Cognex Corporation (CGNX)

The undisputed leader in industrial machine vision — "the robot's eyes" — seven quarters into a margin turnaround and freshly re-platformed on edge AI.

Investment Research · Photoncap-style deep dive · v1 of "Cognex" · May 14, 2026

What Cognex physically does

Cognex makes the systems that let a machine see and make a decision about what it sees. A machine-vision system is, at the hardware level, an industrial camera with a lens, controlled lighting, an image sensor and an onboard processor — but the value is in the software stack that runs on it. Cognex's software takes the raw image and answers a question a factory or warehouse cares about: is this part defective, what code is on this box, where exactly is this object so a robot can pick it. The company's product DNA is two software families — VisionPro (the PC-based vision software platform) and In-Sight (self-contained "smart cameras" that run vision software on the camera itself) — plus the ID products (barcode and code-reading, the DataMan family) and the deep-learning suite (historically ViDi, now folded into the broader AI vision offering).

Why this is the binding constraint for robotics: a robot cannot manipulate what it cannot locate. Vision-guided robotics — where a camera tells the robot arm where the object is, in three dimensions, in real time — is the function that turns a blind, pre-programmed arm into something that can handle variation. Cognex sits at exactly that chokepoint. Its machine-vision systems are the perception layer between the camera and the robot controller, and in a world moving toward general-purpose robots that operate in unstructured environments, the vision problem gets harder and more valuable, not easier.

The structural shift inside Cognex's own technology is the move to edge AI. Traditional machine vision was rule-based — an engineer programmed the system to look for specific features. The new generation runs neural-network inference on the camera itself, so the system can be trained on examples rather than hand-programmed, and can handle the messy variation that rule-based vision could not. Cognex's Q1 2026 product launches — the In-Sight 6900 powered by NVIDIA and the In-Sight 3900 powered by Qualcomm — are the company re-platforming its entire smart-camera line onto AI silicon. This is the bet that the next decade of factory and warehouse vision is AI-native, and that Cognex's installed base and software moat carry it across that transition.

Product roadmap

Cognex's roadmap in 2025–2026 is dominated by the edge-AI re-platforming. The In-Sight smart-camera family — historically the In-Sight 8000 series (2D smart cameras) and the higher-end vision systems — is being rebuilt around AI accelerators. The In-Sight 8900 brought embedded AI to OEM deployments. The flagship 2026 launches came in early May: the In-Sight 6900, an AI vision system powered by NVIDIA silicon, and the In-Sight 3900, an embedded AI vision system launched May 5, 2026 on Qualcomm's Dragonwing platform — Cognex says the 3900 delivers up to 4x faster processing than prior Cognex generations, supports up to 25-megapixel imaging, and runs PC-free for real-time edge inspection. On the ID/code-reading side, the DataMan 290 (recent) added AI-enabled auto-setup and advanced code filtering.

The cadence shows a company pushing AI down its entire price stack — from high-end NVIDIA-powered systems for demanding applications to Qualcomm-powered embedded units for cost-sensitive OEM volume. At its June 2025 Investor Day and a March 2026 conference appearance, management laid out the strategic frame: AI-driven growth, deliberate exits from low-margin legacy product lines, and a raised EBITDA-margin target of 25–31%. What Cognex does not make is the robot, the actuator, or the end-to-end robot software — it is the perception layer, sold to robot integrators, OEMs and end users. It also is not a 3D-vision-only or a camera-component-only player; its differentiation is the full hardware-plus-software-plus-AI stack, sold across 30-plus industries.

The financial print

Cognex closed FY2025 (year ended December 31, 2025) with revenue of $994 million, up 9% year-over-year (up 8% constant-currency, or 7% excluding a one-time medical-lab channel-partner benefit), and a full-year operating margin of 16.3% — results released February 11, 2026. Adjusted EBITDA margin was 21.5%, up 440 basis points year-over-year, clearing the company's 20% milestone ahead of plan. The story is the margin trajectory: Cognex spent 2022–2024 over-earning then under-earning through the post-COVID factory-automation cycle, and FY2025 was the year the cost discipline and mix improvement showed up.

