Financebotresearch desk研究台

Discovery › screen

Screen robotics automation machine vision physical ai

Date: 2026-07-29 · Agent: Scout / Screener · Framework: analysis_notes.md §0, §1, §2, §3, §4

Mandate: Find emerging-technology exposure at the layer where AI meets the physical world — explicitly not another datacenter-capex bet. The portfolio is already saturated with AI infrastructure (AVGO, ANET, INTC, QCOM, ORCL, GOOGL, MSFT, META + power/nuclear).

Exclusions honored: ISRG, OUST, PTC, U, AMBA, ANET, AVGO, INTC, QCOM, AMD, NVTS, ADI, NXPI, MCHP, EMR, ETN, AME, HUBB, ITRI, ROCK, CODA, the entire Microstocks sleeve, and the previously-screened semi/infra names.

0. Headline — read this first

The field's cycle has already turned, and the market got there first.

This screen was commissioned on the hypothesis that industrial automation might still be sitting in a 2024–26 capex trough while the rest of industrials re-rated. That hypothesis is wrong as of mid-2026. The trough ended somewhere in H2 2025 and the recovery is now visible in hard order data:

Consequence for §0: the highest-quality pure-plays in this field are now in the "Great company / Expensive price" quadrant, not the "Great / Cheap" one. Rockwell sits at 32x forward within 5% of its 52-week high. Cognex is at ~35x forward after a 79% run off its low. Teradyne is at 35x forward. Buying the automation recovery today means paying for it.

The two genuinely mispriced things I found are not the pure-play robot names. They are (a) a warehouse-automation/machine-vision compounder the market still prices as a fading barcode-scanner company, and (b) a motion-control roll-up the market still prices as a levered 2023 M&A accident. Both are real §0 candidates. Neither is glamorous. That is the point.

On humanoids: reject the entire category. Global 2025 shipments were 16,000–18,000 units (+508%), but fewer than 3,000–4,000 are in actual productive commercial deployment — the rest are R&D units and internal training fleets. ~90% of volume is Chinese. Tesla's external Optimus deliveries are targeted "late 2026." A Chinese survey found only 23% of humanoid buyers satisfied across 150 competing vendors. (Axis Intelligence, TNW, Interact Analysis) There is no investable US-listed humanoid pure-play with revenue. There is only narrative. §6's 90%-failure prior applies at full strength.

1. Candidate funnel

18 names considered, 14 pulled through fin.py. Quotes as of 2026-07-29.

Ticker Sub-field Mkt cap Price P/E ttm P/E fwd EV/EBITDA % off 52wH Verdict
ZBRA Warehouse automation, machine vision, AMRs $13.8B 288.94 34.9 14.0 15.8 −18.1% ✅ #1
RRX Motion control / actuation $13.5B 202.14 47.0 15.0 14.1 −18.4% ✅ #2
GMED Surgical robotics + spine $10.9B 80.33 18.8 15.6 11.2 −20.8% ✅ #3
CGNX Machine vision (pure-play) $10.3B 62.02 72.9 34.8 41.5 −14.9% ⏸ Great co, wrong price
ROK Industrial control (pure-play) $52.7B 473.34 49.1 32.4 27.8 −4.8% ⏸ Fully re-rated
TDY Sensing / imaging $29.8B 643.35 31.2 24.1 19.4 −7.2% ⏸ Fair, not cheap
TER Robotics (UR/MiR) + semi test $56.5B 360.86 49.6 34.9 33.1 −26.0% ❌ Datacenter proxy
MKSI Photonics / semi-cap $19.6B 290.69 60.7 18.9 22.5 −35.1% ❌ Datacenter proxy
NOVT Precision motion, medical robotics $5.0B 139.10 100.1 33.6 26.9 −19.1% ❌ Deteriorating
IPGP Industrial fiber lasers $3.6B 83.78 123.2 36.3 28.4 −46.2% ❌ Value trap
SYM Warehouse automation $25.8B 42.72 n/a 59.2 154.2 −51.4% ❌ Speculative
HSAI Lidar ~$17–21B* 16.59 34.6 18.5 n/a −46.2% ❌ China ADR
GXO Contract logistics $5.8B 50.08 44.7 14.2 11.9 −25.1% ❌ Not automation
PRCT Surgical robotics (micro) $1.0B 17.84 n/a −19.9 −8.3 −65.3% ❌ §6 territory

* HSAI market-cap and EV figures returned inconsistently by the data source (EV printed as −$2.35B). Data gap — do not rely on HSAI numbers here.

