Reference Index — Sector 03: Humanoid and Robot Makers

Compiled 2026-07-14. Every figure below is sourced to a working link. Market caps carry an as-of date; private valuations carry the round and date.

Tesla, Inc. (the Optimus program)

What it is: The most-capitalized captive humanoid program in the world — the thesis' proof that a vertically integrated maker can build the robot, the factory, the data pipeline, the fleet, and the financing seat itself, and therefore never become addressable demand for anyone else. Status: Public (NASDAQ: TSLA). Optimus is an internal program, not a subsidiary. Market cap / valuation: ~$1.53 trillion as of 2026-07-13 (Macrotrends, TSLA market-cap series). Latest financials: Q1 2026 (quarter ended 2026-03-31): total revenue $22.4B, up ~15.8% YoY; automotive revenue $16.2B. 2026 capex guidance raised to >$25B (AI compute, Cybercab, Optimus production) — the thesis' outline cites "$20B+ from $8.53B in 2025," which the Q1-2026 update now supersedes at >$25B. Cited for: ~1,000+ Optimus units on live factory tasks by Jan 2026 and 2026 capex as captive-demand evidence (I; OUTLINE §Tesla (Optimus), ARGUMENT §"in-house collection operators"). Optimus Gen2 joint architecture — 14 rotary strain-wave joints + 14 linear planetary-roller-screw joints, ~11 more hand actuators, zero RV reducers (III.1; ARGUMENT claim 7). Musk's Q1-2025 call (2025-04-22) stating rare-earth magnet licensing was constraining Optimus production (III.5; ARGUMENT §"April 2025"). Morgan Stanley teardown: Optimus Gen2 BoM ~$46,000 with Chinese suppliers → ~$131,000 without (~2.85x) (III.4; V.5 "ex-China Optimus ~$46k"). Tesla Insurance as the captive-underwriting analog (V.2; ARGUMENT §2). Giga Texas 5.2M sq ft expansion and ~1M-unit/yr Fremont Optimus line (VI.6). Kill-case "captive capture" (V; OUTLINE Kill #3). Links:

Figure AI

What it is: The best-funded western humanoid pure-play and the thesis' flagship forward-deployed case — the BMW Spartanburg pilot is the most fully disclosed humanoid production deployment in existence, and Figure is one of the captives that could take the financing seat itself. Status: Private (Sunnyvale, CA). Investors include Parkway Venture Capital (Series C lead), Brookfield Asset Management, NVIDIA, Intel Capital, Macquarie Capital, LG Technology Ventures, Salesforce, T-Mobile Ventures, Qualcomm Ventures, Align Ventures, Tamarack Global. Market cap / valuation: $39B post-money, Series C (>$1B committed), announced 2025-09-16. Up ~15x from the $2.6B post-money of its Feb 2024 Series B. Latest financials: No audited financials disclosed (private). Operating disclosures instead: BotQ manufacturing line producing "one robot every 90 minutes" (Apr 2026); ~740 robots operating by end-June 2026; BotQ design target 12,000 units/yr. Cited for: Series C >$1B at $39B post-money (Sept 2025) and ~740 robots operating (I; OUTLINE §Figure AI (Helix)). Helix trained on ~500 teleop hours; Project Go-Big / Brookfield partnership using Brookfield's 100K residential units + 500M sq ft commercial as a captive human-video estate; navigation trained on 100% human video, zero robot demos (I; ARGUMENT §captive datasets). BMW Spartanburg: Figure 02 contributed to production of 30,000+ BMW X3 vehicles over ~10 months, moved 90,000+ components, ~1.2M steps (>200 miles), ~1,250 operating hours, 10-hour weekday shifts, 84-second cycle / 37-second load time, >99% successful placement per shift; Figure 03 now on logistics sequencing at the same plant (IV.1; ARGUMENT §Figure AI at BMW). Captive-capture kill case (V; OUTLINE Kill #3) and captive-underwriting threat (V.2). Links:

