FormFactor, Inc. (NASDAQ: FORM) — The Ramp Printed; the Multiple Still Demands More

Independent research report dated August 30, 2026. Market data through August 28, 2026; financial data through the quarter ended June 27, 2026 unless noted.

⚡ Claude’s Take

The author’s independent opinion; general information only and not investment advice. The analytical body below carries no position.

Verdict: HOLD / WAIT for a wider margin of safety; not a short. I would accumulate in a roughly $65–80 zone, not chase at $101.70. This is a materially less negative view than at the June 27 baseline. FormFactor did what an expensive cyclical is supposed to do: Q2 revenue reached $258.2M, GAAP operating margin reached 22.4%, Foundry & Logic and Systems both grew more than 20%, and Q3 guidance implies another record. Meanwhile, the stock fell 22% from the prior report’s $130.74 close. The evidence for a structural earnings step is stronger and the price is lower. But the remaining valuation still asks investors to capitalize a favorable mix and unusually high utilization as if they were durable: the stock trades at roughly 8.5x trailing sales, 50x trailing EBITDA, 58x trailing EBIT, and 68x trailing GAAP EPS. Q3 non-GAAP gross margin also includes about 300 basis points of one-time tariff refunds, while FormFactor, Technoprobe, and Micronics Japan are all adding capacity into the same demand signal.

The framing has changed from a late-June one-way momentum vehicle to a high-beta quality reset near its 200-day trend. Factor data still show large market, semiconductor, and quality exposure, negative low-volatility exposure, roughly 51% specific volatility, and a five-year maximum drawdown of 62.7%. A scenario set centered on a roughly $73 base, about $30 in a cyclical bear case, and about $145 in a successful 2030 bull case leaves an unattractive weighted payoff at $101.70, but the net-cash balance sheet, real HBM/advanced-logic qualifications, and record operating momentum make a short thesis unnecessarily hazardous. Conviction: medium. Flips bullish if FORM sustains roughly $260–280M quarterly revenue and normalized margins after tariff refunds while non-HBM demand broadens through 2027. Flips bearish if normalized gross margin falls below roughly 48%, revenue drops below $240M while AI capital spending remains strong, or Farmers Branch ramps into poor utilization.

Tag: “The proof improved faster than the margin of safety.”

📈 Stock Price Action — Five-Year Event Map

FORM’s five-year close-basis range runs from $18.19 on November 3, 2022 to $159.93 on June 30, 2026. The August 28 close of $101.70 is 36.4% below that high, near the $101.83 200-day exponential moving average, and within a trailing 52-week range of $27.81–$160.27. Price moves below are facts; causal attributions are interpretations supported by dated filings and market evidence.

The 2022 trough followed a classic memory-capital downturn: utilization and test demand weakened just as discount rates rose. The 2023–24 recovery brought advanced-logic optimism and a focused portfolio, but a second drawdown into 2025 demonstrated that qualifications do not eliminate cyclicality. The HBM-led inflection then produced repeated earnings gaps and an extraordinary 2026 re-rating. Late June added a technical index-and-momentum extension; July removed much of that excess during a semiconductor sell-off. The July 30 jump was company-specific evidence—Q2 results and Q3 guidance—not merely beta. The stock’s subsequent inability to reclaim the peak despite stronger fundamentals shows that the debate has shifted from whether the ramp exists to how long it lasts and what multiple it deserves. Sources: AZI adjusted price history through August 28, 2026; FormFactor’s July 29 earnings release; and contemporaneous sector-market reports dated July 2 and August 28, 2026.

Changes since the June 27, 2026 baseline

The prior June 27 baseline described a real but narrow probe-card advantage and required three things to disprove its skeptical stance: quarterly revenue above $240M, operating margin of at least 16%, and diversified demand that persisted through a memory digestion. Q2 cleared the first two numerical gates decisively and offered meaningful diversification: Foundry & Logic rose 22.4% year over year and Systems rose 44.1%; HBM supplied only about one-third of the company’s total year-over-year revenue increase. The duration requirement remains open. Q3 shifts DRAM mix from HBM toward DDR, 2027 introduces factory-ramp expense, and direct rivals are adding capacity. The evidence score is therefore passed for one quarter; persistence not yet established.

The valuation and tape changed almost as much as operations. FORM fell from $130.74 on June 26 to $101.70 on August 28, while trailing revenue and EBIT stepped higher. Its AZI composite own-history valuation percentile declined from roughly the 98th percentile to 87.7; P/B and P/S remain near their 95th percentiles, whereas the P/E percentile fell to 73.4 because earnings caught up. The new conclusion is not the old view repeated at a lower price. It explicitly credits better earnings power, raises the normalization range, and treats the stock as a reset quality cyclical rather than an index-driven momentum extreme.

1. Executive Summary

FormFactor designs and manufactures probe cards—the custom, precision interface that temporarily connects an automated semiconductor tester to a wafer—and a smaller portfolio of engineering systems used in electrical, optical, thermal, and cryogenic test. The probe card looks like an accessory beside an expensive tester, but it performs a critical economic function: finding defective die before packaging. That function becomes more valuable when a package combines several costly die, as in high-bandwidth memory, chiplets, and co-packaged optics. A bad die discovered after stacking can destroy the value of the entire package. FormFactor’s product therefore benefits from rising test intensity even if semiconductor unit growth is modest.

The current operating inflection is genuine. Q2 2026 revenue was $258.2M, up 31.9% year over year and 14.2% sequentially. Probe Cards produced $209.7M and Systems $48.5M. Foundry & Logic revenue grew to $121.8M, DRAM to $85.0M, and Systems to $48.5M; Flash declined to $2.9M. GAAP gross margin was 50.7%, GAAP operating income was $57.8M, and operating margin was 22.4%. Operating cash flow reached $61.8M and free cash flow $52.6M. Management guided Q3 revenue to $270M plus or minus $10M, GAAP gross margin to 52.0% plus or minus 150 basis points, and GAAP EPS to $0.75 plus or minus $0.09. These are not presentation-adjusted anecdotes; the revenue, GAAP profit, and cash are visible in the filed statements.

The earnings quality is strong but not pure. Management attributed roughly one-third of Q2’s sequential non-GAAP gross-margin improvement to durable cost actions, one-third to volume, and one-third to nonrecurring tariff refunds and precious-metal recovery. It placed a normalized Q2 baseline near 51% non-GAAP gross margin at Q2 volumes, below the reported 53.3%. Q3’s 54% non-GAAP gross-margin guide includes $7–9M of IEEPA tariff refunds—about 300 basis points. Moreover, HBM was approximately two-thirds of Q2 DRAM revenue, but management expects Q3 DRAM dollars to remain roughly flat while mix rotates materially toward DDR because constrained DDR5 wafer starts are more profitable for customers. That disclosure is healthy realism and a warning against straight-line HBM extrapolation.

FormFactor’s advantage is real and narrow. SmartMatrix and related architectures combine large-area MEMS manufacturing, high contact density, signal integrity, thermal control, cleaning behavior, and design automation. Customers qualify a card for a specific device generation because a failed interface can damage yield, delay a production ramp, or consume tester time. These costs create product-generation captivity and favor incumbents with application engineers, manufacturing learning, and service infrastructure. They do not create a network effect or permanent account lock-in. Cards are design-specific, customers can dual-source, and the next chip generation provides a natural re-sourcing point. Direct competitors Technoprobe and Micronics Japan are formidable and currently growing at least as quickly as the market.

Industry structure is attractive at the leading edge but mixed overall. HBM4, more stack layers, tighter pitches, chiplets, advanced packaging, and optical I/O all raise the economic value of known-good-die testing. Yet the customer base is concentrated and sophisticated: SK hynix and TSMC alone generated 35.3% of Q2 revenue. Supply is also reacting. FormFactor’s Farmers Branch factory starts production in late 2026; Technoprobe is executing a roughly €350M capacity program through early 2027; Micronics Japan is accelerating its Aomori expansion. The capital-cycle question is not whether AI demand is real. It is whether new supply and customers’ dual-sourcing leverage allow suppliers to retain today’s utilization, lead times, and margins.

Financially, the company enters this investment cycle from strength. June cash of $109.8M and marketable securities of $235.9M exceed roughly $11.6M of term debt; a $65.9M equity investment provides additional optional value but is excluded from conservative net cash. The balance sheet can fund a 2026 capex program of $140–170M. The cost is near-term free-cash-flow volatility and a larger fixed-cost base. The initial Farmers Branch capacity is roughly comparable to the California probe-card footprint, begins ramping in Q4 2026, and should not be meaningfully accretive until 2028. Underutilization or qualification delays could turn strategic capacity into a margin burden before it becomes an advantage.

At $101.70, approximately 78.1M shares imply a $7.94B equity value and roughly $7.61B enterprise value using conservative net cash. Against trailing revenue of $902.2M, EBITDA of $153.3M, EBIT of $130.5M, and GAAP EPS of $1.49, the stock trades near 8.4x sales, 49.7x EBITDA, 58.3x EBIT, and 68.4x earnings. The multiple has compressed sharply, but the stock still discounts a large part of management’s 2030 ambition before the factory ramp and competitive capacity wave have been tested. The body of this report therefore treats the central uncertainty as duration and normalization, not the existence of the current upturn.

2. Business Overview

The product and its place in semiconductor economics

A wafer-probe system has three conceptually separate layers. The automated test equipment provides electrical stimulus and measurement. The prober positions and controls the wafer. The probe card is the customized interface that translates the tester’s resources into thousands—or, in advanced full-wafer memory applications, more than 150,000—temporary low-impedance contacts. Each application imposes different constraints: pad layout, pitch, current, frequency, signal loss, thermal expansion, contact force, particulate contamination, and cleaning frequency. The card must maintain those properties over hundreds of thousands or millions of contacts while limiting damage to the wafer.