2. Top 3

🥇 #1 — ZBRA · Zebra Technologies

Sub-field: Warehouse & logistics automation, machine vision, autonomous mobile robots Market cap $13.76B · Price $288.94 · 18.1% off 52wk high ($352.66) · 52wk low $199.05

Why it's a gem

The market still prices Zebra as a cyclical barcode-scanner and rugged-handheld company that had a bad 2023. What it has actually become — through the Matrox Imaging, Fetch Robotics, Adaptive Vision and Photoneo acquisitions — is one of the two credible Western machine-vision platforms and a warehouse-robotics vendor, sold into an installed base of enterprise logistics customers no competitor has access to. At ~15.6x management's guided 2026 adjusted EPS of $18.50, it trades at roughly half its own 3-year average multiple (~39.9x) and 12% below its 10-year median trailing P/E (31.4 vs 27.7) (GuruFocus, FullRatio). That gap exists because the tariff story dominated the 2025 narrative — and management has now guided to full tariff mitigation by Q2 FY2026.

Metrics (§1)

Metric Value Note
P/E ttm / fwd 34.9 / 14.0 (GAAP fwd); ~15.6x on adj $18.50 vs 3yr avg 39.9x, 5yr 37.2x, 10yr median 31.4x
EV / EBITDA 15.80 down from 17.2x in Q4'25
FCF (2025) $831M FCF yield 6.0% on cap, 5.1% on EV
FCF 3yr CAGR +26.2% 2023 was −$91M (destocking trough) → clean inflection
OCF 3yr CAGR +23.4%
Revenue 3yr CAGR −2.3% ⚠️ trough-distorted: $5.78B (2022) → $4.58B (2023) → $5.40B (2025)
Rev growth YoY +14%; Q2'26 guided +14–17% recovery is real and accelerating
Net income 3yr CAGR −3.3% 2025 GAAP hit by acquisition/amort
Debt / Assets 31.7% $2.70B debt, $125M cash — ⚠️ cash drained by the $1.36B Elo acquisition
Current ratio 0.96 ⚠️ tight; the one real balance-sheet flag
Shares (dil) 3yr CAGR −0.9% $587M bought back in 2025
ROE / ROA 12% / 6%
Gross margin 48% stable-to-rising; was 45% at the 2023 trough
Rev/share · FCF/share $105.37 · $16.23 price/FCF/share = 17.8x
Graham IV $115.52 Not meaningful — asset-light tech with $4.73B goodwill

Where in the cycle

Early-mid recovery, roughly 12–18 months in. 2023 was the destocking bottom (negative FCF, revenue down 21%). 2024–25 rebuilt. 2026 is the first year of clean double-digit growth with margin expansion. This is the "cyclically depressed earnings normalizing" case, not the trap — the tell is that FCF recovered before revenue did, which is what a real inventory correction looks like, not structural impairment.

China exposure

Meaningful on the supply side, modest on the demand side. Nearly 50% of finished-goods production has been moved out of China; the component supply chain remains largely China-based. This is a tariff/COGS risk, not a competitive-share risk — Zebra does not lose enterprise logistics deals to Inovance or Estun. Management targets full tariff-cost mitigation by Q2 FY2026 (Q3 2025 call; Supply Chain Dive).

What they actually sell today — with revenue attached

$5.40B FY2025. Concrete, shipping products, no TAM slides: - Asset Intelligence & Tracking — barcode/RFID printers, labels, supplies. Consumable, recurring. - Enterprise Visibility & Mobility — rugged mobile computers and scanners; the core installed base at retailers, warehouses, hospitals, parcel carriers. - Machine vision — Matrox Imaging + Adaptive Vision + the fixed industrial scanning line; the CV70 ultra-compact high-speed camera launched at Automate 2026 (Intellectia). - 3D vision-guided robotics (Photoneo) — bin picking, box depalletizing, picking from custom dunnage, 3D volume measurement. Shipping today (Yahoo Finance / Automate 2026). - AMRs (Fetch) — warehouse mobile robots.

⚠️ Data gap: Zebra does not break out machine-vision/robotics revenue separately. I cannot size it. Treat the vision business as an optionality kicker on a profitable core, not as the thesis.