Agility Robotics (incl. the Churchill Capital Corp XI SPAC)

What it is: The most-deployed western humanoid platform (Digit) and the only one with a real Robots-as-a-Service commercial record — the thesis' central evidence that robot-hours, not robots, are what gets sold, and, via the SPAC, the first US-listed humanoid pure-play. Status: Private (Corvallis / Salem, OR; Pittsburgh). Announced 2026-06-24 a definitive business-combination agreement with Churchill Capital Corp XI (NASDAQ: CCXI); the combined company is expected to trade on Nasdaq as AGLT, closing expected in 2026 subject to shareholder vote, SEC S-4 review and exchange approval. Market cap / valuation: ~$2.5B implied enterprise value in the Churchill XI business combination announced 2026-06-24. Transaction expected to generate >$620M in gross proceeds. Latest financials: No audited revenue disclosed pre-S-4. Operating figures: Digit has logged 65,000+ hours of real-world operation across customers; >100 units deployed (Amazon, GXO, Spanx, Toyota Motor Manufacturing Canada, Mercado Libre, Schaeffler); Digit operating cost cited at ~$10–12/hour today against a target of $2–3/hour at scale; RaaS pricing cited around $30/hr and $2,000–8,000/month. Cited for: First formal commercial and first RaaS humanoid deployment, at GXO Flowery Branch, GA — pilot → RaaS contract → 100,000-tote milestone (IV.1; ARGUMENT §Agility/GXO). $10–12/hr operating cost, $2–3/hr target, $2,000–8,000/mo RaaS (V.4; V ARGUMENT §"Agility's Digit costs ~$10–12/hr"). >100 units across Amazon/GXO/Spanx/Toyota TMMC/Mercado Libre as the most-deployed platform (VI.1; ARGUMENT §495). Toyota Woodstock 7-unit RaaS signed Feb 2026 (VI). $2.5B Churchill SPAC to list on Nasdaq, mid-2026 (IV.1). Links:

Apptronik

What it is: The Austin humanoid maker (Apollo) that is simultaneously a Mercedes-Benz/GXO deployer, Google DeepMind's physical embodiment partner, and — via "Robot Park" — the captive-OEM version of the neutral data-generation seat the thesis says is still open. Status: Private (Austin, TX; spun out of UT Austin's Human Centered Robotics Lab). Investors include B Capital, Google, Mercedes-Benz, PEAK6, AT&T Ventures, John Deere, Qatar Investment Authority, Rockaway Ventures. Manufacturing partner: Jabil. Market cap / valuation: ~$5.3B post-money on the $520M Series A-X extension, announced 2026-02-11 (CNBC reported the extension as $520M at ~$5B; Apptronik's own release and Forbes put total Series A at >$935M). Roughly 3x the initial Series A valuation. Latest financials: No revenue disclosed. Funding: $415M initial oversubscribed Series A (2025) + $520M Series A-X extension (Feb 2026) = >$935M total Series A; proceeds earmarked for Apollo production ramp, commercial/pilot deployments, robot training and data-collection facilities, and a new robot debuting in 2026. Cited for: ~$935M Series A (Feb 2026) and ~$1B total raised, with production handed to Jabil targeting commercial quantities by 2027 (IV.1; ARGUMENT §77 and §532). Apollo at Mercedes-Benz Digital Factory Campus (Berlin-Marienfelde) and Kecskemét, Hungary, plus GXO (IV.1). "Robot Park" (Austin, 90k sq ft, opened June 2026, expanded mid-2026) as the Google DeepMind data-capture facility and the captive answer to the neutral-operator seat (I; VI.6; ARGUMENT §512). Named as a possible 11th–13th addressable buyer or merely an arm of DeepMind's demand (I; OUTLINE open questions). Links:

1X Technologies (NEO / Redwood)

