FormFactor

Phase A — Understand the business

Lens 1 · Company Overview

FormFactor sells the pick-and-shovel of chip test: it does not make chips, it makes the precision interface that touches a wafer and tells you whether the silicon works. Three things to internalize:

1. It is a probe-card company first, everything else second. Two reportable segments — Probe Cards ($637.9M, 81% of FY2025 revenue) and Systems ($147.1M, 19%). Within Probe Cards, the markets are Foundry & Logic ($369.9M), DRAM ($247.4M), and Flash ($20.6M). The 10-K is blunt: "We derive the majority of our revenues from the sale of our probe card products… We anticipate that sales of probe cards will represent a substantial majority of our revenues for the foreseeable future". Systems (probe stations, thermal chucks, cryogenic systems) is the "Lab" half of the "Lab-to-Fab" story — strategically interesting (quantum, photonics) but a minority of the P&L.
2. The product is a custom consumable, not a tool. A probe card is engineered to one customer's specific wafer/chip design and wears out over "hundreds of thousands, even millions, of compression cycles". Present tech puts >150,000 contact elements on a card at 40-micron spacing, testing at >80 GHz. That custom-consumable structure is the whole investment case: more complex chips → more test content → more probe cards per wafer, and they get replaced.
3. The end markets are exactly the AI-infrastructure stack. Customers test GPUs/CPUs/custom ASICs (Foundry & Logic), HBM/DRAM, NAND, RF, image sensors, co-packaged optical ICs, and quantum processors. The growth engine right now is HBM (memory stacks for AI accelerators).

Contract structure: revenue recognized at point of control transfer (ship/deliver); no take-or-pay, generally purchase-order driven with installation/service contracts (1–3 yr) as separate performance obligations. Because cards are custom, cancellations create non-recoverable cost / inventory impairment — a real downside-asymmetry the company flags itself. Founded 1993, HQ Livermore CA, ~2,153 employees, Mike Slessor CEO. No dividend.

Lens 2 · Supply Chain

Map: upstream component/material suppliers → FormFactor manufacturing → chipmaker/foundry → AI accelerator → hyperscaler.

• Upstream into FORM (named chokepoints): the 10-K states it depends on suppliers for "ceramic and organic substrates and complex printed circuit boards" and certain "contact elements and interconnects," some single-vendor / sole-source, generally on purchase orders, not long-term contracts — explicitly flagged as a price-increase and shortage risk. In Feb 2025 FORM bought a 20% equity stake in FICT Limited (Nagano, Japan) — a maker of complex multi-layer organic substrates and PCBs. Read that as vertical-integration insurance on its single most strategic input (the organic substrate that carries the probe). It cost $67.2M of cash in FY2025 and is currently dilutive — $2.0M equity-method loss in FY2025.
• FORM manufacturing footprint: probe cards in Livermore CA + Beaverton OR (primary), smaller Yokohama JP; Carlsbad + Baldwin Park CA being consolidated/closed by end-FY2026; a new Farmers Branch, Texas site (50,000 sq ft cleanroom, four structures) ramping late-FY2026. Systems/thermal/cryo in Boulder CO + Thiendorf/Munich/Karlsruhe Germany.
• Downstream named customers (this is where it's concentrated): SK hynix (>10% of revenue in every one of the last 8 quarters, 19.2%–25.0%), Intel (12–17% in several quarters), TSMC (10.4% in Q2-25), Samsung (12.4% in Q1-24). NVIDIA disclosed as a 10%+ customer for the first time in Q1-26 at 10.2%, with SK hynix at 29.5%. So the chain terminates squarely in the HBM→Nvidia-accelerator→hyperscaler loop.
• Chokepoint verdict: FORM is itself a chokepoint (one of three firms that can do leading-edge MEMS probe cards), but it sits on a thin sole-sourced substrate supply and a customer base of three or four giants. Concentration risk runs in both directions.

Lens 3 · Competitive Advantages (moats)

Real but contested — a "narrow moat that is being actively attacked."