Moat — switching costs + installed base/aftermarket

Named explicitly, and both are the durable kind this field rewards. A Zebra scanner fleet is wired into a retailer's or 3PL's WMS through years of integration work; the printers pull recurring label/supply consumables; and a vision system qualified into a production line is genuinely painful to swap. Secondary: efficient scale in enterprise device management (Zebra's DNA/Savanna software layer), which no point-solution vision vendor can match. Gross margin holding 48% through a violent destocking cycle is the quantitative confirmation (§2.1) — the moat did not leak when volumes collapsed.

Bear case (adversarial)

  1. This is a device-refresh cycle dressed up as automation. The bulk of revenue is still handhelds and printers on a ~5-year enterprise replacement clock. If 2026 is a pull-forward, 2027–28 is an air pocket, and the "cheap" 15.6x is on peak-cycle EPS. This is the single strongest bear argument and I cannot fully refute it.
  2. Balance sheet has no slack. $125M cash against $2.70B debt and a 0.96 current ratio after the Elo deal. A demand stumble is not fatal but removes buyback flexibility.
  3. Goodwill is 56% of assets ($4.73B of $8.50B). The acquisition roll-up strategy has not yet proven it earns its cost of capital — ROE of 12% on a 48% gross margin business is unimpressive.
  4. Chinese competition: low direct threat in enterprise logistics, but real in commodity scanning hardware, and the component supply chain is a tariff hostage.
  5. Smartphone substitution — the perennial bear thesis on rugged handhelds. It has been wrong for 15 years, but it never fully dies.

Diversification check

✅ Genuine diversifier. Revenue is driven by retail/e-commerce/3PL/healthcare capex and enterprise device refresh — not hyperscaler spend. There is no datacenter line item. Correlation to the existing AVGO/ANET/ORCL cluster runs through general risk appetite, not through shared end demand.

Valuation & entry (§3)

  • Bogle: 0% yield + ~10–12% earnings growth + modest multiple recovery → low-to-mid teens expected return.
  • Multiple-based: $18.50 adj EPS 2026, ~$21 in 2027. At 18–20x (still below its 10-yr median) → fair value $333–$370.
  • FCF-based: $831M FCF, ~5% required yield on EV → ~$14.4B equity → ~$302/sh. Conservative anchor.
  • Graham IV of $115 is not applicable — ignore it for a goodwill-heavy asset-light business.
  • DYT/DDM: N/A, no dividend.

Fair value range: $310–$370. Entry zone: below $285. Conviction: 7/10.

🥈 #2 — RRX · Regal Rexnord

Sub-field: Motion control, actuation, precision drives — the actuation layer of physical AI Market cap $13.45B · Price $202.14 · 18.4% off 52wk high ($247.80) · 52wk low $127.96

Why it's a gem

Regal Rexnord is still priced as the wreckage of its 2023 Altra acquisition — a $4.87B deal that pushed debt to $6.55B and produced a −$57M net loss year. That story is now two years stale. Debt is down $1.61B in 24 months to $4.94B, FCF has compounded at 36.3% over three years to $893M, and the Automation & Motion Control segment just printed orders +34% YoY (+28% excluding datacenter). At 15.0x forward with a 6.6% FCF yield, the market is paying for the deleveraging but not for the automation franchise.

Metrics (§1)

Metric Value Note
P/E ttm / fwd 47.0 / 15.0 ttm distorted by amortization; fwd is the honest number
EV / EBITDA 14.13 reasonable for a leveraged industrial mid-recovery
FCF (2025) $893.1M FCF yield 6.6% on cap, 5.3% on EV
FCF 3yr CAGR +36.3% $352M → $893M
OCF 3yr CAGR +31.5%
Revenue 3yr CAGR +4.4% inorganic (Altra); organic was negative until 2026
Rev growth YoY +4%; 2026 guide raised to ~$6.20B (+4.5%)
Net income 3yr CAGR −17.0% −$57.4M (2023) → $279.5M (2025); the trough is behind
Debt / Assets 35.5% $4.94B, down from $6.55B in 2023
Interest expense $349.2M (2025) vs $431.0M (2023) −19%, direct EPS tailwind
Current ratio 2.17 comfortable
Shares (dil) 3yr CAGR −0.2% no dilution; buybacks paused for deleveraging
ROE / ROA 4% / 3% ⚠️ the weakest number in this report
Gross margin 38%, rising (32% in 2022 → 38%) 600bp expansion = mix shift toward AMC working
Dividend 0.75% yield, 33% payout 5yr avg yield 0.93% → DYT says not cheap on yield
Rev/share · FCF/share $89.11 · $13.41 price/FCF/share = 15.1x
Graham IV $99.43 vs $202 — flags the goodwill problem, correctly