What it is: The Norwegian-American home-humanoid maker whose "Expert Mode" teleoperation is the thesis' purest case of a customer-funded data flywheel — the buyer pays for the robot and, in escalations, supplies the training data. Status: Private (Moss, Norway; US HQ and manufacturing in Hayward, CA). Backers: OpenAI Startup Fund (Series A2, Mar 2023, $23.5M), EQT Ventures (Series B lead, Jan 2024, $100M), Samsung NEXT, Tiger Global, Nvidia. Market cap / valuation: Last publicly *closed* round is the $100M Series B (Jan 2024). Since Sept 2025 the company has been reported to be raising ~$1B at a valuation of at least $10B; that round is not confirmed closed in any primary source as of 2026-07-14. Treat "$10B" as a reported target, not a struck valuation. Total disclosed funding to date: ~$125M+ across the OpenAI-led A2 and the EQT-led B, plus undisclosed earlier rounds. Latest financials: No revenue disclosed. Commercial disclosures: NEO priced at $20,000 outright or $499/month (six-month minimum), pre-orders opened 2025-10-28, US deliveries targeted late 2026; first year of production capacity reportedly booked out in five days. Hayward factory (opened May 2026) billed as America's first vertically integrated humanoid factory. Dec 2025 EQT partnership to deploy up to 10,000 NEOs across EQT's 300+ portfolio companies, 2026–2030. Cited for: NEO at $20K / $499-mo, delivery late 2026, 60–70% autonomous initially with remote Expert-Mode operators filling the gap, targeting 80–90% by 2027; ~90% door-opening / ~80% fetch autonomy on easy tasks; privacy controls (no-go zones, face blur, opt-out) (I; IV.1; IV.3; ARGUMENT §545). Escalation rate ~30–40% today vs the 90% 2027 target as the driver of the margin structure (IV; ARGUMENT §446). The open question the thesis flags: 1X publishes an autonomy *percentage*, not a robots-per-operator ratio (IV; ARGUMENT §184). Links:

Unitree Robotics (宇树科技)

What it is: The Chinese price-floor setter — the reason the thesis' residual-value argument holds, because a G1 at ~$13,500 collapses the depreciation curve underneath every western unit, and the platform teleop-arbitrage operators buy in bulk. Status: Private, transitioning to public. IPO status as of mid-2026: the China Securities Regulatory Commission approved Unitree's STAR Market (Shanghai) IPO registration in early July 2026; filing accepted 2026-03-20, listing-committee review passed 2026-06-01. Debut targeted as early as late July 2026. Not yet trading as of 2026-07-14. Market cap / valuation: IPO seeks ~RMB 4.2B (~$618M) for at least a 10% stake — an implied post-money of roughly RMB 42B (~$5.9–6.2B), per the approved registration (July 2026). Latest financials: Per the IPO prospectus as reported: revenue RMB 1.17B for the first nine months of 2025; net profit RMB 105M; humanoid robots ~51.5% of revenue. Cited for: Unitree G1 base ~$13,500 official store price (down from $16k) as the Chinese-floor mark against ARK's ~$16k hardware break-even (III.6; V.5; ARGUMENT §76). Full price ladder mid-2026: G1 twelve configs to $73,900; H2 from $29,900 commercial / $40,900 EDU, shipping Apr 2026; H1 ~$90,000; R1 $4,900–5,900 (July 2025) (V.4). Unitree has qualified domestic harmonic reducers for production — the "warm base" existence proof (III.1; III.6; ARGUMENT claim on domestic reducers). Shuanghuan's Suzhou plant supplies both Optimus and Unitree (III.1). G1 as the platform Eastworld's teleop-arbitrage operation buys in bulk (IV.3). Links:

UBTech Robotics (优必选)