• Source of the moat: (1) Customer-specific qualification. Each card is co-designed to a customer's wafer; the 10-K names "high capital investment… and the time and high cost of the customer evaluation process" as the barrier to entry in MEMS probe cards. Once you're qualified into a device family, you're sticky for that family's life. (2) High-frequency / high-parallelism leadership — >80 GHz, >150k contacts — which the company argues is what lets it win HBM (testing stacks of 8/12/16 DRAM die). (3) Lab-to-Fab breadth — it is the only player spanning analytical probes, probe stations, cryogenic systems AND production probe cards, giving it the photonics/quantum optionality rivals lack.
• Bargaining power: weak-to-balanced vs customers (a handful of giants, custom product, cancellable orders → customers hold the whip), and mixed vs suppliers (sole-source substrates → suppliers hold pricing power, which is exactly why FORM bought into FICT).
• The moat-erosion fact you cannot ignore: Technoprobe's fully-integrated MEMS-tip production won ~30% of TSMC's 2nm qualifications, "eroding FormFactor's historical lead" in foundry probe cards. FORM's own FY2025 Foundry & Logic revenue fell 3.0% while the AI cycle roared — a yellow flag that the moat is strongest in memory/HBM and weakest where Technoprobe is pressing.
• Verdict: durable in HBM and high-parallelism memory (Smart Matrix is a genuine edge); contested and slipping in leading-edge foundry. Not a wide moat — a capability lead that must be continuously re-earned each node.

Lens 4 · Segments

All `` unless noted.

Segment gross margin: Probe Cards 37.2% (FY23) → 41.4% (FY24) → 40.5% (FY25); Systems 51.3% → 43.2% → 41.8% (Systems margin has halved its premium as mix shifted to lower-margin products). Within Probe Cards, DRAM is structurally lower margin than Foundry & Logic, so the HBM-driven mix shift (DRAM 36.3%→38.8% of probe-card sales) is a margin headwind even as it's a growth tailwind — a key tension.

Geography: South Korea 30.3% (HBM/SK hynix), Taiwan 25.8% (foundry/TSMC), US 19.4%, China 7.4% (down from 13.5% in FY24 — export controls), Japan 5.5%. The China decline is policy-driven, not demand-driven, and it's nearly done washing through.

The acceleration story is entirely in DRAM/HBM. Q1-FY2026 makes it vivid: total revenue +32% YoY to $226.1M, with DRAM probe-card revenue +70% YoY to $82.9M and Foundry & Logic +30% to $111.2M. Systems actually fell 20% YoY ($34.8M→$27.9M) as legacy probe stations gave way ahead of the Triton CPO ramp.

---

Phase B — Measure performance

Lens 5 · Earnings Result (latest print: Q1-FY2026, period ended 2026-03-28)

A genuine beat-and-raise, partially masked by a restructuring charge in GAAP.

• Revenue $226.1M, +32.0% YoY (vs $171.4M); above the company's $220–230M guide and a Q1 record.
• Gross profit $86.8M, GAAP GM 38.4% — but COGS absorbed $21.5M of 2026-restructuring charges; underlying GM ≈ 47.9%. Probe Cards segment GM hit 50.5% (vs 37.8% a year ago) — the operationally important number.
• Operating income $16.6M (GAAP) vs $3.3M; net income $20.4M (+218%); GAAP diluted EPS $0.26 vs $0.08. Non-GAAP diluted EPS was $0.56 — the gap is restructuring + SBC + amortization. Always read FORM on non-GAAP for run-rate, but respect that GAAP operating margin is structurally only ~7%.
• Guidance (Q2-FY2026): revenue $240M ±5M, non-GAAP GM 49.5% ±150bp, non-GAAP EPS $0.61 ±0.04; HBM guided to "another record" as a second customer adopts Smart Matrix for at-speed stack test. Tone: clearly upbeat / accelerating.
• Balance-sheet flags: cash $123.5M + marketable securities $179.7M = ~$303M; AR $132.2M (up with revenue — not outrunning it); inventory $112.9M (well-controlled, +2% q/q on +revenue); goodwill $215.4M; minimal term debt. CFO Q1 $45.0M vs $23.5M — operating cash nearly doubled. Healthy.
• Market reaction: despite the beat-and-raise, the stock fell ~11.6% in late April — pre-earnings profit-taking after a ~354% 12-month run and valuation vertigo. The print was good; the stock was priced for perfect. That is the single most important sentence in this dossier.