The Q1 2026 print on May 6, 2026 accelerated it. Revenue was $268.4 million, up 24.3% year-over-year and ahead of expectations, with double-digit growth across logistics, packaging, electronics and semiconductor. Operating margin was 22.3% and adjusted EBITDA margin reached 26.9%, up 1,010 basis points year-over-year — the seventh consecutive quarter of margin improvement. Adjusted diluted EPS was $0.34, up 113% year-over-year, the seventh straight quarter of EPS growth. Management guided Q2 2026 to revenue of $280–300 million (midpoint ~16.5% growth) and adjusted EPS of $0.40–0.44 (midpoint implying ~68% growth). At $64.61 the stock trades at a forward P/E of 37.2 with a market cap of $10.75B — a premium multiple, but on a company that is compounding both revenue and margin, which is a different proposition than a flat-revenue name on a high multiple. The next binary is Q2 2026 earnings, expected around July 30, 2026.

Customer mix today

Cognex breaks its revenue out by end market, and the FY2025 mix is well-disclosed. Logistics was roughly 26% of full-year 2025 revenue and delivered its ninth consecutive quarter of double-digit growth — this is the warehouse-automation engine, driven by e-commerce parcel sortation and the structural build-out of automated distribution. Packaging was roughly 21% of 2025 revenue and posted double-digit Q1 2026 growth, strong enough that management raised its 2026 packaging outlook to high-single-digit. Consumer electronics was roughly 19% of revenue, with double-digit Q1 growth and a high-single-digit to double-digit full-year 2026 outlook — this is the historically lumpy Apple-supply-chain exposure that whipsaws on smartphone build cycles. Automotive sat in the low teens of revenue and was the one soft spot, down high single digits in 2025, though management said on the Q4 2025 call that the market has bottomed and guided flat-to-low-single-digit growth for 2026.

The 2024-versus-2026 change worth highlighting is the diversification away from automotive and consumer-electronics lumpiness toward the more durable logistics and packaging demand, plus the layering-on of the AI-vision product cycle. Cognex does not have the single-customer concentration of a supplier — its largest customers are integrators and end users spread across 30-plus industries — which is a structural strength versus several names in this batch. The robotics/humanoid-specific exposure is not yet broken out as its own line; it is embedded across logistics (warehouse robots) and the broader vision-guided-robotics demand, and it is best understood as an option on the theme rather than a quantified revenue stream today.

What's actually happening in logistics and AI vision

The two mechanisms driving Cognex right now are the logistics flywheel and the AI-vision product transition. On logistics: nine consecutive quarters of double-digit growth is not a cycle, it is a structural build — every automated distribution center needs barcode reading at every sortation point and increasingly needs vision-guided picking and induction. Cognex's installed base in logistics compounds because once a DC standardizes on Cognex code-reading, the expansion revenue is sticky. Management was deliberately conservative on the 2026 logistics outlook (mid-to-high single digit) despite the Q1 strength, which reads as sandbagging given the run-rate.

On AI vision: the In-Sight 6900 (NVIDIA) and In-Sight 3900 (Qualcomm, launched May 5, 2026) are the proof points of the re-platforming. The strategic logic is that AI-native vision expands Cognex's addressable market — applications that were impossible or uneconomic with rule-based vision (handling high variation, training on examples rather than code) become solvable, and Cognex can sell into them. The seven consecutive quarters of margin expansion suggest the transition is being executed without the margin destruction that usually accompanies a platform shift, because Cognex is simultaneously exiting low-margin legacy lines (the "portfolio exits" management flagged in March 2026). The skeptical read: AI vision is also the door through which competition and commoditization walk in, which is the terminal-risk section. The bull read: Cognex's software moat and installed base are exactly what carry an incumbent across a platform transition, and the early evidence — 24% revenue growth, 27% EBITDA margin in Q1 — says it is working so far.