What it is: The listed Chinese humanoid maker whose Walker S order book and Liuzhou production line are the thesis' hard volume signal — the only humanoid company with audited financials and a mass-production delivery record. Status: Public (SEHK: 9880), listed in Hong Kong Dec 2023. Market cap / valuation: HK$41.99B (~US$5.4B) at HK$83.55/share, as of 2026-07-14. The stock has fallen materially from ~HK$128/share in Jan 2026 — worth noting when the thesis leans on UBTech as the "volume signal." Latest financials: FY2025 (year ended 2025-12-31): revenue RMB 2.00B, up ~53% YoY; net loss RMB 703M (a ~37% narrowing vs FY2024). Still loss-making. Cited for: Walker S2 mass production and delivery begun late 2025; ~RMB 800M (~$112M) in Walker-series orders since early 2025; 1,000th unit off the Liuzhou line; capacity targeted at 5,000 units/yr in 2026 and 10,000/yr in 2027; deployed at BYD, Geely, FAW-Volkswagen, Foxconn, SF Express, Airbus (Jan 2026) (IV.1; ARGUMENT §UBTech/Walker S2). ~$80M of robots procured into Chinese national training centres (II; OUTLINE §237). Anchor tenant of the Beijing Robot Mall (Yizhuang/E-Town, opened 2025-08-08) (IV.4; VI.6). Links:

AgiBot / Zhiyuan Robotics (智元机器人)

What it is: The world's #1 humanoid shipper by units and the open-dataset publisher (AgiBot World) — the thesis' evidence that China's "warm base" is compounding volume and public training data faster than any western captive. Status: Private (Shanghai). Co-founded by ex-Huawei "Genius Youth" engineer Peng Zhihui. Backers include Tencent, BYD, HongShan/Sequoia China, Hillhouse, and Shanghai state funds. Reportedly preparing a Hong Kong IPO in 2026. Market cap / valuation: Reported at ~US$2.09B as a private unicorn (Dealroom, 2026); Chinese press has put a later round near RMB 45B+ and the planned HK IPO is reportedly targeting HK$40–50B (~US$5.1–6.4B). Could not source a single company-confirmed post-money to a primary link as of 2026-07-14 — AgiBot has never issued a valuation press release. Use the shipment and production milestones, which are primary, rather than the valuation. Latest financials: No audited financials public. Production: 5,168 humanoid units shipped in 2025 (39% of a ~13,000-unit global market, Omdia — the #1 position worldwide); 10,000th robot off the line late 2025/early 2026; 15,000th robot off the production line announced 2026-06-29. Cited for: AgiBot the leading shipper at ~5,168 units, and AgiBot World open datasets (1M+ trajectories; "Rich Interaction" 2026 theme) as the China data-flywheel evidence (I; OUTLINE §247–250). AgiBot has qualified domestic harmonic reducers for production — the warm-base existence proof alongside Unitree (III.1; III.6; ARGUMENT claim on domestic reducers). Links:

Galaxea AI (星海图)

What it is: The Beijing embodied-AI robot maker whose 2026 funding pace is the thesis' clearest single illustration of the Chinese capital frenzy — three up-rounds in eight months — and an anchor tenant of the state-run Beijing Robot Mall. Status: Private (Beijing). Investors include Lens Technology (hardware partner) and Chinese state-backed funds; nearly 20 participants in the latest round. Market cap / valuation: >RMB 20B (~US$2.8–2.9B) after a RMB 2B (~$291M) round reported 2026-04-02. Prior marks: RMB 10B (~$1.4B) on a RMB 1B (~$144M) Series B, Feb 2026; ~$700M valuation on a $100M round, Aug 2025. Caution: at least one English outlet (The AI Insider, 2026-04-04) rendered the latest mark as "$29B USD" — that is a unit error on the RMB 20B figure; the correct dollar equivalent is ~$2.9B. Do not carry the $29B number. Latest financials: No revenue disclosed. Cumulative funding ~RMB 3B+ (~$430M+) as of Apr 2026. Products: industrial robotic arms and humanoid/embodied-AI platforms for handling, assembly, inspection and logistics. Cited for: Named as one of the ten anchor tenants of the Beijing Robot Mall (Yizhuang / E-Town), alongside the Beijing Humanoid Robot Innovation Center and UBTech (IV.4 / OUTLINE §313). Links:

Boston Dynamics (Hyundai Motor Group)

What it is: The oldest western humanoid program, now the captive robotics arm of an automaker — and the sharpest case in the thesis of an OEM keeping all of its robots and all of its data inside the house. Status: Private subsidiary. Hyundai Motor Group holds a controlling stake (acquired ~80% from SoftBank in a deal closing June 2021 that valued Boston Dynamics at ~$1.1B); SoftBank retained a minority. AI partner: Google DeepMind. Market cap / valuation: ~$1.1B enterprise value at the 2021 Hyundai control acquisition — the last disclosed mark. No newer valuation has been published; Hyundai does not carve out Boston Dynamics in its market cap. Hyundai Motor Group has announced a $26B US investment programme covering robotics infrastructure, which is the number the thesis should use for scale, not a BD valuation. Latest financials: Boston Dynamics does not publish standalone financials. Operating disclosures: production-ready electric Atlas launched at CES 2026; all 2026 Atlas production is committed — to Hyundai's Robotics Metaplant Application Center (RMAC) and to Google DeepMind; a robotics factory targeting 30,000 Atlas units/yr is planned for 2028; Hyundai has committed to deploying >25,000 Atlas robots across Hyundai/Kia plants (≈83% of that capacity), first at Metaplant America (Savannah, GA) in 2028 and Kia Georgia in 2029. Cited for: All 2026 electric Atlas units committed to Hyundai's RMAC and to Google DeepMind — i.e. zero addressable supply, the captive-demand argument (I; OUTLINE §255–260; ARGUMENT §155). RMAC explicitly framed as a "data factory" building a humanoid manufacturing dataset (I). Hyundai/BD counted as one of the "12–13 buyers" that has replaced the corpus' "~10 buyers" framing (I; OUTLINE §286). Links:

Clone Robotics

What it is: The artificial-muscle humanoid maker (Protoclone / Clone Alpha) that the thesis uses as the sharpest *substitute* threat to the actuator chokepoint — a body with zero reducers and zero roller screws — and then defeats on the ground that it has shipped no qualified fleet. Status: Private. Founded 2021 in Wrocław, Poland; second office opened in Mountain View, CA (2026). Y Combinator-adjacent; investors include Initialized Capital (seed lead), Pioneer Fund, Access VC, Tango VC, Wikus Ventures, plus a Republic crowdfunding raise. Market cap / valuation: No priced institutional round disclosed. Total funding ~$7.1M (Crunchbase/Tracxn, 2026), anchored by a ~$6.5M seed led by Initialized Capital (2023); its Republic community round carried a ~$30M valuation cap. This is the only humanoid maker in this sector file whose entire capital base is smaller than a single week of Tesla's Optimus capex — which is precisely the thesis' point. Latest financials: No revenue. Clone Alpha announced for limited release with a stated ambition of a sub-$20,000 "synthetic human" build cost (CEO Dhanush Radhakrishnan, 2026 Abundance Summit). No qualified or deployed fleet as of July 2026. Cited for: Protoclone / Clone Alpha replace motors and gears entirely with 1,000+ water-powered "Myofiber" muscles over 206 3D-printed bones, contracting in under 15–50 ms, ~1 kg of force per 3-gram fiber, coordinated by a Jetson Orin and 500+ sensors; 200+ degrees of freedom. Zero reducers, zero roller screws — the steelman substitute in III.2(b) (ARGUMENT §117–118 and claim 9). The rebuttal the thesis rests on: "Clone Alpha is real and impressive and it has shipped exactly zero units into a qualified fleet" (III; ARGUMENT §156). Links:

Reflex Robotics

What it is: The cheap wheeled-humanoid maker that is the third brand in GXO's multi-brand incubator — the thesis' evidence that a neutral deployer can already run three vendors' machines side by side in one warehouse. Status: Private (New York, NY). Investors: Khosla Ventures, 8090 Industries, Crossover VC, Invariantes Fund, Julian Capital, SNR. Market cap / valuation: No valuation disclosed. Total funding ~$7.0–7.3M, raised in a single seed round (Khosla Ventures and 8090 Industries participating), per Crunchbase and Tracxn as of 2026. Could not source a post-money valuation to a primary link as of 2026-07-14; the seed total is the best available figure. Latest financials: No revenue disclosed. Product: a wheeled mobile-manipulator humanoid targeting a ~$10k price point; reaches operational capability within 60 minutes of deployment and improves toward autonomy by learning from human demonstrations. Second-generation commercial system in pilot deployments as of 2026. GXO's second Robots-as-a-Service agreement. Cited for: Named as the third brand in GXO's three-brand humanoid incubator alongside Agility's Digit and Apptronik's Apollo — running in a live warehouse, picking cardboard off Locus bots at a sports-apparel client (IV.2; VI.1; ARGUMENT §92–95, §164, §510). Links:

Foundation Robotics / Foundation Future Industries (Phantom)

What it is: The defense-humanoid startup whose Pentagon award is the thesis' first sighting of a *creditworthy* demand guarantee for robot-hours — government as the anchor counterparty — and, on inspection, the cautionary tale about what that award actually is. Status: Private (San Francisco). Backers include Eric Trump (publicly associated); the company markets the Phantom MK-1 as a US-made counter to Chinese humanoids. Market cap / valuation: No closed priced round disclosed. Reported to be seeking ~$500M at a $3B+ valuation (2026); total funding raised to date is roughly $21M. Treat the $3B as a sought valuation, not a struck one. Latest financials: No revenue disclosed. Government contracts: ~$24M in Pentagon contracts announced April 2026, on top of ~$10M in prior government work. Unit price cited at ~$150k. Stated plan: 50,000 units by end-2027 — from a base of roughly 40 units, i.e. a ~250x scale-up on ~$21M of capital, which is the reason the thesis flags the number rather than banking it. Two Phantom MK-1 units sent to Ukraine (Feb 2026) for logistics and reconnaissance testing. Phantom MK-1: 5'9", 176 lb, 19 upper-body DoF, 8 cameras, no LiDAR, proprietary cycloidal actuators to ~160 N·m, 1.7 m/s walk, 44-lb payload. Cited for: The ~$24M Pentagon award, ~$10M prior government contracts, ~$150k/unit, and the 50,000-by-2027 plan as evidence that "first creditworthy demand guarantees are being signed on the defense side" (VI.3; OUTLINE §62). Critically, the thesis' own correction: the ~$24M is SBIR research contracts (incl. an SBIR Phase 3 designation), not a procurement order (VI; ARGUMENT §451, §555) — the SBIR Phase 3 designation qualifies Foundation as an approved military vendor but does not itself buy 50,000 robots. Links:

XPeng (IRON)

What it is: The Chinese EV maker running the exact Tesla playbook one step behind — car plant, VLA model, robotaxi, humanoid — and therefore the second data point for the thesis' claim that vertically integrated automakers, not robot startups, are the ones with the capital to be captives. Status: Public (NYSE: XPEV; HKEX: 9868). IRON is an internal robotics division; He Xiaopeng took the robotics-division CEO role concurrently with the group CEO role in an internal letter dated 2026-06-10. Market cap / valuation: ~$12.68B at $13.26/share as of 2026-07-14 (StockAnalysis). Down ~26% over the trailing year. Latest financials: TTM revenue ~$10.72B (up ~47.5%); TTM net income −$327.6M. FY2025 revenue RMB 76.72B (+87.7% YoY), with losses narrowing ~80% vs 2024. Q2 2026 deliveries: 103,295 vehicles. Cited for: XPeng (IRON) is flagged in the thesis as surfaced adjacent but not profiled — Part I explicitly lists it, with Samsung/Rainbow Robotics, as a maker that came up in the sweep but was not written up (I; OUTLINE §295). The reference facts a profile would need: next-generation IRON unveiled at XPeng AI Day (2025-11-05) with a biomimetic spine, flexible outer skin, and 22 DoF per hand; large-scale production targeted by end-2026, first commercial deployment expected Q1 2027, initial deployment as retail shopping guides in XPeng stores. Links:

Sanctuary AI

What it is: The Vancouver humanoid maker that got to industry-leading hydraulic hands and then could not get to a fleet — and in 2026 abandoned selling whole robots to sell software and hands to other people's robots. It is the sector's clearest case of the thesis' central claim that hardware novelty without a deployment record is not a business. Status: Private (Vancouver, BC). Daniel Friedmann appointed CEO 2026-06-26 amid a strategic pivot; the prior CEO was removed in Nov 2024 amid investor concern about competitiveness. Market cap / valuation: No post-money valuation disclosed. Total funding ~US$130M cumulative through early 2026, anchored by a US$76M Series A (Nov 2022) plus follow-ons and a ~CAD$30M Canadian Strategic Innovation Fund grant. The Logic reported in 2026 that Sanctuary was attempting to raise more money — i.e. no fresh priced round on the record. Could not source a company-confirmed valuation to a primary link as of 2026-07-14. Latest financials: No revenue disclosed. ~165 employees as of 2026-05-31. Strategic position as of 2026: Phoenix is a research/pilot platform used to capture teleoperation data, not a product for sale; the company now sells its "Physical AI" software stack and its hydraulic dexterous hands into other makers' robots. Cited for: Not named in the current thesis drafts. A grep of all six OUTLINE.md and ARGUMENT.md files on 2026-07-14 returns zero hits for "Sanctuary." It is documented here because it is the negative control for the sector: a well-funded western humanoid maker with a genuine hardware edge (hands/tactile) that never accumulated a deployment record and has now exited whole-robot sales — the exact failure mode Part V's residual-value and Part VI's operator-record arguments predict. Links:

Fourier (Fourier Intelligence, 傅利叶)

What it is: The Shanghai humanoid maker that came from rehabilitation robotics and has stayed in a real, paying vertical (care and rehab) rather than chasing general-purpose factory work — the counter-example to the "general-purpose or nothing" framing. Status: Private (Shanghai). Founded 2015 by Alex Gu and Zen Koh. Investors include Guoxin Investment, Prosperity7 (Aramco), Saudi Aramco-linked and Chinese state vehicles, Runyang Technology. No announced IPO as of 2026-07-14. Market cap / valuation: No post-money valuation disclosed by the company. Total disclosed funding ~US$246M. Most recent disclosed round: a CNY 300M (~$43M) Series E+ corporate minority investment closed 2025-08-01, led by Runyang Technology; the earlier Series E was ~CNY 800M (~$110M) with Guoxin Investment and Prosperity7. Could not source a company-confirmed valuation to a primary link as of 2026-07-14 — total funding is the honest figure. Latest financials: No revenue disclosed. Product line: GR-2 general humanoid; GR-3 "Care-Bot" unveiled Aug 2025 and shown at CES 2026 (its first major US trade show) — 55 DoF, four-microphone array, 31 pressure sensors, structured-light vision. Cited for: Not named in the current thesis drafts — a grep of all six OUTLINE.md and ARGUMENT.md files on 2026-07-14 returns zero hits for "Fourier." Documented here because it is the credible Chinese humanoid maker with an actual healthcare revenue line, which is the strongest available counter to Part IV's assumption that logistics and manufacturing are the only near-term humanoid markets. Links:

Galbot (银河通用 / Galaxy General)