Lens 6 · Earnings Calls (sentiment trend)

No transcripts on the local shelf (transcripts/ empty) — this lens is ``, drawn from Q4-25 and Q1-26 call coverage.

• Tone trajectory (last ~3 calls): Q3-25 → Q4-25 → Q1-26 shows steadily rising confidence, with the vocabulary migrating from "cyclical recovery / diversification" to "HBM4 inflection," "Smart Matrix adoption," and "record". Q4-25 stock +14.4% on a raised 2026 outlook.
• What management keeps repeating: HBM test-intensity, Smart Matrix full-wafer contactor for HBM4, the Triton CPO platform (co-developed with Advantest + Tokyo Electron), the "$1.6B revenue / doubling by 2030" long-range target. Capital-allocation language centers on the Texas factory build and gross-margin discipline.
• What they've stopped emphasizing: the China/export-control drag (now small at ~5–7% of revenue) and Systems-segment growth (de-emphasized while it transitions to CPO). The narrative has become purely an AI-memory-test story — which is both the bull case and the concentration risk.

Lens 7 · Comps

Read: the only directly-comparable public pure-play is Technoprobe, and it trades even richer than FORM (~$26B cap, ~49x fwd P/E) on €628.4M FY2025 revenue (+16%, net income €98.8M +55%). So the market is paying a structural premium for the probe-card duopoly's AI leverage — FORM is expensive, but not an outlier within its own niche. It IS an outlier vs the broader semi-cap group. The honest framing: FORM at ~57x forward earnings and ~11.5x EV/sales is priced as an AI-secular-grower, not as the ~18–20% cyclical test-equipment supplier its 10-year financials describe. ATE peers (Teradyne, Advantest) and the Japanese probe names would anchor the table properly but I could not source clean current multiples for them — flagged rather than fabricated.

Lens 8 · Stock-Price Catalysts (what actually moves FORM)

Mostly ; the 5-year total-return path is .

• 5-yr cumulative total return (from the 10-K's own performance graph, $100 invested 2020-12-26): $104.5 (2021) → $52.2 (2022) → $98.0 (2023) → $107.8 (2024) → $136.4 (2025). That 2022 halving is the tell.
• The pattern: FORM trades as a high-beta memory-cycle + AI-sentiment proxy. The >50% drawdown in 2022 mapped the DRAM/NAND oversupply and ~20% YoY semi-revenue decline; revenue itself fell 11.3% in 2023. The ~354% rally into 2026 mapped the HBM/AI capex super-cycle.
• What the market reacts to, ranked: (1) HBM/DRAM demand signals and HBM4 timing (the dominant driver now); (2) memory-customer capex (SK hynix/Samsung/Micron 2026 capacity is "sold out" ); (3) its own guidance vs a stretched bar — Q1-26 proves a beat-and-raise can still sell off ~12% when expectations are extreme; (4) export-control / China headlines (now muted). Earnings + HBM roadmap move this stock far more than valuation does — until a cycle turn, when valuation moves it violently.