Where in the cycle

Early recovery, and this is the clearest cycle signal in the whole screen. AMC orders +34% YoY with 28 points of that excluding datacenter is a genuine broad discrete-automation inflection — management called out discrete automation, aerospace & defense and medical alongside datacenter (Q1 2026 release; transcript). Orders lead revenue by 2–4 quarters in this business, so 2026 revenue guidance likely understates 2027.

China exposure

⚠️ Data gap — RRX does not cleanly disclose China revenue %, and I will not invent a number. Directional read: RRX's industrial powertrain and motion businesses are more North America/Europe weighted than a Rockwell or a Cognex, and the AMC growth cited is US-led (datacenter, A&D, medical). Chinese motion-control competitors (Inovance especially) compete hard in commodity servos and VFDs — a real long-run margin threat in the low end, less so in the engineered/aerospace/medical tail. Verify this in a full /analyze before sizing.

What they actually sell today — with revenue attached

$5.93B FY2025 across three segments — all physical, all shipping, zero narrative: - Automation & Motion Control (AMC) — precision servo motors, linear actuators, conveying, gearing, encoders. The physical-AI-relevant segment. Organic sales +12.1% in Q1'26. - Industrial Powertrain Solutions — couplings, bearings, gearboxes, clutches (the Altra assets). Heavy aftermarket content. - Power Efficiency Solutions — commercial/residential motors. The commodity, lower-margin ballast.

Moat — installed base / aftermarket + switching costs

Named explicitly. The Industrial Powertrain business is a classic aftermarket annuity: couplings and bearings specified into installed machinery get replaced from the original supplier for decades. In AMC, precision motion components are designed into OEM machine platforms — swapping a servo means requalifying the machine. This is the same switching-cost mechanic as machine vision, just at the actuation layer. Weaker than Zebra's, though: many of these components are more substitutable, and the 4% ROE says the moat is not yet earning its capital.

Bear case (adversarial)

  1. ROE of 4% and $6.61B goodwill against $6.84B of equity — tangible book is effectively zero. This is the crux. Regal paid up for Altra and has not yet demonstrated the combined entity earns above its cost of capital. If it never does, 15x forward is fair, not cheap, and there is no re-rating.
  2. The order surge is partly the bet the user is trying to avoid. 34% headline, 28% ex-datacenter — so ~6 points of AMC order growth is datacenter. Not disqualifying, but it means part of the good news here is correlated with existing holdings.
  3. Tariff margin pressure is live and explicitly flagged in Q1'26; management claims a path to neutrality but has not yet delivered it.
  4. Chinese competition in commodity motors and servos is structural and will keep pressure on Power Efficiency Solutions indefinitely. That segment may be permanently impaired.
  5. Deleveraging is a finite catalyst. Once net leverage normalizes, the story needs organic growth to carry it — and organic growth has been the weak link for three years.

Diversification check

✅ Mostly a genuine diversifier — with an asterisk. ~80%+ of the business is industrial/commercial/aerospace/medical end markets with no hyperscaler linkage. But management is actively marketing a datacenter angle (Yahoo Finance), and ~6pts of the order growth is datacenter. Call it 85% diversifier / 15% repackaged AI-capex.

Valuation & entry (§3)

  • Forward EPS implied ~$13.51 (2026). At 16–18x → $216–$243.
  • FCF: $893M at a 6% required yield → ~$14.9B → ~$224/sh.
  • Graham IV $99.43 — far below price; correctly penalizes the goodwill. Do not dismiss this; it is the quantitative form of the bear case.
  • DYT: current 0.75% yield vs 0.93% 5yr average → yield says overvalued, not cheap. Honest tension, surfaced not buried.
  • Analyst mean target $262.60 (strong_buy consensus) is above my range. I am deliberately more conservative.

Fair value range: $215–$245. Entry zone: below $190. Conviction: 6/10.

🥉 #3 — GMED · Globus Medical

Sub-field: Surgical robotics (with a large spine-implant core) Market cap $10.90B · Price $80.33 · 20.8% off 52wk high ($101.40) · 52wk low $51.79

⚠️ Honesty first — what this is and is not

This is not a robotics pure-play, and I will not sell it as one. Enabling Technologies (the ExcelsiusGPS robot and capital portfolio) was $26.9M of revenue in Q1 2026, +21% YoY — roughly 3–4% of company revenue. GMED is a $2.94B spine-implant company that uses a robot as a razor-and-blades pull-through mechanism. The robot's job is to lock a surgeon into Globus implants.

That framing is exactly why it qualifies: unlike almost everything else in this field, you are not paying a robotics multiple for the robotics narrative. You get the physical-AI optionality embedded inside a business trading at 15.6x forward earnings with a net-cash balance sheet.

Metrics (§1 + §4 med-device lens)

Metric Value Note
P/E ttm / fwd 18.8 / 15.6 cheapest quality name in the screen
EV / EBITDA 11.15 lowest multiple of any name here with >20% op margin
FCF (2025) $579.0M FCF yield 5.3% on cap, 5.5% on EV
FCF 3yr CAGR +77.0% $104M → $579M
Revenue 3yr CAGR +42.2% ⚠️ almost entirely the NuVasive merger — inorganic. Organic 2026 guide is 9%.
Rev growth YoY +27%; 2026 guide $3.18–3.22B (+9%) material deceleration — the key bear input
Net income 3yr CAGR +41.4% ⚠️ 2025 NI of $537.9M exceeds operating income of $537.2M → one-off tax benefit. Use ttm EPS $4.28, not GAAP $3.92.
Debt / Assets 2.24% $118.7M debt vs $526.2M cash → net cash $407M. Fortress.
Current ratio 4.56
Shares (dil) 3yr CAGR +10.1% ⚠️ NuVasive stock deal. But $300M repurchased in 2025 — reversing.
ROE / ROA 13% / 8%
Gross margin 68%, rising (2024: 59% → 2025: 67%) merger synergies landing
Op margin 21% (from 8.7% in 2024) large, real operating leverage
Rev/share · FCF/share $21.44 · $4.22
Graham IV $57.96 vs $80.33 closest to price of any name screened

⚠️ Data gap: the source returned FY2025 shares outstanding as 112.63M against a diluted count of 137.06M and an implied ~135.7M from market cap. The 112.63M figure is almost certainly a data error — do not use it. Diluted ~137M is the right basis.

Where in the cycle

Not a capex cycle at all — hospital surgical volumes, which are secularly growing and largely recession-resistant. This is the deepest structural break from the industrial-capex risk the rest of this screen carries. The relevant "cycle" is the NuVasive integration curve, which is roughly two years in with margins inflecting hard.

China exposure

Negligible. US-dominant medical device revenue; spine robotics is a US/EU procedure market. This is the single cleanest China profile in the screen.

What they actually sell today — with revenue attached

$2.94B FY2025: - Musculoskeletal Solutions (~96%) — spinal implants, screws, interbody devices, trauma, orthopedics. The cash engine. - Enabling Technologies (~3–4%, $26.9M in Q1'26, +21% YoY) — ExcelsiusGPS robotic navigation, imaging, and a growing rental/placement program. ~130,000 robotic procedures performed to date. Robot placements +59% YoY, record units and dollars (Q1 2026 8-K).

The strategic shift is from one-time capital sales to placement + recurring disposables/software — which, if it works, converts a lumpy capital line into an annuity (Spine Market Group).

Moat — switching costs (surgeon workflow lock-in) + intangibles (regulatory/clinical)

Named explicitly. A surgeon trained on ExcelsiusGPS and a hospital that has installed it face high friction switching to Medtronic Mazor or Stryker Mako — retraining, revalidation, and OR workflow disruption. The robot then pulls Globus implants through for the life of the placement. Secondary moat: regulatory intangibles — FDA clearance plus accumulated clinical evidence across 130k procedures is a multi-year barrier. 68% gross margin, rising, is the §2.1 quantitative confirmation.