The competitive threat / Keyence

Cognex's primary named competitor is Keyence (6861, also covered in this batch). Per coherent market-share estimates, Cognex holds roughly 21% of the global machine-vision market and Keyence roughly 19% — together close to half the market, with the remaining leading-five players controlling another slice and a long tail of Teledyne (DALSA), Basler, Sick, Omron, Datalogic, Canon and others. Keyence is the more formidable competitor: it is structurally more profitable (operating margins above 50% versus Cognex's mid-teens-to-low-20s), it dominates the Asia-Pacific market, and its direct-sales model gives it deep application-engineering reach. Where Cognex is differentiated is depth of vision software and the AI/deep-learning stack — VisionPro and the AI suite are genuinely best-in-class — and a stronger position in logistics and code-reading specifically.

There is no active IP litigation to flag between the two. The competitive dynamic is a share contest, not a courtroom one: both companies are racing to AI-native vision, and the question is whether Cognex's software lead or Keyence's profitability and sales reach wins the next product cycle. The honest assessment is that this is a structural duopoly-plus where both can grow with the market — machine vision is a ~$12B market in 2026 growing mid-to-high single digits — but Keyence's margin structure means it can out-invest and out-price Cognex if it chooses to. Cognex's defense is that it is the vision-software specialist while Keyence is a broad-line sensor company for which vision is one of many products; in a robotics world where vision software is the hard part, that specialization should matter.

The terminal risk

The terminal risk for Cognex is that general-purpose AI eats dedicated machine vision. The transition technology is the vision-language model — large multimodal AI models that can look at an image from a commodity camera and answer perception questions without Cognex's specialized hardware or software. If a warehouse can run a VLM on a cheap camera and a standard GPU and get "good enough" defect detection, code reading and object localization, the value of a dedicated Cognex system erodes. This is the SaaS-displaced-by-LLM analogy applied to industrial vision: the moat was the specialized software, and general-purpose AI is a solvent for specialized software.

The timing is genuinely uncertain. The skeptic case says this is a 3–5 year threat as VLMs get faster, cheaper and reliable enough for industrial uptime requirements. The bull rebuttal — and it is a real one — is that industrial vision has requirements that general-purpose AI does not naturally meet: deterministic latency, 99.9%+ reliability, traceability for regulated industries, and integration with PLCs and robot controllers. Cognex's response is to be the company that productizes AI for industrial requirements rather than the company that gets displaced by it — the In-Sight 6900/3900 launches are precisely that bet, putting NVIDIA and Qualcomm AI silicon inside ruggedized, deterministic, factory-grade hardware. Named alternative beneficiaries of the displacement scenario would be NVIDIA's own robotics/vision stack, hyperscaler vision APIs, and AI-native vision startups. The constraint this puts on the multiple: 37x forward earnings already prices Cognex as a durable AI winner, so the terminal risk is less "the stock goes to zero" and more "the multiple compresses if the AI-native transition turns out to commoditize the category."

Bull / Gap / Optionality (Photoncap framing)

1. Seven straight quarters of margin expansion, and accelerating. Q1 2026 adjusted EBITDA margin of 26.9% was up 1,010 basis points year-over-year (company release, May 6, 2026), and management has raised the long-term EBITDA target to 25–31%. This is a structural re-rating of the earnings power, not a cyclical bounce — the cost discipline and portfolio exits flagged in March 2026 are durable.

2. The logistics flywheel is a structural, not cyclical, demand engine. Nine consecutive quarters of double-digit logistics growth (Q1 2026 call), with management still guiding conservatively to mid-to-high single digit for 2026 — the sandbag is visible. Logistics at ~26% of revenue is the most durable leg of the mix and compounds on installed-base expansion.

3. The AI-vision re-platforming expands the addressable market. The In-Sight 6900 (NVIDIA) and In-Sight 3900 (Qualcomm, launched May 5, 2026) make previously-uneconomic applications solvable. Cognex is putting AI silicon across its full price stack while exiting low-margin legacy lines — the cleanest version of an incumbent navigating a platform shift.

4. Best balance-sheet and customer-concentration profile in the batch. Cognex sells across 30-plus industries with no single-supplier-style customer concentration, carries a strong net-cash position, and is the global market-share leader. In a batch that includes a thin-margin turnaround (VPG) and an auto-heavy cyclical (Sensata), Cognex is the quality name.