That combination makes a probe card both consumable and project-like. Contact elements wear and require cleaning, repair, or replacement, but the dominant order flow follows new chip designs, qualifications, wafer starts, and production ramps. The company does not have subscription revenue. A design win can persist through a device’s production life and generate replacement/service demand, yet it must often be re-earned at the next node or package. Revenue consequently has higher visibility than an unqualified piece of capital equipment once a program is ramping, but lower recurrence than a software or aftermarket model.

The economic value rises with package complexity. Traditional testing could tolerate some defect discovery after wafer sort because the package contained one relatively inexpensive die. In HBM, several DRAM die sit on a base die and connect to an accelerator through advanced packaging. A defective component discovered after stacking wastes not just one die but scarce packaging capacity and the other known-good components. High-speed stack testing and wafer-level known-good-die screening can therefore save multiples of the probe card’s cost. Similar logic applies to chiplets and co-packaged optics: more heterogeneous, costly components increase the penalty for late failure.

Segments and revenue mix

The company reports Probe Cards and Systems. Probe Cards generated $209.7M, or 81.2% of Q2 revenue; Systems generated $48.5M, or 18.8%. Within total company revenue, Foundry & Logic was 47.2%, DRAM 32.9%, Flash 1.1%, and Systems 18.8%. Probe Cards earned 54.4% gross margin in Q2, while Systems earned 48.5%. Both segment margins improved sharply year over year, which matters because the current profit step is not confined to one reporting unit.

Foundry & Logic covers microprocessors, graphics processors, networking silicon, mobile application processors, foundry customers, and custom accelerators. The category grew 22.4% year over year in Q2. Management cited CPU, networking, and initial hyperscaler custom-ASIC activity. A new advanced-MEMS GPU card qualified and began production shipments in Q3, but management acknowledges that GPU is currently more heavily served by a competitor. This is an opportunity with evidence of entry, not evidence of category leadership.

DRAM includes conventional memory and HBM. Q2 DRAM revenue of $85.0M rose 49.0% year over year. Management estimated HBM at about two-thirds of the category and described strong high-speed stack-test share at two of three major HBM producers using SmartMatrix. The company supports HBM4 final-stack testing at more than 10 gigabits per second. The technological claim is credible and economically important, but precise customer wallet share is not public. Q3’s expected shift toward DDR also shows that the product mix responds to memory producers’ opportunity cost rather than following a single end-market script.

Flash is currently immaterial at $2.9M, down 48.4% year over year. Its decline illustrates a benefit of portfolio breadth: weakness in NAND no longer determines company direction. It also warns that FormFactor has not escaped memory cyclicality; one submarket can contract by half even during record consolidated results.

Systems combines engineering probe stations, analytical probes, thermal systems, cryogenic products, and newer optical-electrical production-test cells. The Triton platform integrates optical alignment, electrical test, automation, and partner equipment for silicon-photonics and co-packaged-optics manufacturing. Management now expects 2026 CPO revenue above $20M, compared with an original $10–20M range. The company has more than 160 prior silicon-photonics systems installed, which supplies application experience, but Triton is still an early high-volume-manufacturing ramp. Adoption timing, customer yields, test time, and partner dependence remain less proven than the installed lab franchise.

Customers, geography, and operating model

SK hynix represented 24.3% of Q2 revenue and TSMC 11.0%; Intel fell below the 10% disclosure threshold. The comparable disclosed concentration was 47.8% a year earlier, so threshold concentration improved to 35.3%. That is real diversification, but it can overstate the reduction in buyer power. HBM4 demand is driven by a few memory producers, leading-edge foundry activity by very few fabs, and the end customers behind those wafers by a handful of AI-platform companies. The customer can also qualify more than one card supplier. FormFactor’s dependence is therefore better measured by economic clusters than by accounting thresholds alone.

Manufacturing is vertically integrated where intellectual property and process control matter. MEMS contact structures, assembly, routing, and design automation require proprietary processes and tight yields. The company also uses external partners and announced a July 2026 relationship with Keystone Microtech for Taiwan assembly, test, repair, and distribution scalability while retaining proprietary technology. This hybrid model can place service closer to customers without handing away the core process. Its success depends on maintaining quality and protecting know-how across a more distributed network.

The new Farmers Branch, Texas facility is the largest operating-model change. Initial capacity is roughly equivalent to the California probe-card footprint. Production begins in late Q4 2026 and ramps through 2027–28. Management expects 2027 ramp inefficiencies and principal accretion in 2028. That timetable means reported depreciation, labor, and start-up expense arrive before fully loaded revenue. A favorable demand environment can absorb the capacity and lower unit costs; a digestion can expose excess fixed cost. The facility is thus both an enabler of growth and the cleanest future test of whether management correctly judged the cycle.

Verdict. FormFactor sells an economically indispensable, technically difficult interface with more product-generation recurrence than ordinary capital equipment but much less contractual recurrence than software. The mix is improving and Systems adds optionality, yet custom programs, concentrated customers, and fixed manufacturing keep the model cyclical. The business is understandable once probe-card value is separated from tester value: FORM monetizes the cost of finding defects early, and its earnings depend on how much of that value it retains after customers and competitors respond.

3. Industry Dynamics

Demand: test intensity, not merely wafer volume

The structural industry case rests on rising test content per wafer. HBM adds die, stack height, I/O density, bandwidth, and multiple test insertions. Chiplets and 2.5D/3D packaging combine separately manufactured components whose value can be destroyed by one latent defect. Leading-edge logic raises current, thermal, signal-integrity, and fine-pitch requirements. Co-packaged optics adds precision optical alignment and electrical-optical coordination. None of these trends requires semiconductor units to grow at AI-like rates for test spending to outgrow wafer volume.

Primary customer evidence supports the mechanism. TSMC is expanding CoWoS toward larger interposers and more HBM and expects a 5.5-reticle solution to enter volume production in 2026. Samsung announced HBM4 mass production and expects 2026 HBM sales to more than triple while adding capacity. Teradyne reported Q2 Semiconductor Test revenue up 128.1% year over year and memory-test revenue above $200M for a third quarter. Cohu grew Q2 revenue 38.4% as AI-compute and industrial demand improved. These companies occupy different layers of the test chain, so their results corroborate broad test-intensity growth rather than a company-specific promotion.

FormFactor’s May Investor Day, using Yole and company estimates, placed its served market at $3.1B in 2025—$2.8B probe cards and $0.3B systems—and $4.5B in 2030—$4.0B and $0.5B respectively—an 8% compound annual growth rate. That is a served market, not an independent total addressable market, and the sponsor has an incentive to define it favorably. Micronics Japan’s separate assumption of roughly 20% probe-card growth in 2024–26 supports a much faster near-term cycle but says little about normalized growth after capacity catches up.

Supply: concentrated expertise, active response

The advanced probe-card market has meaningful entry barriers. MEMS process development, contact performance, signal/thermal simulation, design software, production yield, customer-specific application support, and multi-quarter qualification require accumulated knowledge. A new supplier cannot simply machine a generic card and compete on labor cost. Leading customers value yield protection and on-time ramp execution enough to avoid an unproven interface. These barriers favor established suppliers and permit attractive economics during technology transitions.

The barriers do not prevent capable incumbents from expanding. Technoprobe’s Q2 revenue grew 64.3% year over year to €277.1M, with 57.2% gross margin and a striking 49.4% EBITDA margin. Its H1 investment was €90.3M, and its updated capacity program totals roughly €350M through Q1 2027, including logic and HBM machinery and a greenfield plant. Micronics Japan’s Q2 probe-card sales rose 50.6% to ¥27.9B, with a 41% segment margin; memory was 95% of probe-card sales. It is accelerating Aomori and guides FY2026 probe-card sales to ¥102B, up 48.9%. The results show a powerful industry upcycle. They also show that FORM is not the only beneficiary and that supply will not remain fixed.

The top-tier structure resembles differentiated oligopoly more than commodity competition. Suppliers specialize by architecture and application, but customers can dual-source and use the next product generation to reopen competition. Two of Technoprobe’s customers each represented about 30% of H1 revenue. This symmetry—few qualified suppliers selling to even fewer very large buyers—creates mutual dependence but leaves bargaining power with customers when capacity loosens. Prices need not collapse for returns to normalize; shorter lead times, lower expedite revenue, more qualification spending, and underutilized incremental capacity can do the work.

Regulation and geography

U.S. controls on advanced semiconductors, HBM, and manufacturing equipment constrain some China-linked demand, although probe cards do not map perfectly to every controlled equipment category. No material new BIS semiconductor/HBM rule affecting FormFactor was identified between June 27 and August 30, 2026. A January 2026 policy allowed case-by-case licensing for certain advanced chips under safeguards, while earlier restrictions remained. FORM’s China revenue was 6.5% in Q2, up from 4.8% a year earlier, and 5.8% in H1 versus 6.3%. The quarter therefore showed partial recovery rather than a new regulatory shock. Investors should not confuse stability over two months with removal of geopolitical risk.

Tariff refunds are a separate accounting issue. FormFactor received $0.8M by Q2-end and expects $7–9M in Q3 from IEEPA-related refunds. These benefits lift reported gross margin but do not reflect customer value, production learning, or recurring pricing. They should be removed from any normalized industry-return judgment.

Capital-cycle conclusion

The Marathon framework asks where capital is entering, not merely where demand is growing. All three leading probe-card suppliers are expanding into record results. The most plausible sequence is: higher test intensity tightens qualified supply; strong utilization and mix raise margins; suppliers accelerate capacity; start-up expense and depreciation appear; customers use greater availability to dual-source or negotiate; and returns normalize even if final demand continues growing. This is not a forecast of collapse. It is a warning that a good demand industry can produce mediocre equity returns when valuation and supply assume the good news will persist.

Verdict. The leading-edge probe-card and optical-test niches are mixed-attractive: structural content growth and qualification barriers support value capture, but concentrated customers, short design generations, semiconductor cyclicality, and visible 2026–27 capacity additions limit structural pricing power. FORM can earn above-history margins during the transition without having become a wide-moat compounder.

4. Competitive Position

What the moat actually is

Greenwald’s framework separates supply advantages, customer captivity, and scale economies. FormFactor has some of each, but none is absolute.

The supply-side advantage is accumulated process and application know-how. A high-parallelism probe card must route tester resources across the wafer, hold low and consistent contact resistance, control force and planarity, preserve high-frequency signals, tolerate temperature changes, shed contamination, and manufacture microscopic contacts at high yield. SmartMatrix’s ATRE resource-sharing architecture can raise parallelism above 3,000 sites per wafer while preserving signal fidelity and reducing tester time. At HBM4 speeds above 10Gbps, small losses or contact variation become production economics. Years of manufacturing learning, design libraries, failure data, and application-engineer experience are difficult to reproduce quickly.

Customer captivity exists at the product generation. Qualification consumes engineering time and wafers, and a poor switch can reduce yield or delay a valuable launch. Once a card is working in production, the buyer has little incentive to change solely for a small unit-price concession because tester time and package yield dwarf interface cost. FormFactor also provides repair and support near major customers. That combination creates meaningful switching friction during a ramp.

The captivity resets. A card is customized to a chip design, not a perpetual platform contract. At the next device generation, customers can qualify another supplier, split volumes, or change the test flow. Large memory and foundry customers have the scale and technical competence to sponsor alternatives. The company discloses neither retention rates nor wallet share, so one should not infer account-wide lock-in from a successful program. There is no network effect: one customer’s adoption does not make the card more valuable to another customer except indirectly through manufacturing learning and reputation.

Scale advantages are present but bounded. FormFactor spreads R&D, design automation, process development, field applications, and service infrastructure across a broad customer base. At high utilization, it absorbs fixed manufacturing cost and produces substantial operating leverage. Technoprobe’s current profitability shows that FORM lacks unique global scale; a direct rival can achieve equal or better process economics. Scale also becomes a disadvantage if the Farmers Branch footprint is underloaded.

Rivalry by category

Technoprobe and Micronics Japan are the most relevant direct comparisons. Technoprobe has leading-edge logic relationships, advanced MEMS capability, high margins, and an aggressive investment program. Micronics Japan is especially strong in memory and is expanding into the same HBM demand. Japan Electronic Materials and regional suppliers add competition in selected categories. Teradyne, Advantest, Cohu, and TEL are better treated as adjacent ecosystem participants: they sell testers, handlers, contactors, or partner equipment and validate test intensity, but they are not clean substitutes for FORM’s core high-end probe cards.

FormFactor appears strongest where full-wafer memory parallelism and complex production support matter. Management’s claim of strong high-speed stack-test share at two of three HBM customers is consistent with revenue and margin acceleration. Yet competitors’ even faster reported growth prevents attribution of the industry ramp to company-specific share gains. Public evidence supports participation and technical relevance, not a comprehensive share victory.

In Foundry & Logic, the position is broad but contested. Networking and CPU demand contributed to Q2, initial custom-ASIC revenue is emerging, and advanced-MEMS GPU cards began production shipments in Q3. Management says the fabless CPU customer remains only a low-single-digit share and constrained until Farmers Branch, while GPU is mostly competitor business today. This creates upside if capacity and qualifications convert, but the low starting share also proves that FormFactor is not the default supplier across leading-edge logic.

Systems may create a differentiated integration advantage. Triton combines nine-axis nano-precision optical alignment, electrical test, software, automation, and partner components from Advantest and TEL. Production customers value a validated cell more than disconnected laboratory instruments. The company’s installed silicon-photonics base and applications expertise can shorten adoption. The counterargument is maturity: CPO is early, customer test time and yield are uncertain, and key ecosystem components come from partners. A promising joint solution is not yet a standalone moat.

Economic proof and disconfirming evidence

Q2 supplies the strongest economic evidence in years. Probe Cards gross margin increased to 54.4% from 38.3% a year earlier; Systems rose to 48.5% from 39.4%. Consolidated operating margin reached 22.4%. The filing attributes the step to favorable mix, higher utilization, restructuring, yield and manufacturing-spend improvements, and shorter cycle times. Several of those mechanisms reflect competence and process learning. Several also reverse when volume and mix weaken. Management’s disclosure that pricing was not the main historical driver remains consistent with a narrow, not wide, moat.

Customer concentration provides another reality check. SK hynix and TSMC at 35.3% of revenue can be interpreted as validation by the industry’s best customers. It can equally be interpreted as buyer leverage and program risk. Intel’s move below 10% improved accounting concentration, but the company still depends on a handful of advanced-production ecosystems. Likewise, Technoprobe’s two 30% customers and Micronics’ 95% memory mix reveal that concentration is an industry feature rather than a FORM-specific failure.

Verdict: narrow / moderate competitive advantage, with evidence upgraded after Q2. Custom engineering, MEMS/process know-how, design tools, qualification risk, and service create real protection within a product generation. Q2 demonstrates that the franchise can earn strong economics at scale. Short generations, customer dual-sourcing, capable rivals, buyer concentration, and a synchronized capacity response prevent a wide-moat conclusion. The most important disconfirming evidence for the narrow-moat view would be several years of stable 50%+ normalized gross margin and sustained share across successive device transitions; the most damaging evidence would be qualification losses or margin erosion despite strong end demand.

5. Growth History and Forward Opportunities

The history before the inflection

Long-term growth has been uneven. Revenue was $769.7M in 2021, $747.9M in 2022, $663.1M in 2023, $763.6M in 2024, and $785.0M in 2025. The five-year endpoint barely exceeded the starting point. Probe Cards moved from $633.3M in 2021 to $637.9M in 2025, while Systems rose from $136.4M to $147.1M, with the FRT disposal complicating comparisons. Foundry & Logic fell from $435.8M in 2021 to $369.9M in 2025; DRAM rose from $156.0M to $247.4M; Flash fell from $41.4M to $20.6M. The portfolio thus shifted toward memory/HBM before consolidated growth convincingly accelerated.

This record matters because it defines the burden of proof. A company can possess strong technology and still grow only at the cycle’s pace if customers capture the savings, products reset each generation, and end-market contractions offset content gains. The 2021–25 numbers describe a cyclical supplier with mix changes, not a secular compounder. They also create operating leverage: R&D, service, and manufacturing built for advanced programs can support much more revenue when utilization rises.

The first half of 2026 shows that leverage. Q1 revenue reached $226.1M and Q2 $258.2M, bringing H1 to $484.4M, up 31.9% from $367.2M a year earlier. Q1 filed GAAP operating margin was only 7.4%, not the 10.5% shown by one aggregator, because the third-party measure omitted $7.1M of factory-start-up cost. Q2 then reached 22.4% after volume, mix, restructuring, and manufacturing improvement. Q3’s $270M midpoint would annualize to $1.08B, but a peak-quarter annualization is not a forecast. It establishes near-term capacity demand; it does not establish through-cycle revenue.

HBM and advanced memory

HBM is the clearest structural opportunity. More layers and faster interfaces raise the number and difficulty of test insertions. Known-good-die economics become more valuable as scarce DRAM and logic die enter costly stacks. FormFactor’s strong high-speed stack-test position at two major HBM customers and support for HBM4 above 10Gbps place it directly in that content increase. Samsung’s HBM4 production and industry-wide capacity plans supply end-market corroboration.

The opportunity is larger than a single HBM revenue line because memory customers allocate wafer starts dynamically. Management expects Q3 DRAM revenue roughly flat while shifting heavily toward DDR5, whose constrained wafers offer customers attractive economics. FORM can participate in either mix if it has qualified products. That breadth is a positive. It also means investors should model memory test as a portfolio of programs with different ASP, utilization, and margin—not multiply an HBM growth rate across the entire category.

Risks include customer concentration, architecture changes, and competitor qualifications. Two customers drive the present HBM4 ramp. Micronics Japan is 95% memory in probe cards and expanding capacity; Technoprobe is investing in HBM machinery. A future generation can change contact architecture or tester-resource needs, and a customer can qualify two suppliers. The durable value lies in repeated engineering relevance, not permanent ownership of today’s socket.

Foundry, logic, and custom compute

Foundry & Logic is the most important diversification test because it is larger than DRAM and contains several independent programs. Q2 growth came from CPU, networking, and initial hyperscaler custom-ASIC activity. Management began shipping production units of an advanced-MEMS GPU card in Q3. The fabless CPU opportunity is only a low-single-digit share today and capacity-constrained until Farmers Branch. These facts create credible paths to growth without assuming share already won.

Custom accelerators and networking silicon can raise test content through large die, advanced nodes, chiplets, and high-speed I/O. Hyperscalers also diversify the end-demand chain beyond merchant GPU suppliers. Yet foundry concentration remains: TSMC was 11% of Q2 company revenue, and the ultimate wafer programs often run through one advanced foundry. A design win can produce large revenue but also sharp program transitions. The appropriate growth model uses probability-weighted ramps and replacement cycles, not software-like retention.

The GPU qualification is strategically important because management says the category is currently mostly competitor business. Success would extend FormFactor’s process platform into a large application and improve California/Texas asset utilization. Failure would reveal a category-specific competitive gap even if memory remains strong. Investors should ask for production-volume evidence, multi-generation retention, and customer breadth rather than relying on the word “qualified.”

Co-packaged optics and Systems

Optical I/O is a smaller but potentially valuable vector. Electrical interconnects consume more power and encounter reach limitations as AI clusters scale. Co-packaged optics moves optical engines closer to switching or compute silicon, introducing precise alignment, thermal, electrical, and optical test requirements in production. Triton addresses that workflow as an integrated cell rather than a single instrument.

Management raised 2026 CPO revenue expectations above $20M and expects the threshold to be crossed by Q3. Against $48.5M of total Q2 Systems revenue, the category is already relevant to growth but not large enough to determine company value. The company-sponsored 2030 served opportunity of $400M should be treated as an addressable pool, not a forecast. Adoption architecture, yield learning, test time, customers’ make-versus-buy decisions, and partner economics will determine capture. Keystone Photonics, acquired for about $20.6M net in December 2025, may accelerate capability; standalone revenue and return are not disclosed.

The existing Systems franchise provides ballast through engineering probe stations, thermal products, and cryogenic tools. It can benefit from quantum and advanced-device research even when high-volume probe-card programs slow. The 2023 FRT sale reduced reported segment comparability, however, and Systems gross margin fell from above 50% in 2022–23 to 41.8% in 2025 before rebounding. This is a differentiated equipment portfolio, not an annuity.

The 2030 ambition and probability bridge

At Investor Day management targeted $1.6B of revenue, 55% non-GAAP gross margin, and more than $5 of non-GAAP EPS by 2030. Revenue would roughly double from 2025. A path can be constructed: a roughly $1.08B Q3 run rate, continued HBM/DDR intensity, larger GPU/custom-ASIC participation, CPO growth, and Farmers Branch capacity. The target is not arithmetically absurd.

The probability question is harder. Reaching $1.6B requires roughly 19.5% annual growth from the $785M 2025 base. The company’s 2021–25 revenue was essentially flat. The near-term cycle supplies a large first step, but 2027–30 must add growth after the industry capacity response, while maintaining price/mix and avoiding a memory digestion. Reaching 55% non-GAAP gross margin requires reported improvements beyond a normalized Q2 baseline near 51%, while absorbing new-factory costs and excluding SBC, amortization, restructuring, and other adjustments. More than $5 non-GAAP EPS may therefore be achieved with a wider gap to GAAP and cash economics than the headline suggests.

Verdict. Growth has shifted from a plausible narrative to visible operating evidence. HBM, advanced logic, custom ASICs, and CPO are distinct drivers, and non-HBM supplied roughly two-thirds of Q2 revenue and most of the total growth. The evidence supports a structural step above the 2021–25 base, but not the full 2030 target. The central forecast error is duration: how much of the current run rate survives customer mix changes, peer capacity, and the 2027 factory transition.

6. Financial Quality

Filed five-year record

The table below uses filed GAAP figures and simple free cash flow defined as operating cash flow less capital expenditures. FY2023 and FY2024 operating income include material divestiture gains and are not clean measures of recurring profitability.

FY2023 included a $73.0M pre-tax gain on the FRT divestiture, leaving roughly $9.8M, or 1.5% margin, after a simple subtraction. FY2024 included a $20.6M gain on the China-operations sale, leaving roughly $44.2M, or 5.8%. FY2025 was cleaner but included $3.0M of Farmers Branch start-up expense. The economic record through 2025 is therefore weaker than unadjusted operating income suggests: gross margin stayed around 39–42%, underlying operating margin was volatile, and free cash flow depended on capex timing.

The prior June baseline used $60.1M and 7.7% for FY2025 operating income; the filed figure is $57.1M and 7.27%. It also used a 10.5% Q1 2026 operating margin from an aggregator that omitted factory start-up costs; filed GAAP operating income was $16.6M and 7.36%. These corrections do not weaken the Q2 result. They make the sequential step more dramatic and reinforce the need to reconcile every ratio to the filing.

Q2 margin bridge and earnings quality

Q2 revenue of $258.2M produced $130.9M of gross profit and 50.7% GAAP gross margin. R&D was $31.1M, SG&A $37.2M, and factory start-up expense $4.9M, leaving $57.8M of operating income. The ROIC.ai “operating income” field of $62.7M excludes the separately presented start-up line; its EBIT field of $57.8M matches the filing. The filed measure is authoritative.

Non-GAAP operating margin was 27.9% after excluding $8.6M of SBC, $4.5M of restructuring, $0.9M of acquisition amortization/fair-value cost, and $0.1M of transaction costs. Some adjustments help compare operations; none disappear economically by definition. SBC transfers value to employees, recurring restructuring can represent a durable cost of adapting capacity, and factory costs become ordinary depreciation and labor after start-up. A normalized view should distinguish temporary cash charges from recurring economic consumption rather than accept or reject every adjustment mechanically.

Management’s gross-margin decomposition is unusually useful. About one-third of the sequential non-GAAP improvement came from baseline cost reduction, one-third from volume, and one-third from temporary tariff refunds and precious-metal recovery. Management estimated roughly 51% normalized non-GAAP gross margin at Q2 volumes versus 53.3% reported. Q3’s 54% guide includes $7–9M of tariff refunds. Removing the midpoint from $270M revenue reduces gross margin by about 300 basis points. Underlying Q3 economics can still be excellent around 51%, but the headline should not be capitalized as the new floor.

Q2 net income of $56.2M also benefited from a low 11.1% effective tax rate and $2.2M of equity-method income from FICT. H1’s tax rate was 8.9%, partly due to foreign-derived deduction and discrete stock-compensation benefits. A normalized long-run tax rate should be higher. FICT moved from prior-year loss to H1 income; because it sits below operating profit and can fluctuate independently, core forecasts should separate it.

Cash flow, working capital, and capex

H1 operating cash flow was $106.8M, versus $42.4M a year earlier. Simple free cash flow was $82.0M after $24.8M of capex; company-defined FCF was $83.3M after additional adjustments. The improvement is real and consistent with profit. It is not representative of full-year investment because management still guides 2026 cash capex to $140–170M, implying $115–145M in H2. H1 working capital consumed $25.9M as receivables rose $30.7M and inventory rose $19.6M, partially offset by payables and accruals.

Calculated Q2 days sales outstanding were about 51, inventory days 84, and payables days 47, producing an approximately 87-day cash-conversion cycle. Those figures are analytical estimates based on average balances, not reported KPIs. They illustrate why strong growth absorbs cash: custom WIP and finished cards must be built, qualified, and collected before the company realizes the full accounting profit.

FY2025 capex of $103.7M was 2.8 times depreciation and amortization of $37.6M. The 2026 guide would be roughly four to five times an annualized H1 D&A run rate. That investment can create future revenue without equal replacement capex, but it means trailing FCF overstates near-term owner earnings if H1 is isolated and understates future earnings if all growth capex is treated as maintenance. The proper model explicitly forecasts build, ramp, and normalized maintenance phases.

Balance sheet and returns on capital

June cash and marketable securities totaled $345.6M against $11.6M of term debt, or $334.0M net cash before lease liabilities. Including $18.3M of operating-lease liabilities still leaves about $315.7M. The $150M revolver is undrawn. Inventory of $121.4M and receivables of $155.8M are supported by current demand; goodwill of $212.6M and a $65.9M equity investment require separate judgment. The balance sheet makes a liquidity-driven permanent loss unlikely and gives the company room to complete Farmers Branch through a volatile year.

Returns historically failed to prove a compounder. A normalized calculation using average invested capital, normalized EBIT, and a 20% tax rate produces approximate operating ROIC of 7.6% in 2022, 1.3% in 2023 after removing the FRT gain, 5.9% in 2024 after removing the China gain, and 6.7% in 2025. ROIC.ai uses different definitions and reports lower values, but agrees on the direction: mostly below a reasonable cost of capital. Annualizing H1 2026 implies around 15%, evidence that utilization can lift returns, but it excludes much of the coming capex and overweights Q2. Sustainable value creation requires the new asset base to earn those higher returns after the ramp.

Shares and dilution

Diluted average shares stayed around 78–79M from 2021 through 2025 because repurchases historically offset awards. H1 SBC was $16.7M, 3.5% of revenue, while period-end shares increased 0.6% from year-end to 78.1M and Q2 diluted average shares increased 2.7% year over year to 79.6M. Shareholders approved five million additional plan shares in May, raising estimated fully diluted plan overhang from 7.5% to 11.3%. Full-value awards consume the reserve faster, so the headline reserve is not a forecast of five million issued common shares, but it is meaningful potential dilution.

Verdict. Financial quality improved substantially in Q2: revenue converted to filed GAAP profit and cash, both segments expanded margin, and the balance sheet is strong. The quality is not yet through-cycle. Temporary refunds, low tax, equity income, restructuring, factory expense, and back-loaded capex must be normalized separately. Pre-2026 ROIC was generally sub-cost-of-capital; the decisive question is whether Farmers Branch and higher test intensity sustain mid-teens returns after competitors’ capacity arrives.

7. Capital Allocation

Portfolio decisions and M&A

Management’s record is more disciplined than the historical ROIC alone suggests. The 2016 Cascade Microtech acquisition, about $325M net, created the modern Systems segment and broadened applications capability. Public disclosure does not permit a clean standalone return calculation, but the assets remain strategically coherent. Smaller acquisitions include Baldwin Park probe-card assets for $35M and High Precision Devices for $16.9M in 2020. The related Baldwin Park facility was closed during the 2026 restructuring as Flash weakened, a mixed outcome; HPD remains a small cryogenic adjacency without disclosed returns.

FRT is the clearest success. It cost about $24.4M net and was sold in 2023 for roughly $100.1M of net cash proceeds, producing a $73M gain. The 2024 sale of China operations generated $21.4M of proceeds and a $20.6M gain while retaining an exclusive distribution arrangement. Both exits converted non-core or geopolitically exposed assets into capital without visibly damaging the core franchise.

The current deployment is tightly linked to advanced packaging. FormFactor invested roughly $67.2M including costs for 20% of FICT, an advanced-substrate company. FICT produced a $2.0M equity loss in 2025 and $3.8M of income in H1 2026; the holding may provide strategic insight and economic participation, but one favorable half-year does not establish the return. Keystone Photonics cost $20.6M net in December 2025 and supports Triton/CPO. The higher 2026 CPO outlook is early validation, while undisclosed standalone revenue, margins, and purchase multiple prevent a quantitative score.

Farmers Branch: the defining capital decision

The Texas facility is not merely maintenance capex. It is a decision to replicate approximately the California probe-card capacity and support logic, HBM, and advanced-packaging growth. Capex of $140–170M in 2026, plus $25–30M of preproduction expense, arrives before meaningful accretion expected around 2028. If customers are genuinely capacity-constrained and new qualifications convert, the investment can reduce cycle times, localize supply, and spread R&D/design infrastructure over a larger revenue base. If demand digests as Technoprobe and MJC finish their own expansions, the same facility can depress returns through depreciation and underutilization.

The decision is strategically defensible because waiting could forfeit programs that require assured capacity. It is financially unproven because public filings do not disclose binding customer commitments, target utilization, or project-level return thresholds. Capital-cycle discipline requires judging the asset on post-ramp cash returns, not on announced square footage or peak-quarter revenue.

Repurchases, SBC, and dividends

The April 2025 $75M repurchase authorization has $70.9M remaining and expires in April 2027. No shares were repurchased in H1 2026 because cash was prioritized for Farmers Branch. Historically, repurchases kept the diluted count roughly flat despite $29–40M of annual SBC. Pausing them during heavy investment preserves balance-sheet flexibility, but the combination of no repurchase, rising shares, and an expanded equity plan means dilution can resume.

There is no dividend. For a cyclical technology supplier with valuable projects and a net-cash balance sheet, that policy is reasonable. The standard is not whether cash is returned each quarter; it is whether retained capital earns more than its cost. The 2027–28 facility outcome will be more informative than the current authorization balance.

Incentives, ownership, and insiders

The 2025 annual incentive was 70% adjusted operating income and 30% corporate goals. The financial component scored 85.6% and corporate goals zero, producing a 59.9% result; the committee showed willingness to withhold pay. The weakness is definition: adjusted operating income excludes SBC, restructuring, transaction items, and other selected costs. Long-term awards are half time-based RSUs and half relative-TSR performance shares. There is no explicit ROIC, FCF, or per-share fundamental measure, despite capital returns being the key strategic issue.

Governance is otherwise sound: independent chair and CEO roles, annual elections, majority voting, one share/one vote, clawback, and anti-hedging and pledging policies. CEO ownership was roughly 484,000 shares after August activity, below 1%, so management is economically exposed but not owner-operator aligned.

The trailing 60-month Form 4 corpus contains no code-P open-market purchase. Adjusted for amendments, insiders sold 589,248 shares for $37.25M; 212,478 shares were explicitly under 10b5-1 plans. Since June 27, CEO Mike Slessor sold 20,737 shares for $2.66M under an existing plan. August vesting largely offset the sales, leaving his holdings near the proxy level. Planned sales do not prove a negative view, but five years without discretionary buying provides no valuation-confidence signal. Awards, tax withholding, and exercises should not be mechanically netted because the same grant can appear in several codes.

Verdict: balanced / improving, not yet excellent. Management has preserved a net-cash balance sheet, kept the historical share count stable, realized exceptional value in FRT, exited China cleanly, and focused R&D and acquisitions on coherent niches. The offsets are a large capacity bet near a cyclical high, paused repurchases, 11.3% equity-plan overhang, modest insider ownership, addback-heavy incentives, and limited deal-level return disclosure. Farmers Branch utilization will determine whether the next chapter upgrades this record or exposes overinvestment.

8. Changes and Headwinds — Last Two Years

The last two years transformed both the operating portfolio and the fixed-cost base. The China-operations sale reduced direct geopolitical exposure while preserving distribution. FICT added substrate participation; Keystone Photonics and Triton added optical-test capability. Farmers Branch moved from construction to a late-2026 production start. A January 2026 restructuring consolidated California operations, affects 200–300 employees, and is now expected to cost $32–40M, chiefly within Probe Cards. Carlsbad manufacturing continues through December 2026. These actions can raise efficiency but make comparisons noisy because savings, charges, and start-up costs overlap.

Demand also changed. HBM emerged as roughly two-thirds of Q2 DRAM revenue, advanced logic recovered, initial custom-ASIC activity began, GPU production shipments started, and CPO expectations moved above $20M. Named-customer concentration above 10% fell in Q2 because Intel moved below the threshold, though H1 concentration was essentially unchanged. China revenue recovered modestly. No material new M&A, financing, litigation, management departure, or semiconductor-control rule appeared after the June baseline.

The most favorable change is economic breadth. Of the $62.4M year-over-year Q2 revenue increase, roughly $21M came from HBM, or about 34%. Foundry & Logic and Systems supplied the rest of the expansion while Flash contracted. This weakens the claim that the entire re-base is one memory product. The second favorable change is cash conversion: H1 OCF more than doubled and capex was temporarily low.

The main headwinds are embedded in the same progress. Q3’s HBM-to-DDR rotation tests mix sensitivity. The expected tariff refund obscures the next gross-margin comparison. H2 capex is sharply back-loaded. Farmers Branch, Technoprobe, and MJC capacity arrives into 2027. Customer concentration and lack of public wallet-share data limit confidence that program wins persist. The equity-plan expansion and zero open-market insider buying weaken per-share alignment at a premium valuation.

The stock’s decline is itself a changed condition, not a business headwind. A 36% fall from the June 30 close high reduced the valuation percentile and the probability that technical index flows dominate price. It did not make the stock statistically low risk: specific volatility is about 51%, five-year volatility 60.6%, and five-year maximum drawdown 62.7%. Investors must distinguish an improved entry context from proof that multiple compression is complete.

Verdict. Operating changes over the last two years are net positive and Q2 strengthened their credibility. The countervailing change is a synchronized industry investment wave and a larger fixed-cost base. The next clean evidence will come not from another adjusted-margin peak but from normalized margins, share, and cash returns through the 2027 ramp.

9. Risk Analysis

The risk matrix separates business failure from price loss. A total loss is remote under current facts because the company has net cash, critical products, positive operating cash flow, and no refinancing wall. A severe mark-to-market loss is entirely plausible because the stock combines a premium multiple, cyclical earnings, concentrated customers, and high empirical volatility. The historical lifetime maximum drawdown of 92.4% is not a forecast, but it demonstrates how operating survival and shareholder experience can diverge.

The most dangerous correlated scenario is not a single failed product. It is a 2027 digestion in which customer wafer allocations slow, direct rivals’ capacity becomes available, FORM absorbs Texas depreciation and labor, and the market de-rates semiconductor-test multiples simultaneously. Revenue, margin, and valuation would then move adversely together. Net cash would protect the enterprise but not the share price.

The constructive correlated scenario is the reverse: HBM4, custom accelerators, GPU, and CPO absorb capacity; normalized gross margin holds near or above 50%; new programs diversify concentration; and factory utilization converts capex into mid-teens or better ROIC. The risk discipline is to track those operating variables rather than treating price volatility itself as proof of opportunity or danger.

Verdict. Permanent impairment risk is low at the corporate level and high at the valuation level. The dominant risks are duration, utilization, and multiple—not solvency. Position sizing, if any, should reflect a stock whose business has improved faster than its empirical drawdown profile.

10. Valuation Discussion — Embedded Expectations

Current bridge and denominator quality

At the August 28 close of $101.70, 78.120725M period-end shares produce a computed basic market capitalization of $7.945B. June cash plus marketable securities were $345.640M and building debt $11.644M, leaving $333.996M net cash before operating leases. Standard enterprise value is therefore $7.611B. Removing the $65.891M FICT equity-method investment as a non-operating asset produces a $7.545B operating enterprise value. Including $18.268M of lease liabilities would change operating EV by only 0.2%; the convention is not thesis-driving.

Trailing figures through Q2 were $902.225M revenue, $153.284M EBITDA, $130.543M filed EBIT, and $136.121M provider-defined free cash flow. The filed EBIT controls because ROIC.ai’s operating-income field omitted Q2 factory start-up expense; its EBIT field reconciled to $57.799M of Q2 filed operating income. The current operating EV is 8.36x revenue, 49.2x EBITDA, and 57.8x EBIT. AZI reports 68.4x P/E on $1.4859 of trailing EPS, 8.76x P/S, and 7.15x P/B.

H1 contained $27.814M of restructuring, of which $17.55M was non-cash. Adding the full H1 charge to trailing EBITDA and EBIT yields $181.098M and $158.357M; current operating EV remains 41.7x and 47.6x. This adjustment recognizes the discrete California consolidation without pretending every future restructuring is free. The valuation does not add back $11.933M of H1 Farmers Branch start-up expense to sustainable profit because management expects those costs to migrate into cost of revenue and acknowledges 2027 ramp inefficiency.

Annualized Q2 revenue is $1.033B and filed EBIT $231M. Adding back only Q2 restructuring produces roughly $249M annualized EBIT. The operating EV is 32.6x the reported peak-quarter annualization and 30.3x the restructuring-adjusted version. These checks show that valuation is less extreme on the current run rate than on trailing results. They are not forecasts: a single quarter ignores mix changes, back-loaded capex, ramp costs, and peer capacity.

Free-cash-flow yield is not a clean shortcut. The trailing provider-defined $136.1M is only a 1.7% equity yield and includes a period when H1 capex was $24.8M against $140–170M full-year guidance. Most factory cash spending sits in H2. EBIT, revenue, and normalized gross margin are more useful until the build and utilization stabilize.

Own history and peers

AZI’s August 28 own-history composite valuation percentile is 87.73, with P/E at 73.41, P/B 95.17, and P/S 94.61. Earnings caught up and the price declined, so the composite is well below June’s extreme. Book and sales remain near the richest 5% of the stock’s own history. These percentiles compare FORM with itself, not with peers, and provide context rather than intrinsic value.

FORM now has the lowest sales multiple among the profitable premium cohort in this table, but the discount reflects lower through-cycle margins and returns. On reported EBITDA and EBIT it remains more expensive than Teradyne and KLA; after restructuring normalization, it remains above Technoprobe’s forward EBITDA multiple. Onto is the principal exception because acquisition effects depress its denominator. Relative cheapness on sales is not automatically a margin of safety when each sales dollar historically generated less durable profit.

Earnings power and static requirements

Net tangible book is roughly $880M after subtracting $212.6M of goodwill and $18.0M of acquired intangibles from $1.111B of equity. That understates organically created know-how and qualifications, so it is a floor diagnostic, not reproduction value. A no-growth earnings-power calculation using $158.4M of restructuring-adjusted trailing EBIT, a 20% tax rate, and a 9.5% capitalization rate produces about $1.33B of operating value. Annualizing Q2 adjusted EBIT raises that diagnostic to roughly $2.10B. Current value is therefore predominantly franchise growth, not assets or present no-growth earnings.

Capitalizing $7.545B of operating EV at 20x, 25x, or 30x sustainable EBIT requires $377M, $302M, or $251M of annual EBIT. At a 20% margin, those profits require $1.89B, $1.51B, or $1.26B of revenue; at 25%, $1.51B, $1.21B, or $1.01B. The lower revenue requirements depend on unusually high mature margins and premium multiples. The table is generous because it does not charge for time.

Compounding current operating EV for four years at a 9.5% required return produces approximately $10.85B of 2030 terminal EV before interim cash flow. If revenue reaches management’s $1.6B target, that value equals 6.8x sales and needs a 33.9x exit EBIT multiple at a 20% margin, 27.1x at 25%, or 22.6x at 30%. Interim cash flow reduces the terminal burden, but current value requires both the revenue destination and exceptional margin or multiple durability.

Scenario coverage

The following scenarios measure present operating-value coverage; they are not assigned probabilities and are not price objectives. All begin with an explicit $1.024B 2026 revenue estimate equal to H1 actual, the Q3 midpoint, and a flat Q4. Annual FCFF and terminal operating EV are discounted at 9.5%; non-operating assets are excluded from numerator and denominator.

The bear assumes a supply/digestion outcome and partial reversion toward historical margins. The base gives FORM growth near or above its company-sponsored served-market rate and a durable margin far above 2021–25, yet covers only 60% of current operating value. The bull gives full credit to management’s revenue destination, a 25% GAAP EBIT margin, high-teens FCFF, and a 28x mature exit multiple; it exceeds current value by only 13%. Current valuation is therefore much closer to a demanding bull than to a reasonable base.

Sensitivity does not change the load-bearing assumptions. In the base, each extra $100M of 2030 revenue adds roughly $300M of present value; each 100 basis points of 2030 EBIT margin adds about $225M; one exit-multiple turn adds about $169M. Moving the discount rate from 9.5% to 8.5% increases coverage only to 0.63x; moving it to 10.5% lowers coverage to 0.58x. Revenue, margin persistence, and terminal multiple dominate modest balance-sheet and discount-rate changes.

A ten-year reverse DCF reinforces the result. Starting from $1.024B of 2026 revenue, a 9.5% discount rate and 3% terminal growth, current operating EV requires about 17.2% annual revenue growth for ten years at a 16% FCFF margin, reaching roughly $5.0B of year-ten revenue. At 18% FCFF margin, required growth is 15.6%; at 20%, 14.2%. These are not forecasts. They show that a cyclical-hardware cash margin in the low-to-mid teens cannot support the current value without exceptional growth duration.

Embedded-expectations verdict

The market may be correctly underwriting a structural step. Q2 was broader than HBM, Q3 revenue guidance is strong, known-good-die economics raise customer value, and Farmers Branch removes a genuine constraint. Technoprobe’s still-higher sales multiple shows that investors are capitalizing an industry-wide probe-card boom rather than a FORM-specific fantasy.

The fragile assumption is supply, not demand enthusiasm. The valuation treats utilization-led margin, restructuring savings, and leading-edge mix as durable before the new capacity wave has been tested. It gives limited weight to H2 cash capex, 2027 ramp inefficiency, customer concentration, dilution overhang, and a Q3 gross-margin guide containing about 300 basis points of refund. Verdict: Q2 materially strengthened the earnings denominator, but current operating value still requires something close to management’s 2030 revenue ambition plus premium terminal economics. Downside protection comes from future execution, not tangible assets or current no-growth earnings.

11. Variant Perception

The prevailing interpretation

The constructive consensus is straightforward: AI raises test intensity faster than wafer volumes; FormFactor leads critical HBM probe-card applications; Q2 proves that revenue converts into much higher margins; and Farmers Branch provides the capacity to serve HBM4, GPU, custom accelerators, and CPO into 2030. Management’s $1.6B revenue, 55% non-GAAP gross-margin, and greater-than-$5 non-GAAP EPS ambition gives the market a convenient endpoint. The net-cash balance sheet and technical qualifications reduce financing and execution risk relative to a speculative equipment supplier.

That interpretation has stronger support than it did in June. Q2 did not merely beat a low estimate: it exceeded the prior report’s explicit $240M revenue and 16% operating-margin hurdles, produced $52.6M of company-defined FCF, and grew non-DRAM categories. Q3 guidance points to a likely second revenue pass. A skeptical view that ignores this evidence would be anchored to the old financial record.

The strongest constructive case

The strongest case is that known-good-die economics have structurally changed the amount customers will spend on wafer-level test. HBM4 and advanced packages combine more expensive components and make late defect discovery intolerable. SmartMatrix’s high parallelism and signal integrity solve a binding production problem, and qualification protects each program. Foundry & Logic adds large CPU, networking, GPU, and hyperscaler custom-ASIC opportunities; Triton adds an optical test insertion that barely existed in prior cycles. If FORM retains share, the 2025 revenue base becomes obsolete rather than mean-reverting.

In that outcome, Q2’s margin is not simply a peak. A portion of restructuring, cycle-time, yield, and manufacturing-spend improvement persists. R&D and corporate expense grow slower than revenue. Farmers Branch removes capacity constraints and eventually improves footprint economics. A strong balance sheet prevents forced retrenchment during the build. The company can reach materially higher revenue and operating profit even if the valuation multiple normalizes.

The evidence that would strengthen this case is successive-generation share, not another HBM headline: production GPU shipments scaling, custom-ASIC programs broadening, CPO repeat orders, normalized gross margin around or above 50% after refunds, and 2027 factory utilization without heavy discounts. Customer concentration falling for economic—not merely disclosure-threshold—reasons would further improve quality.

The strongest skeptical case

The skeptical case is not that AI or HBM is fictional. It is that capital markets are capitalizing a favorable point in a supply-constrained cycle before the supply response arrives. FORM’s 2021–25 revenue was essentially flat, normalized ROIC mostly below a reasonable cost of capital, and gross margin around 39–42%. Q2 combined favorable mix, high utilization, restructuring savings, temporary recoveries, and low tax. The next quarter includes an even larger tariff refund. Those conditions can produce real cash and still be above normalized earnings.

Direct rivals make the skepticism concrete. Technoprobe grew faster, earns higher current margins, and is investing roughly €350M through early 2027. Micronics Japan is expanding a memory-heavy franchise after 50% growth. FORM itself is replicating its California capacity. Customers can dual-source at the next device generation and have more bargaining power once lead times ease. A 2027 utilization pause could therefore compress revenue growth and margins at the same time depreciation rises.

The multiple magnifies this operational risk. At roughly 50x trailing EBITDA and near the 95th percentile of its own P/S and P/B history, FORM does not need a recession for the equity to disappoint. It only needs growth, margin, or duration to land below a bullish path. The July–August tape supports this variant: record results restored part of the decline but did not recover the June multiple.

The differentiated synthesis

The useful variant is between the extremes. FormFactor has likely achieved a structural earnings step above the 2021–25 base, but Q2 is also above clean through-cycle economics. The narrow moat can protect qualifications and produce high returns at full utilization without guaranteeing pricing power after a synchronized capacity build. The price reset partially recognizes that distinction; it does not eliminate it.

Four swing variables organize the debate: (1) normalized gross margin after refunds and start-up noise; (2) revenue breadth and successor-generation qualification; (3) Farmers Branch and peer utilization during 2027; and (4) the terminal multiple applied to a still-cyclical earnings stream. The variant perception is therefore empirical and falsifiable: neither “AI winner” nor “cycle peak” is sufficient without the next four to six quarters of mix, margin, and capacity evidence.

12. Fact vs. Interpretation Table

13. Open Questions

1. What share of HBM4 wafer and stack-test revenue is sole-sourced, dual-sourced, or in active requalification at each of the three major memory customers?
2. How much of Q2 Probe Cards gross-margin expansion came from utilization, mix, structural yield improvement, restructuring, customer expedite behavior, and precious-metal recovery in dollar terms?
3. What normalized gross margin does management expect after the Q3 tariff refund, the HBM-to-DDR mix shift, and full Farmers Branch depreciation?
4. What binding customer commitments, qualification milestones, and target utilization supported the decision to replicate California capacity in Texas?
5. What revenue level and mix are required for Farmers Branch to become accretive, and how much underutilization expense would appear at 60%, 70%, and 80% loading?
6. Do advanced-MEMS GPU production shipments represent one program or a platform qualification capable of carrying into successive generations?
7. What is FORM’s current wallet share in fabless CPU, networking, custom accelerator, and GPU categories, and how does it compare with Technoprobe and MJC?
8. How much of Triton’s greater-than-$20M 2026 CPO revenue reflects repeat production cells versus initial evaluation or line-qualification systems?
9. What gross margin and service burden accompany Triton, and how are economics shared with Advantest, TEL, and other partners?
10. What standalone revenue, operating contribution, and invested-capital return have Cascade, HPD, FICT, and Keystone generated?
11. What portion of the five-million-share plan reserve is expected to become full-value awards, and what repurchase cadence would keep per-share dilution neutral during the factory build?
12. Why does long-term incentive design omit ROIC, free cash flow, and per-share operating measures when the largest strategic decision is capital deployment?
13. How sensitive is HBM probe-card demand to memory customers reallocating constrained wafer starts toward DDR, and what lead time does FORM receive before those changes?
14. Are Taiwan outsourced assembly and service activities reducing cycle time without creating IP leakage, quality variation, or dependency on a new partner?
15. What evidence would management accept as showing that the industry added too much 2027 capacity, and how quickly could capex or headcount be flexed?

These questions are not gaps that make analysis impossible. They identify where public evidence is weakest and where the next filing or call can most efficiently update the thesis.

14. What Must Be True

Constructive case

The constructive case requires more than continued AI spending. Quarterly revenue must remain broadly above the old $240M hurdle as mix rotates; non-DRAM categories must continue to supply a majority of revenue and a material share of growth; normalized non-GAAP gross margin must hold around 50% or better after tariff refunds; and successor-generation qualifications must convert in GPU, CPU, networking, custom ASIC, and HBM. Farmers Branch must ramp without a long period of underutilization, while peer expansions do not force material price concessions. Capitalized returns on the enlarged asset base should move sustainably into the mid-teens.

Falsification test: the constructive case fails if, during otherwise healthy AI capital spending, revenue falls below $240M for two consecutive quarters, normalized gross margin drops below roughly 48%, and the weakness is associated with qualification loss, excess capacity, or customer sourcing changes rather than a temporary shipment schedule. A delayed Texas ramp combined with rising depreciation and declining share would be especially decisive.

Skeptical case

The skeptical case requires the current margin step to contain substantial cyclical and temporary benefit. Industry capacity must loosen qualified supply during 2027, customers must use the next generation to dual-source or negotiate, and Texas start-up/depreciation must arrive before enough volume. HBM growth alone must prove insufficient to offset a digestion or mix rotation, while CPO and custom-ASIC revenue remain too small or lumpy to stabilize the model. The valuation multiple must continue to normalize toward a cyclical-test framework.

Falsification test: the skeptical case fails if FORM reports at least four quarters around $260–280M or better, holds normalized GAAP operating margin in the high teens or above after tariff refunds and restructuring, grows Foundry & Logic and Systems through a DRAM digestion, and loads Farmers Branch without price or lead-time deterioration. Success across two successive device generations would be stronger evidence than any single quarter.

Base-case requirements

A middle outcome assumes a genuine structural revenue step but lower economics than the 2030 ambition: annual revenue eventually around $1.2–1.3B, normalized gross margin around 49–51%, operating margin in the high teens, and positive but volatile FCF during the factory ramp. This case requires neither collapse nor flawless execution. It does require that the company’s technical relevance offsets customer and competitive pressure enough to earn its cost of capital on new capacity.

The monitoring sequence is Q3 refund-adjusted margin and mix; Q4 Farmers Branch start; 2027 qualification, utilization, and peer-capacity data; then 2028 accretion and ROIC. Each observation should update both earnings power and the multiple appropriate for its durability.

15. Source Appendix

This memo prioritizes FormFactor’s SEC filings, filed earnings exhibits, proxy, company product and Investor Day materials, and primary competitor/customer disclosures. The detailed, dated bibliography—including accession links, market-data sources, factor-model sources, and public market context—is provided in the separate Source Appendix and carried into Appendix B of the combined report.

Key primary anchors are FormFactor’s FY2021–FY2025 Forms 10-K; Q1 and Q2 2026 Forms 10-Q; July 29, 2026 earnings Form 8-K and Exhibit 99.1; April 2, 2026 proxy; August 4, 2026 S-8; Form 4 corpus through August 17, 2026; May 11 Investor Day; SmartMatrix and Triton product materials; and the July 28 Keystone Microtech partnership release. Industry comparisons use Technoprobe H1 2026, Micronics Japan Q2 2026, Teradyne and Cohu Q2 filings, TSMC packaging materials, Samsung HBM4 disclosure, and BIS rules. Third-party AZI, FactorsToday, and ROIC.ai data are used for price, factor, valuation-history, and cross-check context; filings control when fields conflict.

APPENDIX A — Standard Diligence Questionnaire

Supplement to the August 30, 2026 research memo. Financial data through June 27, 2026; market data through August 28, 2026.

General — What thoughtful questions have other investors asked?

The recurring high-value questions are: How much HBM test content grows per stack generation? Is FormFactor gaining wallet share or merely participating in a strong industry? How durable is gross margin after utilization, tariff refunds, and precious-metal recovery normalize? Can GPU, custom-ASIC, and CPO programs diversify the two-customer HBM exposure? What utilization is required for Farmers Branch to become accretive? How much capacity are Technoprobe and Micronics Japan adding? Why should a historically mid-single-digit-return supplier sustain a premium multiple? These are the correct questions because each ties a technology narrative to a measurable economic variable.

Cyclicality & Earnings Nature

Are earnings at a cyclical high or low? Fact: Q2 revenue of $258.2M and GAAP operating margin of 22.4% were records, and Q3 guidance points higher. Interpretation: earnings are in a strong upcycle and may be near a favorable utilization/mix point, but the cycle can still expand as HBM4 and new logic programs ramp. The clean description is “cyclically high with possible structural re-base,” not a definitive peak.

External environment or company actions? Both. External HBM, advanced packaging, AI networking, and custom-accelerator demand raised volume and mix. Company-led restructuring, yield, manufacturing-spend, and cycle-time work improved the baseline. Management attributed roughly one-third of sequential non-GAAP gross-margin improvement to durable cost actions, one-third to volume, and one-third to nonrecurring items. This decomposition prevents attributing all improvement to either management or the cycle.

Revenue stability and outlook. Probe cards are customized and qualified for chip programs, providing visibility during a production ramp but limited contractual recurrence. Device generations and wafer starts drive orders; customers can dual-source at transitions. Q2 Foundry & Logic, DRAM, and Systems grew 22.4%, 49.0%, and 44.1%, respectively, while Flash fell 48.4%. The diversity is better than an HBM-only story, but quarterly revenue remains program- and cycle-sensitive.

Market size and geography. FORM’s company/Yole served-market estimate grows from $3.1B in 2025 to $4.5B in 2030, an 8% CAGR; it is sponsored, not independent. Growth is global and centered in Asian semiconductor production. Structural drivers are HBM layers/speed, known-good-die value, chiplets, advanced logic, and optical I/O. China was 6.5% of Q2 revenue and remains exposed to export policy.

Business Quality & Competitive Moat

Is competition increasing? Yes at the capacity level. Technoprobe is investing roughly €350M through early 2027, Micronics Japan is accelerating Aomori, and FORM is replicating its California probe-card footprint in Texas. Qualification barriers limit new entrants, but capable incumbents are responding aggressively.

Profitability and returns. Filed gross margin stayed around 39–42% in 2021–25 and normalized ROIC estimates were generally mid-single digit, below a reasonable cost of capital. Q2 gross margin reached 50.7% and annualized H1 returns improved materially. Interpretation: high utilization demonstrates attractive incremental economics, but post-ramp ROIC—not a peak-quarter annualization—is the quality test.

Barriers to entry and nature of competition. The business is technically understandable: a probe card makes thousands of reliable temporary electrical contacts so a tester can screen wafer die. Barriers include MEMS/contact manufacturing, signal and thermal engineering, design software, yields, customer qualification, and global service. Competition is differentiated oligopoly, not brand-driven consumer competition. Low-cost foreign labor alone cannot undermine the product because precision, process IP, and qualification dominate assembly labor; established Asian competitors can undermine it through comparable engineering and customer relationships.

Switching costs. Meaningful within a product generation, bounded across generations. A failed switch can harm yield or delay a ramp, so customers avoid needless mid-cycle changes. New chip designs reopen qualification and enable dual-sourcing. There is no network effect or permanent account captivity.

Financial Condition & Balance Sheet

Underrecognized assets. Proprietary MEMS know-how, design libraries, qualification history, application data, customer relationships, and the service network are not fully capitalized. The $65.9M FICT equity-method stake may contain strategic value beyond carrying value, but public evidence is insufficient. These assets justify value above tangible book; they do not remove cyclicality.

Off-balance-sheet liabilities. No material hidden financing was identified. Operating lease liabilities were about $18.3M. The larger economic commitments are the planned Farmers Branch capex and future labor/depreciation, plus potential equity issuance under the amended plan. Product warranty, purchase, and employment obligations appear ordinary for the model.

Accounting conservatism. Filed GAAP reporting is usable, but presentation requires normalization. FY2023 and FY2024 operating income included $73.0M and $20.6M divestiture gains. Q2 had restructuring and start-up costs; Q3 expects $7–9M of tariff refunds. Non-GAAP results exclude SBC, amortization, restructuring, transaction, and factory items. The accounting is not unusually aggressive, but adjusted measures are addback-heavy.

Capex intensity. High during the current expansion. FY2025 capex was $103.7M and 2026 guidance is $140–170M, versus $37.6M FY2025 D&A. Farmers Branch approximately replicates the California probe-card footprint. Maintenance capex should eventually be lower than build capex, but the incremental asset base must be utilized.

Capital Allocation & Management

FCF and its use. H1 simple FCF was $82.0M, but most 2026 capex is back-loaded. Management is prioritizing Farmers Branch, R&D, and small strategic investments over repurchases. There is no dividend. The philosophy is growth-and-technology reinvestment backed by net cash, with opportunistic dilution-offset repurchases.

Acquisitions and divestitures. Cascade created Systems; returns are undisclosed. FRT was an excellent realized outcome: roughly $24.4M net cost and about $100M net sale proceeds. The China exit generated a large gain while preserving distribution. FICT and Keystone align with substrates/CPO but remain too early for a return judgment. Baldwin Park was mixed because related manufacturing later closed as Flash weakened.

Repurchases and dilution. Historically, repurchases kept diluted shares near 78–79M despite SBC. No H1 2026 repurchases occurred; $70.9M authorization remains. Period-end shares rose 0.6% from year-end and Q2 diluted average shares 2.7% year over year. The May plan amendment raised estimated overhang from 7.5% to 11.3%. Actual issuance will be lower than the headline reserve if full-value awards consume 1.7 shares each, and depends on forfeitures and performance outcomes.

Compensation and motivation. Short-term pay emphasizes adjusted operating income; long-term awards combine time-based RSUs and relative-TSR PRSUs. The committee awarded zero on 2025 corporate goals, demonstrating some discipline. The weakness is no ROIC, FCF, or per-share fundamental metric. CEO ownership is below 1%. Five years of Form 4 data show zero open-market purchases and $37.25M of sales, although recent CEO sales were planned and vesting kept holdings approximately flat. Management appears professional and strategy-motivated, not owner-operator aligned.

Valuation & Market Data

Security form and dividend. FORM is a domestic NASDAQ common stock, not an ADR, MLP, partnership, or K-1 issuer. It pays no dividend.

Profitability. TTM revenue, EBITDA, and filed EBIT were $902.2M, $153.3M, and $130.5M. At August 28, operating EV was about 8.4x sales, 49.2x EBITDA, and 57.8x EBIT; restructuring-adjusted multiples were 41.7x and 47.6x. AZI’s own-history P/S and P/B ranks remained around the 95th percentile.

Net income versus operating cash. H1 net income was $76.6M and OCF $106.8M. Cash exceeded earnings despite working-capital investment because non-cash restructuring, SBC, and depreciation contributed. The relationship is healthy in H1, but it does not settle owner earnings because full-year capex is heavily back-loaded and the tax rate was low.

Risks & Downside

What could cause a decline? A semiconductor multiple de-rating; HBM/DDR digestion; qualification or wallet-share loss; 2027 capacity oversupply; Farmers Branch underutilization; normalized gross margin below roughly 49%; customer concentration; CPO delays; export-control changes; higher dilution; or a reversal in tax/equity-method benefits. Several can correlate during a downcycle.

Catastrophic or total loss risk. Low under current facts. The company has roughly $334M net cash before leases, an undrawn revolver, positive earnings, and indispensable products. Permanent enterprise failure would likely require prolonged technology displacement plus severe capital-allocation error. A large equity drawdown is much more likely than total loss: five-year maximum drawdown was 62.7% and lifetime maximum drawdown 92.4%.

Recent News & Events

The July 29 earnings release and August 4 10-Q showed record Q2 and a record Q3 guide. The July 28 Keystone Microtech partnership expands Taiwan assembly, repair, test, and distribution capability while FORM retains proprietary technology. The August S-8 registered five million additional equity-plan shares approved in May. Farmers Branch remains scheduled to begin production in late Q4. No post-June M&A, financing, litigation, executive departure, or material new BIS rule was identified. The business environment improved operationally; the chief new risk is the industry-wide capacity response rather than a fresh external shock.

APPENDIX B — Source Appendix

Sources accessed August 30, 2026 unless noted. SEC filings and company-filed exhibits control when third-party data disagree. Management estimates and transcript claims are identified as such in the memo.

Primary company filings

1. FormFactor, FY2025 Form 10-K, filed February 20, 2026: https://www.sec.gov/Archives/edgar/data/1039399/000103939926000009/form-20251227.htm
2. FormFactor, Q1 2026 Form 10-Q, filed May 5, 2026: https://www.sec.gov/Archives/edgar/data/1039399/000103939926000023/form-20260328.htm
3. FormFactor, Q2 2026 Form 10-Q, filed August 4, 2026: https://www.sec.gov/Archives/edgar/data/1039399/000103939926000033/form-20260627.htm
4. FormFactor, Q2 2026 earnings Form 8-K, filed July 29, 2026: https://www.sec.gov/Archives/edgar/data/1039399/000103939926000030/form-20260729.htm
5. FormFactor, Q2 2026 earnings Exhibit 99.1, dated July 29, 2026: https://www.sec.gov/Archives/edgar/data/1039399/000103939926000030/ex9901-earningsreleasexq226.htm
6. FormFactor, Q2 supplemental financial information, dated July 29, 2026: https://investors.formfactor.com/static-files/855b6a03-72f0-46b0-a64e-e29aebf2651c
7. FormFactor, 2026 proxy statement, filed April 2, 2026: https://www.sec.gov/Archives/edgar/data/1039399/000103939926000016/form-20260401.htm
8. FormFactor, 2026 Form S-8, filed August 4, 2026: https://www.sec.gov/Archives/edgar/data/1039399/000103939926000035/formfactor-sx8fy2026.htm
9. FormFactor, January 2026 restructuring Form 8-K: https://www.sec.gov/Archives/edgar/data/1039399/000103939926000002/form-20260105.htm
10. SEC submissions history, CIK 0001039399: https://data.sec.gov/submissions/CIK0001039399.json

Historical filings used for trend reconciliation

1. FY2021 Form 10-K: https://www.sec.gov/Archives/edgar/data/1039399/000103939922000014/form-20211225.htm
2. FY2022 Form 10-K: https://www.sec.gov/Archives/edgar/data/1039399/000103939923000010/form-20221231.htm
3. FY2023 Form 10-K: https://www.sec.gov/Archives/edgar/data/1039399/000103939924000007/form-20231230.htm
4. FY2024 Form 10-K: https://www.sec.gov/Archives/edgar/data/1039399/000103939925000023/form-20241228.htm
5. FRT sale closing Form 8-K: https://www.sec.gov/Archives/edgar/data/1039399/000103939923000043/form-20231101.htm
6. 2025 credit-facility Form 8-K: https://www.sec.gov/Archives/edgar/data/1039399/000103939925000077/form-20250729.htm

Ownership and governance

1. Mike Slessor Form 4, filed July 16, 2026: https://www.sec.gov/Archives/edgar/data/1039399/000162828026048529/wk-form4_1784237440.xml
2. Mike Slessor Form 4, filed August 17, 2026: https://www.sec.gov/Archives/edgar/data/1039399/000162828026057485/wk-form4_1787004992.xml
3. FormFactor 2025 proxy equity-plan proposal: https://www.sec.gov/Archives/edgar/data/1039399/000114036125011850/ny20036714x1_def14a.htm
4. FormFactor 2022 proxy equity-plan proposal: https://www.sec.gov/Archives/edgar/data/1039399/000114036122014333/ny20001965x2_def14a.htm
5. BlackRock 2026 Schedule 13G amendment: https://www.sec.gov/Archives/edgar/data/1039399/000201238326002572/primary_doc.xml
6. Earnest Partners 2026 Schedule 13G amendment: https://www.sec.gov/Archives/edgar/data/1039399/000110257826000046/primary_doc.xml

Company strategy, products, and management commentary

1. FormFactor Investor Day 2026, May 11, 2026: https://investors.formfactor.com/static-files/29f1154b-e1e4-4b57-afbd-f296b3f36504
2. SmartMatrix product page: https://www.formfactor.com/product/probe-cards/dram/smartmatrix/
3. Memory-test application page: https://www.formfactor.com/applications/high-volume-test-on-wafer/memory-test/
4. Triton product discussion, March 20, 2026: https://www.formfactor.com/blog/2026/triton-scaling-silicon-photonics-wafer-test-for-high-volume-manufacturing/
5. FormFactor–Keystone Microtech partnership, July 28, 2026: https://investors.formfactor.com/news-releases/news-release-details/formfactor-and-keystone-microtech-announce-strategic-partnership
6. Q2 2026 earnings call, dated July 29, 2026, obtained through ROIC.ai’s direct latest-call endpoint for NASDAQ:FORM. All material filed figures were cross-checked to the 10-Q and earnings exhibit; forward mix, margin decomposition, capacity timing, and customer-share claims remain management statements.

Competitors and adjacent test ecosystem

1. Technoprobe H1 2026 results presentation, August 5, 2026: https://www.technoprobe.com/wp-content/uploads/2026/08/H1-2026-Results.pdf
2. Technoprobe H1 2026 financial report: https://www.technoprobe.com/wp-content/uploads/2026/08/Technoprobe-SpA-Half-Year-Financial-Report-as-of-30th-June-2026.pdf
3. Micronics Japan Q2 2026 supplemental results, August 12, 2026: https://ssl4.eir-parts.net/doc/6871/tdnet/2871330/00.pdf
4. Micronics Japan FV26 strategy: https://www.mjc.co.jp/en/ir/strategy/fv26/
5. Teradyne Q2 2026 Form 10-Q: https://investors.teradyne.com/sec-filings/all-sec-filings/content/0001193125-26-327715/ter-20260628.htm
6. Cohu Q2 2026 Form 10-Q: https://www.sec.gov/Archives/edgar/data/21535/000143774926025144/cohu20260627_10q.htm

Peer valuation bridges also use August 28, 2026 market prices and latest filing/ROIC.ai statement data for FORM, TPRO, TER, ONTO, CAMT, COHU, and KLAC. Foreign-currency and non-GAAP definitions differ, so the peer table is directional rather than a precision ranking.

Customers, packaging, and regulation

1. TSMC HPC and 3DFabric platform: https://www.tsmc.com/english/dedicatedFoundry/technology/platform_HPC_tech_WLSI
2. TSMC Q2 2026 results: https://investor.tsmc.com/english/quarterly-results/2026/q2
3. Samsung HBM4 production announcement: https://semiconductor.samsung.com/news-events/news/samsung-ships-industry-first-commercial-hbm4-with-ultimate-performance-for-ai-computing/
4. U.S. Bureau of Industry and Security news and updates: https://www.bis.gov/news-updates
5. January 13, 2026 BIS licensing-policy release: https://media.bis.gov/sites/default/files/documents/DoC Revises License Review Policy for Semiconductors Exports.pdf

Market, factor, and third-party data

1. AZI adjusted daily price history for FORM through August 28, 2026: https://azitrading.com/controls/download-data.php?t=FORM
2. AZI valuation index dated August 28, 2026: composite 87.73; P/E 73.41; P/B 95.17; P/S 94.61. Used only for the stock’s own-history context.
3. FactorsToday API, model data through August 28–30, 2026: https://www.factorstoday.com/api. Used for loadings, specific volatility, risk-adjusted history, and related-stock context; these are statistical estimates, not causal or fundamental evidence.
4. ROIC.ai NASDAQ:FORM profile and financial endpoints. Used as a third-party accelerator and cross-check; filed statements control. The Q2 provider operating-income field excluded factory start-up expense, while the EBIT field reconciled to the filing.
5. Kiplinger, “Nasdaq Sinks as Sandisk Sell-Off Continues,” July 2, 2026, for broad semiconductor-sector context.
6. Kiplinger, “Stocks Turn Down as Warsh Talks Up Rates,” August 28, 2026, for the late-August rates/sector context.