What it is: China's highest-valued unlisted embodied-AI company and the operator of the country's most-photographed live humanoid retail deployment — the FamilyMart store robot that, on close reading, still needs four human staff. Status: Private (Beijing). Founded May 2023. Investors include the third phase of China's National IC "Big Fund," SMIC-affiliated Zhongxin Juyuan, CATL, and Chinese state vehicles. Reportedly eyeing a Hong Kong IPO. Market cap / valuation: >RMB 20B (~US$3B) after a RMB 2.5B (~$362M) round announced March 2026 — described by Caixin as making it the highest-valued unlisted humanoid-robotics firm in China. Cumulative funding ~RMB 7B (~$1B) in three years. Latest financials: No revenue disclosed. Deployment record: humanoid store clerks in retail (a FamilyMart in Beijing's Zhongguancun district), where the robot retrieves items from a refrigerated case but cannot complete a transaction; the store still runs four full-time employees — the same headcount as a comparable FamilyMart without a robot. Mass production of Galbot units projected for 2026. Cited for: Not named in the current thesis drafts — grep of all six OUTLINE.md and ARGUMENT.md files on 2026-07-14 returns zero hits for "Galbot." Documented here because the FamilyMart deployment is the single best public data point against the labour-substitution assumption the whole humanoid TAM rests on (Part III's ARK ~$24T full-substitution TAM; Part V's RaaS unit economics): a live, funded, state-backed humanoid retail deployment that displaced zero headcount. Links:

Dexterity

What it is: The non-humanoid warehouse-robot maker with real paying logistics customers (FedEx, UPS, GXO) — in the thesis it appears as a named customer of Foxglove's data/observability platform, and it functions as the sanity check on humanoid economics: purpose-built arms already do the work humanoids are being funded to learn. Status: Private (Redwood City, CA). Investors include Lightspeed Venture Partners and Sumitomo Corporation (co-leads of the latest round), plus Kleiner Perkins and Obvious Ventures historically. Market cap / valuation: $1.65B post-money, on a $95M round announced 2025-03-11 (led by Lightspeed and Sumitomo). Total capital raised: ~$300M. Latest financials: No revenue disclosed. Product/commercial: DexR truck-trailer-loading robot co-developed with FedEx; "Mech," a two-armed robot with a 16-foot arm span lifting 130+ lb on a mobile base, for extreme-temperature environments. Named customers: FedEx, UPS, GXO. Cited for: Listed among Foxglove's customers (with NVIDIA, Amazon, Anduril, Wayve) in the argument that the record substrate for physical AI is being claimed by a well-capitalized observability vendor rather than by a neutral operator (VI.2; OUTLINE §43; ARGUMENT §440). Links:

Anduril Industries

What it is: The defense-tech company that is the sector's proof that a private hardware firm can reach real, audited-scale revenue against a government counterparty — the model Foundation's Pentagon route is chasing, and a named customer of the physical-AI record substrate. Status: Private (Costa Mesa, CA). Investors include Thrive Capital and Andreessen Horowitz (Series H co-leads), Founders Fund. Market cap / valuation: $61B post-money on a $5B Series H, announced 2026-05-13 — more than double the $30.5B mark set under a year earlier on a $2.5B round led by Founders Fund. Total raised: >$11B. Latest financials: Revenue doubled to ~$2.2B in 2025; the company has told investors it expects roughly $4.3B in 2026. These are company-provided figures reported by CNBC/TechCrunch, not audited public filings. Cited for: Named as a customer of Foxglove's multimodal data/observability platform for physical AI, alongside NVIDIA, Amazon, Wayve and Dexterity — the evidence that the "record substrate" seat is being taken by a vendor, not a neutral operator (VI.2; OUTLINE §43; ARGUMENT §440). Its scale is also the implicit benchmark behind Part VI's claim that government is the first creditworthy counterparty for autonomous-machine hours. Links: