charge in FY24) make GAAP net income noisy year-to-year, but cash conversion is excellent and not diverging from earnings adversely.
Tax. Effective rate 20% (FY25), 29% (FY24, one-time Singapore charge), 22% (FY23). Benefits from Singapore (incentive expires Jul 2029) and Malaysia (expired Oct 2025, renewal in process) incentives. Watch: Malaysia incentive renewal + Pillar Two minimum tax ($13M expense in FY25) are structural rate risks.
Regulatory findings (required sub-section).
Built bottom-up from FY2025 actuals + H1-FY2026 + the raised guide. Non-GAAP EPS basis (the consensus convention), every input labeled. Diluted share count ~173M, drifting down ~1%/yr on buybacks.
Anchor: FY2025 non-GAAP EPS ≈ $6.60. H1-FY26 non-GAAP EPS run-rate ≈ $5.7-6.0 (Q1 + Q2 $2.87) ``.
FY2026 (base):
FY2027 (base): AI runway intact but laps a huge FY26; revenue +~10-12%, margin ~26%. EPS ≈ $13.3-13.8.
FY2028 (base): normalize to mid-cycle T&M growth ~6-8% + AI premium; revenue +~8%. EPS ≈ $14.5-15.3.
Scenario band (FY2026 non-GAAP EPS):
Cross-check vs price: at ~$329 and base FY26 EPS ~$11.8, forward P/E ≈ 27.9x — close to the ~32x the web cites on a slightly more conservative EPS. Either way, mid-to-high-20s/low-30s forward P/E for ~10-13% out-year EPS growth = priced for the AI cycle to keep compounding. Consensus 1-yr PT ~$383 (Buy), range $250-$426 ``.
Per --watchlist rules, no forecast.ts create is logged in this unattended sweep — the base case is recorded here for a later human-gated /thesis pass.
Bull case. Keysight is the #1, most-diversified "picks-and-shovels" vendor to the AI-infrastructure build — it sells the test gear that every 800G/1.6T/3.2T transceiver, switch, optical interconnect and AI cluster must pass through before volume. AI revenue doubled to $500-600M in H1 FY26 (≈ all of FY25) and management calls it "early innings" with a multi-year runway across four pillars (clusters, speed transitions, CPO/optical, system emulation). The moat (switching costs, first-to-market, vertical integration of proprietary components, scale/breadth) is durable and high-margin (segment op margins 24-27%, FCF margin ~24%, asset-light at ~$127M capex). Capital allocation is disciplined — no dividend, $1.5B fresh buyback, accretive software/test bolt-ons (ESI, OSG, Spirent) widening the moat. Orders +56% and book-to-bill 1.19x say the recovery has runway. Plus a defense-modernization second engine and semiconductor R&D demand that are secular and counter-cyclical to commercial comms.
Bear case (the three that could permanently impair or de-rate):
Pre-mortem (it's Dec 2027, the thesis broke — what happened?): Hyperscaler AI-capex growth decelerated through 2027 as the first wave of clusters was built out and utilization (not new build) became the focus; transceiver/optical customers that double-ordered 1.6T parts in 2025-26 worked down inventory, and Keysight's orders fell from +56% to negative YoY by mid-2027. The tariff refund didn't repeat. The stock, having been a 32x-forward AI proxy, de-rated to ~22x on flat EPS — a ~45% drawdown — even though the business is structurally fine.
Is the multiple too high? Relative to ATE peers (Teradyne ~48x fwd), no. Relative to its own history and its ~10-13% steady-state EPS growth, yes — it embeds a continuation of the AI super-cycle.
Contrarian view (what the market is refusing to see): The bears under-weight that Keysight's AI exposure is "second-derivative and diversified" — it doesn't need any one chip or hyperscaler to win; it needs the act of designing and validating next-gen interconnect to continue, which spans every competitor and every speed transition for a decade. And the defense + semiconductor R&D legs are secular and partly counter-cyclical, cushioning a commercial-comms air-pocket. The quality is under-appreciated; the entry price is the problem, not the franchise.
Dismantling the bull case. The short thesis on KEYS is not a fraud or accounting short — it is a "great company, terrible price, cyclical-peak-earnings" short.
A licensing-fortified cash machine being paid ~20x forward NOT to lose a $7B Apple leg it is already losing — the rerate only comes if Snapdragon-X PCs + custom data-center silicon replace Apple faster than handsets fade, and the tape (rev −3% YoY, Q3 guide −7%) says it isn't there yet.
A genuinely elite fabless analog compounder (43% 5yr ROIC) that has become a high-beta levered call on Nvidia's power-delivery socket — priced at ~112x trailing / ~49x NTM EV-EBITDA (peer median ~21x) while carrying an unremediated material weakness, an adverse ICFR opinion, and a live securities class action. Own the franchise, not this multiple; the gap between the Vera Rubin "70% share" dream and the Blackwell "allocation-at-risk" reality is the whole trade.