Bear case (adversarial)

  1. Growth is decelerating from 16.5% to 9%, and the 42% 3-year revenue CAGR is a merger artifact that will not repeat. If organic settles at mid-single-digits, 15.6x is fair value, not a discount.
  2. The robot is 3–4% of revenue. If you want physical-AI exposure, this is a homeopathic dose. Buying GMED for the robotics is buying a spine company and telling yourself a story.
  3. Competition is far larger. Medtronic (Mazor X) and Stryker (Mako) can outspend Globus indefinitely and bundle across a wider hospital relationship. Globus is the #3 player in an oligopoly.
  4. Dilution history. +10.1% 3yr share CAGR from the NuVasive deal. Per-share value creation is materially weaker than the headline growth (§0: "growth funded by dilution isn't growth for me"). The $300M buyback is a start, not a fix.
  5. 2025 net income is flattered by a tax item — GAAP earnings quality for that year is poor. Normalize before trusting the 18.8x.
  6. Hospital capital budgets are not immune to macro; capital robot placements can be deferred even when procedure volumes hold.

Diversification check

✅ The strongest genuine diversifier in the screen. Zero datacenter linkage, zero industrial-capex linkage, zero China linkage, beta 0.95. Demand is driven by demographics and surgical volumes. It also diversifies sector exposure — the portfolio's healthcare weight is thin relative to tech.

Valuation & entry (§3)

  • Forward EPS implied ~$5.15. At 18–20x (a defensible multiple for 68% GM, net cash, 9% growth) → $93–$103.
  • FCF: $579M at 5.5% required yield → ~$10.5B → ~$78/sh. This says the stock is already near fair on cash flow — an honest tension with the earnings multiple.
  • Graham IV $57.96 — the most credible Graham reading in this screen (real book value, genuinely profitable). Says $80 is not a bargain.
  • DYT/DDM: N/A, no dividend.
  • Analyst mean target $103.92 — I am below it.

Fair value range: $80–$100 (wide, because FCF and earnings-multiple approaches disagree). Entry zone: below $75. Conviction: 6.5/10.

3. Rejected — and precisely why

⏸ CGNX · Cognex — great company, wrong price. The hardest call in this screen. Everything about the business is working: Q1'26 revenue $268.4M, +24.3% YoY, beating consensus by 9.1%; adjusted EPS $0.34 vs $0.25 expected; adjusted EBITDA margin 26.9%, +1,010bp YoY, the seventh consecutive quarter of margin expansion; Q2 guided to +16.5% revenue and ~68% EPS growth; Greater China +36% (PR Newswire, Investing.com). 68% gross margin, essentially net cash ($263M vs $77M debt), consistent buybacks, and the purest machine-vision moat (switching costs — a vision system qualified into a line does not get swapped) in the field.

But §0 has two questions, and the second one fails. At $62 the stock has run 79% off its $34.60 low, trades at ~35x forward / 73x trailing / 41.5x EV/EBITDA, and its Graham IV is $13 against a $62 price. Meanwhile: the 3-year revenue CAGR is −0.4% — 2025 revenue of $994M is still below 2022's $1.01B. Cognex has spent a decade failing to grow through cycles. Management itself is "maintaining a cautious full-year outlook due to macro uncertainty and limited visibility into the second half." Logistics — 26% of revenue and the growth engine for eight quarters — is guided to decelerate to mid-to-high single digits. China is 15–20% of revenue. One source flatly calls it "over 40x forward earnings, priced for perfection."

Verdict: watchlist, not buy. Great company, bad entry. Revisit below $48 (~27x forward), or if two more quarters confirm that the revenue plateau of 2015–2025 is genuinely broken. This is the name I would most want to own at the right price — it is simply not the right price.

⏸ ROK · Rockwell Automation — The purest industrial-automation franchise available, and the market knows it. 4.8% off its 52-week high at 32.4x forward, 27.8x EV/EBITDA, 14.95x book, with the mean analyst target ($474.58) at the current price ($473.34) — i.e. zero implied upside from the sell side. FY26 guidance was raised to ~$8.9B on double-digit order growth. Great business, fully re-rated with the overbought industrials sector. Nothing to do here. ⚠️ Data gap: I could not source reliable China revenue % or quantified Inovance share-loss data — a genuine open question for any future Rockwell work.

⏸ TDY · Teledyne — Quality sensing/imaging conglomerate at 24.1x forward, 7.2% off high, beta 0.92. Fairly valued, not mispriced. Would be a reasonable core-industrial holding; is not a gem.

❌ TER · Teradyne — Owns Universal Robots and MiR, the most credible collaborative-robot franchise on this list. But revenue grew 104% YoY and earnings 386% — that is semiconductor test riding the AI wave, not robots. At 12.65x sales and 35x forward, you would be paying an AI-capex multiple to get a small robotics business. Directly the correlated bet the user is trying to avoid. Reject on diversification grounds alone.

❌ MKSI · MKS Inc. — 18.9x forward looks cheap and it is 35% off its high, but the 52-week range ($88 → $448) tells you this is a semi-cap cyclical. D/E of 153%. Same objection as TER: repackaged datacenter exposure.

❌ NOVT · Novanta — On paper the ideal pick-and-shovel: precision motion, photonics, medical robotics components, $5B cap, 19% off high. The numbers reject it. EPS has fallen four straight years ($2.06 → $1.77 → $1.47). FCF collapsed from $141.4M to $48.4M in 2025 (−66%), 3yr FCF CAGR −11.4%. Gross margin 44% and flat. Shares rising 0.7%/yr. And for that deterioration you pay 100x trailing / 33.6x forward. Short interest 17% of float. Great story, deteriorating business, premium price — the worst quadrant.

❌ IPGP · IPG Photonics — the value trap, and worth studying as the counter-example. This is exactly the "cyclically depressed vs structurally impaired" test the brief asked me to resolve, and IPGP is unambiguously impaired. Revenue: $1.43B (2022) → $1.00B (2025), 3yr CAGR −11.1%. Gross margin collapsed from ~55% historically to 38%. Operating margin 2%. FCF was negative $3.45M in 2025 with capex ($78.8M) exceeding operating cash flow ($75.3M). And the multiple is not cheap — 123x trailing, 36.3x forward. The cause is structural: Chinese fiber-laser makers (Raycus, Maxphotonics) destroyed IPG's price umbrella permanently. A vertically-integrated cost moat that has been undercut does not come back. 46% off its high and still not cheap. Textbook §0 value trap — poor business, expensive price.

❌ SYM · Symbotic — The most-hyped warehouse automation name. 51% off its high, 154x EV/EBITDA, 59x forward, 20% gross margin, negative net margin, 30% short interest. Essentially a Walmart-dependent systems integrator carrying customer-deposit float (which is why EV of $3.49B looks absurd against a $25.8B cap). §6 territory, and the brief's central test — story vs revenue — rejects it. The revenue is real but the margin structure is a systems-integration business wearing a software valuation.

❌ HSAI · Hesai — Profitable lidar with +30% revenue growth at 18.5x forward is genuinely interesting on the surface. Killed on three counts: it is a Chinese ADR (delisting/VIE/geopolitical risk the portfolio does not need), it is fundamentally an automotive supplier not an automation one, and the data source returned internally inconsistent market cap and a negative EV — I do not trust the numbers and will not build a case on them.

❌ GXO · Logistics — 14.2x forward and 25% off its high, but this is a contract-logistics operator: 15% gross margin, 3% operating margin, 198% D/E, $11.13B EV on $5.76B of equity. It buys automation; it does not sell it. Wrong layer of the stack.

❌ PRCT · PROCEPT BioRobotics — Down 65% from its high, unprofitable (−39% operating margin), −$1.83 EPS, 14% short interest. Real product (Aquablation) and 64% gross margin, but this belongs in the §6 small-cap framework with runway and dilution analysis leading, not in a value screen. If the user wants it, run /analyze-smallcap.

❌ The entire humanoid category — See §0. Fewer than 3,000–4,000 humanoids in productive commercial deployment worldwide, 90% of shipments Chinese, 23% buyer satisfaction, no investable US-listed pure-play with revenue. The pick-and-shovel suppliers (Harmonic Drive, THK, Sanyo Denki) are Japanese-listed and effectively inaccessible; the US-listed "humanoid supply chain" names are the Microstocks sleeve the agency has already covered. No position. The brief asked for skepticism here and the evidence fully justifies it.

4. Verdict on the field

The field is not picked over — it was genuinely never screened — but it is no longer cheap, and I want to be plain about that rather than manufacture a pick.

Three findings that matter more than any individual name:

1. The cycle question is resolved, and the answer is unhelpful for a value buyer. Automation was in a real trough in 2023–24. It is not anymore. Machine tool orders +27.8% in Q1'26, robot orders inflecting from pilot to deployment, ISM customer inventories too low. The pure-plays have already re-rated with the sector — ROK is 4.8% off its high at 32x forward with zero sell-side implied upside; CGNX has run 79% off its low to ~35x forward. You can buy the automation recovery today. You will pay 30–35x forward for it. Per §0, that is the "Great / Expensive" quadrant: watch, don't buy.

2. The mispricing that does exist is in businesses the market has mis-categorized, not mis-priced within the category. Zebra is a machine-vision and warehouse-robotics platform priced as a fading barcode-scanner company. Regal Rexnord is a motion-control franchise priced as a levered 2023 M&A accident. Globus is a robotics-enabled surgical franchise priced as a decelerating spine implant maker. In every case the mispricing comes from the market's label, not from the market's arithmetic — and in every case you must accept a real, named flaw (Zebra's device-refresh cyclicality, Regal's near-zero tangible book and 4% ROE, Globus's 3%-of-revenue robotics and dilution history) to get the discount. There is no clean gem here. I am not going to pretend otherwise.

3. The diversification mandate was the most useful filter in this screen, and it disqualified the highest-quality names. Teradyne owns the best collaborative-robotics assets on the list and is uninvestable for this portfolio because 104% revenue growth is semiconductor test riding the exact AI wave the user is already levered to. MKS the same. The user's constraint is doing real work — it is pushing toward genuinely uncorrelated end markets (retail/3PL logistics, industrial powertrain aftermarket, surgical volumes) and away from the comfortable, well-performing names that would have quietly doubled down on an existing bet.

On humanoids and "physical AI" as a theme: the honest answer is that the investable surface is nearly empty. The narrative is enormous and the revenue is not. Where revenue exists it is Chinese, unlisted, or Japanese-listed. The US-listed way to own physical AI today is to own boring, profitable companies that sell sensors, actuators, and vision systems into real factories and warehouses — and to buy them when the cycle is against them, which it currently is not.

If the user wants exposure now: ZBRA at ~15.6x forward is the most defensible entry, with the explicit understanding that the bear case (device refresh cycle, not automation secular growth) is unresolved. If the user can wait: CGNX below $48 is the highest-quality asset in the field and the one I would most want to own. Patience is a legitimate answer to this screen.

🎯 Single name most worth a full /analyze: ZBRA

It carries the widest gap between what the market thinks it owns and what it actually owns, it is the cheapest quality name relative to its own history (15.6x forward vs a 3-year average near 40x), it is a genuine diversifier away from the datacenter bet, and it has the two unresolved questions a full pipeline is built to answer: (a) how much of the 2026 revenue surge is a one-time enterprise device refresh versus durable automation demand, and (b) how large is the machine-vision/robotics business actually — which Zebra does not disclose and which is the entire optionality in the thesis.

Runner-up for a full workup: CGNX, as a valuation/entry-timing exercise rather than a buy decision.

5. Data gaps & caveats — flagged per §0

Gap Impact
ZBRA does not break out machine vision / robotics revenue Cannot size the core optionality in the #1 pick. Material.
RRX China revenue % not disclosed / not sourced Cannot quantify Inovance competitive risk. Verify before sizing.
ROK China % and quantified Chinese share loss unavailable Open question; searches returned nothing credible.
GMED FY2025 shares outstanding returned as 112.63M vs 137.06M diluted Almost certainly a source error. Use ~137M diluted.
GMED FY2025 net income ($537.9M) exceeds operating income ($537.2M) One-off tax benefit. GAAP EPS of $3.92 overstates run-rate quality.
HSAI market cap / EV internally inconsistent (EV printed as −$2.35B) All HSAI figures untrusted. Excluded on this basis.
LAZR ticker resolved to an ETF, not Luminar Luminar not evaluated.
Forward P/E figures mix GAAP and adjusted across sources ZBRA's 14.0 GAAP fwd vs ~15.6x on guided adjusted EPS. Adjusted used in valuation.
fin.py returns ~4 annual periods only All CAGRs are 3-year, not the framework's preferred 5–8yr. Escalate to roic.ai for real depth on any name taken forward.
Q2 2026 results not yet reported for ZBRA (due Aug 4, 2026) Thesis rests on Q1 actuals + Q2 guidance.

Sources