---

Phase C — Judge people & books

Lens 9 · Management

• CEO Michael (Mike) Slessor — President Oct-2013, CEO since Dec-2014 (~11.5 yr tenure); joined via the 2012 MicroProbe acquisition (he was its CEO); prior KLA-Tencor. Track record: steered FORM from a ~$300M sub-scale probe-card maker through the 2022 trough to a $785M record-revenue, two-segment, Lab-to-Fab platform; 3-yr TSR to mid-2024 at the 73rd percentile of the S&P Semis index. Capital allocation has been acquisition-and-capacity, not return-of-capital: bought MicroProbe, Cascade Microtech, FRT (later sold), Keystone Photonics (Dec-2025, $20.6M, silicon-photonics test), the FICT 20% stake, and the Texas factory. Buybacks are modest and roughly offset dilution (see Lens 10).
• Skin in the game / comp: total comp ~$5.3M (≈11% salary, ≈89% equity/bonus); direct ownership 0.69% ($29.6M at recent prices). Founder? No — a professional operator who has run this asset for over a decade. For a serial-acquirer test-equipment platform, that's the right archetype; the risk is the empire-building bias acquirers carry.
• Red flags: none acute. The FICT minority stake is a related-supplier investment to watch (currently loss-making, $2.0M drag ); the acquisition cadence means goodwill is now $215.4M (~17% of assets). No comp scandal, no strategy whiplash, no promotional behavior in the disclosures.

Lens 10 · Forensic Red Flags

Forensic read of the income statement, balance sheet, and cash flow — every figure ``.

• Earnings quality / non-GAAP gap: the headline risk is the GAAP-vs-non-GAAP wedge. FY2025 SBC was $38.6M — i.e., 71% of the $54.4M GAAP net income. Non-GAAP EPS ($0.56 in Q1-26) flatters GAAP ($0.26) by adding back restructuring + SBC + amortization. This is normal for semi-cap, but at a ~57x non-GAAP multiple the SBC dilution is doing real work and must be counted. Share count is, to management's credit, flat (~77.3M diluted FY23→FY25) — buybacks ($26.2M FY25) roughly neutralize RSU/ESPP issuance, so the dilution is contained even if the expense is large.
• Gain-on-sale noise: FY2023 and FY2024 net income was inflated by one-time gains — $73.0M (FRT sale, FY23) and $20.6M (China ops sale, FY24). So the optics of "net income fell from $82.4M→$69.6M→$54.4M" is misleading — operating income ex-gains actually troughed and is now recovering. Watch that analysts don't anchor on the gain-boosted prior years.
• Cash flow vs earnings: clean. FY2025 CFO $115.4M comfortably exceeded GAAP net income $54.4M; FCF was depressed to ~$11.7M only because of one-time growth capex (Texas land/build) + the $67.2M FICT stake + $20.6M Keystone — i.e., investment, not deterioration. FY2024 FCF was ~$79.1M. AR and inventory are growing with — not ahead of — revenue (AR +$20.2M, inventory +$20.7M on +$21.4M revenue in FY25; well-behaved in Q1-26). No receivables/inventory red flag.
• Inventory provisioning: FORM books a steady ~$13–15M/yr excess-and-obsolete provision — appropriate given custom-card obsolescence; not aggressive, not suspiciously low.
• Effective tax rate rose to 18.7% (FY25) from 12.3% on OBBBA changes — a modeling input, not a flag.

Regulatory findings (required sub-section):

• SEC Litigation Releases / AAERs: None. Verified via SEC EDGAR EFTS (LR + AAER), search window 2021-06-30 to 2026-06-30.
• Non-SEC enforcement (FTC/DOJ/FDA/export): web search returned no enforcement action, lawsuit, or settlement against FormFactor. FORM is affected by US→China export controls (it lost China revenue, 13.5%→7.4%) but is a compliant party, not a target. ~7% of FY2025 revenue was China-derived, down from 14% FY2024, under the expanded license regime.
• 10-K Item 3 (Legal Proceedings): incorporates Note 12; the company discloses only "legal proceedings and claims in the ordinary course of business" with no specific material matter quantified.
• Conclusion: No material regulatory or legal findings — verified via SEC EDGAR EFTS (LR, AAER), web search, and 10-K Item 3 / 10-Q Note 15 as of 2026-06-30.

---

Phase D — Project & stress-test

Lens 11 · Forward Projection (FY2026 / FY2027 / FY2028 non-GAAP EPS)

Built bottom-up from FY2025 actuals + Q1-26 actual + Q2-26 guide. Every input labeled; output ``. No forecast.ts create in --watchlist mode (per skill rules) — the base call is logged below for Connor to commit on a /thesis pass.

Anchors: FY2025 revenue $785.0M, non-GAAP EPS ~$1.90–1.93 (TTM proxy); Q1-26 non-GAAP EPS $0.56 actual, Q2-26 guide $0.61 mid; consensus FY26 non-GAAP EPS ~$2.48 (range $2.21–2.64), FY27 ~$2.34 — note the Street's FY27 < FY26 print looks like a stale/cyclical-reversion estimate; treat with caution.

Margin bridge note: the base case hinges on the non-GAAP GM holding at the guided ~49.5% (vs FY2025's underlying ~40% blended / 47.9% Q1 underlying) — that step-up is the 2026 restructuring payoff (Carlsbad/Baldwin Park consolidation) plus HBM mix and utilization. If the restructuring doesn't convert to durable margin, the base collapses toward the bear. Capex of $140–170M in FY26 means FCF stays suppressed through the Texas build — this is not a FCF story until FY2027.

Base call to log on a /thesis pass (NOT logged here): FORM FY2026 non-GAAP EPS >= $2.45, p≈0.60, resolves 2026-12-26, tags formfactor,deep-dive.

Lens 12 · Bull vs Bear

Bull case. FormFactor is the only Western pure-play on rising test-intensity in AI silicon. The physics are on its side: every generation of HBM (8→12→16-high) and every chiplet/CPO design adds contacts, frequency, and test time, and the cards are consumables. Smart Matrix is a genuine technical lead for at-speed HBM4 stack test, with a second customer now adopting it. The customer set — SK hynix, NVIDIA, TSMC — is the exact list you'd want exposure to. Memory makers have sold out 2026 capacity and are booking 2027. Management's "$1.6B by 2030" doubling is aggressive but directionally credible given the HBM ramp. The 2026 restructuring is a real, self-help margin lever (49.5% guided non-GAAP GM). Balance sheet is fortress-clean (~$300M cash, negligible debt).

Bear case (2–3 things that could permanently impair, plus the priced-in problem).

1. Customer concentration is worsening, not improving. SK hynix 29.5% + NVIDIA 10.2% = ~40% of revenue in two names in Q1-26. A single HBM roadmap slip at SK hynix, or NVIDIA dual-sourcing its test, removes a quarter of the business overnight. The custom-card model means lost designs are non-recoverable.
2. Technoprobe is out-executing FORM in leading-edge foundry (~30% of TSMC 2nm quals ) and FORM's own Foundry & Logic revenue fell in FY2025. The "moat" is half a moat.
3. It's a cyclical wearing a secular multiple. The 2022 share-price halving and 2023 revenue decline are recent history. At ~57x forward non-GAAP EPS and ~11.5x EV/sales, the stock has priced out the cycle. GAAP operating margin is structurally ~7%.

Pre-mortem (it's Dec-2027, the thesis broke — what happened?). HBM4 either pushed right or commoditized faster than test-content grew; SK hynix's 2027 capacity got digested and orders air-pocketed; Technoprobe extended from foundry into memory; the Texas factory ramped into a softening cycle, leaving fixed cost stranded; and the ~57x multiple compressed to ~25x even as EPS held — a −50% price move on no fundamental catastrophe, purely de-rating. That is the most likely way to lose money here.

Are multiples too high? Yes on an absolute and semi-cap-relative basis; no relative to the only pure-play peer (Technoprobe, ~49x fwd). The risk isn't that FORM is a bad business — it's that the entry price already assumes the bull case.

Contrarian view (what the market is refusing to see). Bulls treat HBM-driven DRAM growth as pure upside, but DRAM probe cards are structurally lower-margin than the foundry cards FORM is losing to Technoprobe. So the AI memory boom may grow revenue while diluting the margin mix — the opposite of the operating-leverage story embedded in the multiple. The bull narrative and the segment economics quietly disagree.

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

Dismantling the long.

• The revenue is renting three landlords. ~40% in SK hynix + NVIDIA; >10% concentration in SK hynix for 8 straight quarters. This isn't a diversified consumables annuity — it's a levered bet on two customers' HBM roadmaps. Custom cards mean a lost socket is gone for good.
• The moat is being breached where it matters most. Technoprobe's vertical MEMS-tip integration is winning the highest-value, leading-edge foundry quals (TSMC 2nm). FORM's foundry revenue is already declining. If memory is the only place the moat holds, FORM is a memory-cycle stock, not a secular compounder.
• Most dangerous underestimated competitor: Technoprobe — same MEMS tech, faster execution, richer balance-sheet runway, and now a foothold in FORM's crown-jewel accounts.
• Capital-allocation skeptic's note: the FICT 20% stake is a related-supplier investment that's currently losing money; goodwill is ~17% of assets after a decade of deals; the Texas factory is a $140–170M FY26 capex bet timed into a cycle that has burned this company before.
• What must hold for today's price: non-GAAP GM steps to ~50%+ and holds; HBM4 ramps on schedule with FORM (not Technoprobe) winning the memory share; foundry stops bleeding; and the ~57x multiple doesn't compress. That's four things, all of which have to go right.
• Down-20–30%-growth scenario: if FY27 revenue comes in ~20% below the bull (i.e., flat-to-down as a cycle rolls), EPS likely halves toward the bear ~$1.40–1.55 and the multiple de-rates simultaneously — a classic cyclical double-whammy, plausibly −50–60% peak-to-trough, exactly as 2022 demonstrated.
• Single permanent-impairment scenario: NVIDIA or SK hynix internalizes / dual-sources HBM test at scale, or a competing at-speed contactor leapfrogs Smart Matrix. Plausibility: low-to-moderate near-term, but non-trivial over 3 years given how fast Technoprobe moved in foundry.

Lens 14 · Management Questions (15, ordered by information value)

1. SK hynix + NVIDIA are now ~40% of revenue — what is the concrete path to getting your top-two concentration below 30%, and over what timeframe?
2. Technoprobe reportedly won ~30% of TSMC's 2nm probe-card qualifications. What is your current win-rate trend at leading-edge foundry nodes (N3/N2/A16), and where specifically are you losing?
3. Smart Matrix HBM4 — how many memory customers are designed in today, what's the qualification pipeline, and what is your assumed HBM4 probe-card content uplift vs HBM3E?
4. Walk us through the non-GAAP gross-margin bridge from ~48% to the ~49.5% guide and to a sustained 50%+ — how much is restructuring (one-time) vs mix vs utilization?
5. DRAM cards are lower-margin than foundry cards. As HBM scales, does revenue mix structurally cap your blended margin, and how do you offset that?
6. The Texas (Farmers Branch) factory: at what revenue level does it reach breakeven utilization, and what's the downside if the memory cycle softens during the ramp?
7. What is your through-cycle FCF-conversion target once the Texas capex normalizes, and when does FCF re-inflect?
8. The FICT 20% stake is currently loss-making — what's the strategic thesis, the path to profitability, and would you increase the stake or fully acquire?
9. How replaceable are your sole-source substrate/interconnect suppliers, and what would a disruption cost in revenue and margin?
10. The Triton CPO platform with Advantest/TEL is guided to only $10–20M in 2026 — what's the realistic 2028–2030 CPO TAM and your expected share?
11. What share of probe-card demand is genuinely "AI test-intensity" (rising content) vs simply higher memory volume that reverses with the cycle?
12. How exposed are you to a 2027 HBM digestion air-pocket if SK hynix/Samsung over-build 2026–27 capacity?
13. Capital allocation: with ~$300M cash and the multiple where it is, why buybacks/M&A over a larger return of capital — and what's your M&A bar now?
14. What is your assumed China revenue trajectory under the current export regime, and is there further downside if controls tighten?
15. At today's valuation, what specific milestones over the next 12–18 months do you think the market is over-crediting, and where are you most likely to disappoint?

---