5. Robotics is free optionality on top of a working base case. The base case — 24% Q1 revenue growth, margin expansion, logistics durability — does not require the humanoid theme to work. Vision-guided robotics and humanoid perception are an unpriced call option layered on a company that is already compounding.

Gap

1. The multiple already prices a durable AI winner. At 37.2x forward earnings, Cognex is not cheap on any absolute measure — the market is paying for the margin trajectory and the AI transition both succeeding. RSI 68.2 and +19.8% above the 50-day average is extended (though far less so than VPG); a growth disappointment re-rates this meaningfully.

2. Consumer electronics and automotive are still lumpy. Roughly 19% of revenue is consumer electronics tied to smartphone build cycles, and automotive (low-teens of revenue) was down high single digits in 2025. Both are guided to recover in 2026, but a slip in either would expose how much of the Q1 strength was end-market timing.

3. The VLM commoditization threat is real and Cognex is partly betting against itself. By moving to AI-native vision, Cognex is making its own category more accessible to general-purpose AI competitors. The In-Sight 6900/3900 strategy is the right one, but it is a bet that industrial requirements protect the moat — and that bet is not yet proven over a full cycle.

4. Keyence is a structurally stronger competitor on margin and reach. Keyence runs 50%+ operating margins versus Cognex's low-20s and has the financial capacity to out-invest or out-price Cognex in any segment it prioritizes. Cognex's software lead is real but it competes against a peer that is simply more profitable per dollar of revenue.

Optionality

The trade

Cognex is the highest-quality name in this sensing/vision batch and the one whose base case does not depend on the humanoid theme working at all. The structural thesis is straightforward: the dominant machine-vision company, seven quarters into a real margin turnaround, re-platforming onto edge AI in a way that expands its market, with the best balance sheet and the least customer concentration in the batch — and robotics/humanoid perception as free optionality on top. The chart is moderately extended (RSI 68.2, +19.8% versus the 50-day) but nowhere near the VPG-style danger zone, so this is a name you can actually initiate rather than only watch. The trade is to initiate in a zone of current ±5%, roughly $61.38–$67.84, accumulating on any pullback toward the lower end; size at 1.0–1.5% of risk capital — a real position, justified by the quality and the working base case, but tempered by the 37x forward multiple that leaves no room for a growth stumble. Stop at roughly $54, below the structural base and the rising 50-day cloud; a break there would say the margin-expansion narrative has cracked. The named catalyst is Q2 2026 earnings around July 30, 2026, where the $280–300M revenue and $0.40–0.44 EPS guide gets tested and the AI-vision product traction starts to show. There is no cleaner expression of the machine-vision thesis — Cognex is the cleaner expression; the only purer-vision alternative, Keyence, comes with a much higher absolute multiple and Japan-listing friction. Conviction: 7 / 10.

Sources referenced inline throughout. Reference v1 of this template format: _Watchlist/hanmi-photoncap-style.md.

\newpage

6861 — Keyence Corporation / キーエンス · WATCH (Tier-2) · Conv 6/10 · Bucket B

ticker: 6861 name: Keyence Corporation / キーエンス theme: Robotics bucket: B conviction: 6 entryzonelo: 73444 entryzonehi: 81176 currentprice: 77310 pricedate: 2026-05-14 positionsizepct: 1.0 stoploss: 64000 thesisoneline: The highest-margin sensor and machine-vision company on earth — vision-guided robotics is one of many growth legs, and the quality is the moat. catalystnext: Q1 FY2026 (June quarter) earnings catalystdate: 2026-08-03 deepdivepath: Theme -- Robotics/6861/6861-deep-dive.md lastupdated: 2026-05-14T00:00:00Z rsi: 63.9 vs50ma: 21.0 forwardpe: 41.4 themecycleposition: early customermixsummary: Highly diversified — no single customer or end market dominant; factory automation, electronics, autos, pharma, logistics, semiconductors spread globally. terminalriskoneline: General-purpose AI vision and low-cost Asian sensor makers compress the pricing power that underwrites Keyence's 50%+ operating margin. bulldriverscount: 5 gapriskscount: 4 optionalitycount: 5 lastearningsdate: 2026-04-27 nextearningsdate: 2026-08-03