Robotics Weekly Analysis — 2026-11-09: Chinese authorities forced the China business out of SoftBank's acquisition of ABB Robotics, and Agility Robotics became the first pure-play humanoid company listed in the US, beginning quarterly disclosure

Briefings covered: 2026-11-09 news briefing (period covered 2026-11-03–11-09, including the GenRobot AI funding item published on 11/02)

Frame: This analysis tracks three areas — humanoids and Physical AI, drive-train components (actuators, reducers, motors), and industrial robots, cobots and logistics AMRs — and divides robot brains into VLA, the world-model approach and WAM (a structure that uses a video world model as the policy backbone), reading them not as rival camps but in terms of which layer of a combined stack holds pricing power, and it accepts only externally reproduced real-robot performance as verification.

This Week's Key Points

Chinese authorities blocked the China portion of SoftBank's acquisition of ABB Robotics, and the two companies are considering carving out the China business. ABB said in a statement on 11/5, and SoftBank in a notice on 11/6, that China's foreign-investment security review body notified them on 11/4 that the acquisition of the China business is prohibited. The deal is structured as a single acquisition of a holding company, so removing only the China portion does not fit the original contract. Reports that the reason is concern about robot operating data leaving the country are unverified, as the authority has given no official explanation. This is the first case in which a transfer of ownership of one of the top four industrial robot makers has been stopped by a Chinese review, and together with last edition's US Section 232 determination, it continues the direction in which the industrial robot supply chain is being divided by region under reviews on both the US and Chinese sides.

A pure-play humanoid company is, for the first time, subject to quarterly disclosure on a US public market. Agility Robotics closed its SPAC merger with Churchill Capital XI on 11/4 and began trading on Nasdaq on 11/5. Gross proceeds were about US$465 million before costs, 2025 revenue was US$1.78 million, the net loss was US$138.1 million, and 64% of revenue came from related-party transactions. Until now, valuations of private humanoid companies were discussed only through round prices, with no disclosed revenue. There is now a comparable whose revenue, operating hours and order conversion are disclosed every quarter.

Reports confirmed a structure in which Atlas's core components are filled by Hyundai Motor Group affiliates and large Korean companies, and four Korean component makers are targeting the same positions with integrated actuators. On 11/4 the Electronic Times reported that Hyundai Mobis will supply all actuators and that Samsung SDI and LG Energy Solution will handle batteries (based on industry sourcing, with no official confirmation from the companies). In the same week at RoboWorld 2026, SPG, SBB Tech, Samhyun and C&M unveiled integrated actuators that combine motor, reducer and controller rather than stand-alone reducers. None disclosed orders tied to humanoid volume production.

In the data layer and the world-model approach, funding and results gathered around approaches that use human motion data in place of robot teleoperation data. China's GenRobot AI raised a US$100 million Series A+ for human-demonstration data collection infrastructure, and LiberAI released Liber-0, a world action model pretrained on 2.2 million hours of human video, with its own evaluation suggesting that instrumented-glove data can replace part of robot teleoperation data. Externally reproduced real-robot performance remained at zero cases this week.

Correction to last week's view. SoftBank's acquisition of ABB Robotics, which last edition's confirmation-date table listed as "closing within the year," can no longer close in its original structure because of the Chinese prohibition, so the tracking criterion changes to whether a carve-out structure is agreed. The number of model-layer rounds that disclosed revenue or external verification was zero again this week, making four consecutive weeks, and under the criterion set in the 10/19 edition, model-layer rounds will from now on be recorded separately as "capital signals" only, not as evidence of demand or bottlenecks.

1. Theme and Value-Chain Implications

(1) China's security review has become a variable that changes the deal structure of industrial robot M&A

• Confirmed fact (SoftBank notice 11/6, ABB statement 11/5): The foreign-investment security review body under China's National Development and Reform Commission notified the parties on 11/4 that the acquisition of ABB Robotics' China business is prohibited. The enterprise value of the whole transaction is US$5.375 billion. China is ABB Robotics' largest market, and 90% of China sales are produced locally.
• Follow-up (citing Bloomberg, 11/7): A plan led by SoftBank to spin the China business into a separate entity and then acquire the rest is under discussion at an early stage. ABB is also considering a sale of the China business to a third party (valued at US$2 billion to US$3 billion) or reviving a spin-off listing. ABB had originally expected to close within the year.
• Reasonable inference: Acquisitions of Japanese and European robot makers with production and sales bases in China are now more likely to be designed around a prior carve-out of the China business. A carve-out lowers both the price and the speed of a deal, and when the China business is sold separately, the buyers are effectively narrowed to Chinese capital. If the reason for the prohibition is robot operating data, as reported, it means data generated by robots in the field is being classified as a national-security asset, and that extends as a jurisdictional risk to Physical AI training-data businesses as well. Because the reason is unverified, this part is a hypothesis.
• Speculation: ABB Robotics without its China business would be smaller in revenue but also less exposed to price competition with Chinese local makers. Whether SoftBank will proceed with the acquisition after a carve-out cannot be known, as the company has made no statement.

(2) With a humanoid company now subject to quarterly disclosure, a benchmark for private valuations exists

• Confirmed fact (Agility press release, The Robot Report, Bloomberg, 11/5): The opening price was US$11.26 (SPAC reference price US$10). Reports of the first-day close are conflicting (US$11.51 versus US$10.83), and the stock returned to around US$10 over the following two days. Proceeds of about US$465 million consist of US$164 million from the trust, a US$201 million PIPE led by Foxconn and a separate US$100 million round. Digit has 65,000 cumulative operating hours (nine customer sites), next-generation Digit 5 has multi-year orders of more than US$300 million (conditional on milestones and specifications), early deliveries are in 2027, and existing shareholders have a 180-day lock-up.
• Profit and loss structure: 2025 revenue of US$1.78 million against a net loss of US$138.1 million means the loss is about 78 times revenue. Excluding related-party revenue, external customer revenue works out to about US$640,000 (simple calculation from disclosed figures). The orders of more than US$300 million are conditional on milestones and specifications and are not confirmed revenue.
• Reasonable inference: Round prices of private humanoid companies with no public revenue will from now on have to be explained against Agility's quarterly revenue, operating hours and order conversion. Market capitalization is not in this briefing and 2026 revenue has not been disclosed, so no multiple is calculated. Because the float is limited until the lock-up expires (May 2027), the share price is hard to use as a valuation benchmark, and only disclosed figures are used.
• Form factor: Digit is a bipedal humanoid, and its operating hours at logistics sites are per the company. This is not external verification, but it has become a figure that can be tracked through quarterly disclosure.

(3) In humanoid drive trains, in-group supply and competition among Korean integrated actuators are proceeding at the same time

• Reported (Electronic Times 11/4, based on industry sourcing): Hyundai Mobis supplies all Atlas actuators, Samsung SDI and LG Energy Solution supply batteries, and camera modules come from Hyundai Mobis, Namuga, MCNEX and Cammsys. The battery specification cited by Robot News on 11/8 (70 Samsung SDI 2170 cylindrical cells, pack of about 1.3 kWh) is unverified, as Boston Dynamics and Samsung SDI have not confirmed it. The fact confirmed last edition is that Hyundai Mobis produces prototypes from November and supplies from 2028.
• Reasonable inference: This report supports the hypothesis written last edition that "the position of external reducer and motor makers narrows to sub-components of Hyundai Mobis's joint module." Batteries use EV cylindrical cells as they are, so there is no demand yet for robot-specific cells.
• Confirmed fact (RoboWorld 2026, 11/4–7): SPG showed its SDD lineup integrating motor, reducer and driver (performance and mass-production timing undisclosed); SBB Tech showed precision reducers for humanoids plus ultra-small reducers and intelligent actuators under development; Samhyun showed three AXLON joint actuators; and C&M (under preliminary KOSDAQ listing review) unveiled five integrated HyperAct models. The only confirmed order is one supply contract between C&M and OpenPath Robotics for quadruped robots covering 2027–2030, with amount and volume undisclosed. C&M's claim of double torque density is per the company.
• Interpretation: All four are trying to move up from stand-alone reducers to integrated modules. As more supply candidates target the same position, the bargaining power of humanoid makers grows and component margins are squeezed (reasonable inference). Outside the Atlas volume, where in-group supply is set, the customers independent component makers can target are Korean humanoid makers not belonging to a group, and quadruped and cobot makers. Whether humanoid volume production has appeared in independent reducer makers' orders can be checked in Harmonic Drive Systems' results on 11/11.

(4) In logistics unloading and delivery robots, the focus of design and business has shifted to serviceability and the operations layer

• Confirmed fact (The Robot Report, 11/6): Boston Dynamics unveiled the second generation of its trailer-unloading robot Stretch. Up to 1,000 cases per hour; multi-pick that grabs several cases at once using a valve on each suction cup; two-shift batteries with a 75-minute charge to 90%; 81% less internal wiring; automotive-grade connectors. It is operating at multiple customer sites, and the program has more than 100,000 cumulative operating hours. The 24% unloading speed improvement for the existing fleet is per the company. Hyundai Motor supports procurement of cable harnesses through its automotive supply chain. Price and unit sales are undisclosed.
• Interpretation: The design changes focus on serviceability, durability and parts standardization rather than new functions, which indicates a product intended for operation in units of hundreds. The entry of automotive component unit costs into robot costs is a concrete example of the cost advantage a robot maker under an automotive group holds over independent robot makers (reasonable inference).
• Confirmed fact (Serve Robotics results, 11/5): Third-quarter revenue of US$2.8 million (US$0.7 million a year earlier, US$3.2 million the previous quarter), GAAP net loss of US$59.2 million, cash of US$198 million. Average daily active robots 917 (312 a year earlier). Full-year revenue guidance raised to US$12 million and cost guidance lowered to US$150 million. It launched Physical AI Services, which provides its fleet-operation technology to other robot makers.
• Interpretation: Even with three times as many robots in operation, sidewalk delivery robots generate quarterly revenue of around US$3 million, and the loss is 20 times revenue. Expanding into selling operations technology externally reads as a judgment that revenue per robot alone cannot sustain the business. As with Bedrock and Foundry last edition, this adds one more case in which capital and business direction are heading toward the operations layer rather than robot manufacturing.
• Korean cobots: Doosan Robotics disclosed on 10/29 third-quarter revenue of KRW 16.175 billion (+59.1%) and an operating loss of KRW 13.4 billion (loss narrowed 12.2%), with growth led by North American end-of-line automation revenue. This is growth in the same direction as Teradyne Robotics last edition (+28%, return to profit), but the bottom line is still a loss.

(5) The data layer and the world-model approach moved toward replacing robot data with human motion data

• Confirmed fact (TechNode and others, 11/2): GenRobot AI (Beijing) raised a US$100 million Series A+ (round closed in the third quarter, valuation undisclosed). Investors include SAIC, Baidu, Galaxy General Robotics, Shoucheng, Eastworlds, GL Ventures and CITIC Goldstone. It operates collection equipment that records video together with hand and body motion for household, manufacturing and logistics tasks, plus a cleaning and labeling platform. More than 1 million cumulative hours of data and more than 30 partner AI companies are per the company. Cumulative funding of more than US$200 million within a year of founding is as reported.
• Interpretation: With an automaker, a search company and a humanoid maker investing together, this is a model in which the data supplier becomes shared infrastructure for the demand-side companies. Because the focus is human motion data rather than robot teleoperation data, the quality gap against real-robot data needs verification.
• Confirmed fact (The Robot Report and others, 11/4): LiberAI (Beijing, founded December 2025) released Liber-0, which the company describes as a WAM structure in which a video world model predicts the next scene and the action together. It uses more than 2.2 million hours of human activity video, hand motion recorded with instrumented gloves and robot teleoperation data, and a unified embodiment representation that takes the body dimensions of humans and robots as input. The 36% zero-shot success rate on 60 unseen tasks, and the higher score for a combination of 100% glove data and 25% robot data than for 100% robot data (75.6 versus 72.8, eight tasks), are all per the company's evaluation. The performance gap between its own robot LEO and third-party platforms is large (average 90.4 versus 62.1 on three tasks). Whether weights are released, and any external reproduction, are unverified.
• Interpretation: On the company's own evaluation, the result is that low-cost glove data can replace part of robot teleoperation data, and the 36% zero-shot rate also indicates limits on long-horizon tasks. Because the gap between its own hardware and third-party hardware is large, the cross-embodiment claim is held in reserve. The model is recorded as WAM per the company's description, with the world-model approach used as the general term.

World-model hypothesis update (this edition): Undecidable, unchanged. To the hypothesis that if video pretraining carries through to policies, the robot demonstration-data moat weakens and the video-asset and compute moat strengthens, this week added one company self-evaluation (LiberAI) finding that "glove data replaces part of robot data" and one human-motion-data infrastructure round (GenRobot AI, US$100 million) as signals in the same direction. However, neither has external reproduction, and the result replacing demonstration data came under a condition that mixed in 25% robot data. Externally reproduced real-robot performance remained at zero cases this week, and there were no new reports of field adoption for deterministic physics-based approaches (Realtime Robotics and others). NVIDIA GR00T N2's release has not been confirmed.

2. Company Watch — Listed

ABB (ABBN.SW)

• Confirmed facts → The China portion of the robotics divestment (enterprise value US$5.375 billion) was prohibited in a Chinese government review. The company is considering a sale of the China business to a third party (valued at US$2 billion to US$3 billion) or reviving a spin-off listing (citing Bloomberg). The expectation of closing within the year can no longer be maintained.
• Questions to verify → ① Is a carve-out structure agreed, and if so, how much is the deal price for the non-China business adjusted from the original US$5.375 billion? ② Does the sale price of the China business fall within the reported valuation range? ③ Does the delay affect the robotics business's orders and staff?

SoftBank Group (9984.T)

• Confirmed facts → In a notice on 11/6 it disclosed the Chinese authority's notification, and it is leading discussions on a plan to acquire the remainder after carving out the China business (early stage, citing Bloomberg).
• Questions to verify → ① Does it proceed with the acquisition after a carve-out, or renegotiate or withdraw? ② Does the company explain how it connects industrial robots and humanoid investments in its robot portfolio strategy?

Agility Robotics (AGLT)

• Confirmed facts → Began trading on Nasdaq on 11/5. Proceeds of about US$465 million (before costs). 2025 revenue US$1.78 million, net loss US$138.1 million, related-party revenue 64%. Digit 65,000 cumulative operating hours, Digit 5 conditional orders of more than US$300 million, early deliveries in 2027. Conflicting reports of the first-day close (US$11.51 versus US$10.83).
• Questions to verify → ① How much external customer revenue appears in the first quarterly report? ② Are the conditions and timing for converting Digit 5 orders into confirmed revenue disclosed? ③ Does the capital raised last until the 2027 delivery date against the quarterly cash burn? ④ Does the shareholder composition change before the lock-up expires (May 2027)?

Hyundai Mobis (012330.KS), Samsung SDI, LG Energy Solution

• Confirmed facts → Hyundai Mobis supplying all Atlas actuators, and Samsung SDI and LG Energy Solution supplying batteries, are reports based on industry sourcing with no official confirmation from the companies. The battery cell specification is unverified. Hyundai Mobis's prototype production from November and supply from 2028 were confirmed in last edition's results announcement.
• Questions to verify → ① Does Hyundai Mobis internalize reducers, motors and encoders, or select external partners? ② Are the battery suppliers and specifications confirmed in company disclosures? ③ Is the 2028 volume disclosed?

Rainbow Robotics (277810.KQ)

• Confirmed facts → It is selling 12 units of its wheeled dual-arm humanoid RB-Y2 this year, industrial safety certification is in progress, sample production has started, mass production is planned for around June next year and full-scale sales for mid next year. Payload per arm is 6 kg (double the previous model). AI middleware for operating the robot in natural language is under development.
• Questions to verify → ① Timing and scope of safety certification. ② Is the start of mass production in the first half of next year confirmed by disclosure? ③ Are the customers for the 12 units inside or outside the group?

ROBOTIS (108490.KQ)

• Confirmed facts → It targets annual mass production of 10,000 units of AI Sapiens by scaling up from next year (open-platform strategy). The most recently confirmed production capacity is around 10 units per month.
• Questions to verify → Is the capacity expansion investment needed to close the gap between the target and current capacity confirmed by disclosure or contract? Until then, 10,000 units is recorded as a target only.

Three listed Korean drive-train makers — SPG (058610.KQ), SBB Tech (389500.KQ), Samhyun (437730.KQ)

• Confirmed facts → All three unveiled integrated actuators or their components at RoboWorld 2026. Performance figures and mass-production timing are mostly undisclosed, and no orders tied to humanoid volume production were disclosed.
• Questions to verify → ① Is actuator (module) revenue disclosed separately from stand-alone reducer revenue? ② Are customers disclosed? ③ Which finished-robot makers do they contract with outside the volume where in-group supply is set?

Serve Robotics (SERV)

• Confirmed facts → Third-quarter revenue US$2.8 million (+313%, down from US$3.2 million the previous quarter), GAAP net loss US$59.2 million, cash US$198 million, average daily active robots 917. Full-year revenue guidance raised to US$12 million, cost guidance lowered to US$150 million. Physical AI Services launched.
• Questions to verify → ① First external customer and contract form for Physical AI Services. ② Is daily revenue per robot maintained as the fleet grows? ③ Does fourth-quarter revenue meet the raised guidance?

Amazon (AMZN)

• Confirmed facts → CNBC reported, based on an analysis of job postings, that a team for a new home robot product has been formed (single source). What the company has confirmed is the March 2026 acquisition of Fauna Robotics and the creation of the Personal Robotics Group; whether the new product is based on Fauna technology is unverified.
• Questions to verify → Are the product form and timing disclosed? If confirmed, it is recorded that companies preparing home humanoids face a competitor with distribution, payments and a voice assistant.

3. Company Watch — Private and Funding

A round in the US$100 million range entered the data layer from China, but the valuation was not disclosed. GenRobot AI's investor composition is one in which data demand-side companies hold equity in the supplier. Revenue and paying customer counts are undisclosed, so no multiple is calculated. Through last edition, data-layer rounds gathered around teleoperation and first-person human video, and this round is also on the human motion data side.

Zipline is the only large round with a disclosed valuation, but it lies outside this analysis's three areas. When reading the Crunchbase third-quarter tally (robotics startup investment of US$9.9 billion, more than double the US$4.3 billion a year earlier; January–September cumulative China US$13.2 billion, US US$12.1 billion), it is noted only that large drone and delivery deals lift the total.

Gap with listed comparables. From this week, humanoids have a listed comparable (Agility) whose revenue and profit and loss are disclosed. However, market capitalization is not in the briefing and no recent round valuations of private humanoid companies are in this briefing either, so the gap calculation is deferred until Agility's first quarterly disclosure and a private round appear in the same period. There is no basis to change last edition's conclusion that the listed-private gap is explained not by revenue but by layer (software versus hardware) and jurisdiction (US private capital versus China). Model-layer rounds have had zero disclosures of revenue or external verification for four consecutive weeks and will from now on be recorded separately as "capital signals" only. There were no new model-layer rounds this week; the LiberAI item is a model release, not funding.

Confirmation-date tracking

4. What to Watch

• SoftBank–ABB carve-out structure and the sale price of the China business — Why now: a transfer of ownership of a top industrial robot maker has been stopped by a Chinese review for the first time, and whether a carve-out happens will set the basic structure for later acquisitions of Japanese and European robot makers. Check: disclosures and press releases from both companies, announcements by Chinese authorities. If a carve-out is agreed and the sale price of the China business is disclosed, the price of China risk can be estimated from the China business's share of total value; if the carve-out fails and the deal collapses, it is recorded that acquisitions of robot makers with Chinese bases are difficult to complete without a prior carve-out.
• Agility Robotics' first quarterly report — Why now: humanoid revenue, operations and order conversion appear as disclosed figures for the first time. Check: the SEC quarterly report (10-Q) and results announcement. If external customer revenue grows from the 2025 level (about US$640,000, simple calculation) and the conditions for confirming Digit 5 orders are disclosed, these figures are used in valuation comparisons for private humanoid rounds; if the related-party share stays the same, it is recorded that humanoid revenue still circulates within investor-affiliated companies.
• Harmonic Drive Systems results (11/11) and Symbotic fiscal fourth-quarter results (11/23) — Why now: Korean and overseas component makers have unveiled integrated actuators but no humanoid volume orders have been disclosed, and in logistics automation the backlog reflection of last edition's Walmart commitment has not been confirmed. Check: orders by application and backlog in each company's results materials. If humanoid applications appear as figures, the start date of drive-train volume demand is moved earlier; if not, the 2027–2028 supply plans remain the baseline.
• Official company confirmation of the Atlas supply chain — Why now: the actuator, battery and camera suppliers have been reported only through industry sourcing. Check: disclosures or press releases from Hyundai Mobis, Samsung SDI, LG Energy Solution and Boston Dynamics, and Hyundai Mobis's selection of sub-component partners. If officially confirmed, the entry points for independent component makers in Atlas volume are narrowed to reducers, motors, encoders and camera modules; if different suppliers emerge, this week's report is downgraded to sourcing-stage information.
• Disclosure confirming Korean humanoid mass-production schedules — Why now: Rainbow Robotics announced mass production in the first half of next year and ROBOTIS a target of 10,000 units a year, but the gap against current unit sales (12 this year) and production capacity (around 10 per month) is large. Check: safety certification disclosures, capacity expansion disclosures and contracts. If certification and expansion are confirmed, Korean humanoid sales are tracked in hundreds rather than tens; if not, the figures are recorded as targets only.
• Release of LiberAI Liber-0 weights and external reproduction — Why now: the result that glove data replaces part of robot data came only from the company's evaluation. Check: the model release page, third-party benchmarks and papers. If external reproduction appears, the first external evidence is recorded for the world-model hypothesis; if performance reproduces much lower outside the company's own hardware, the cross-embodiment claim is removed from the hypothesis.
• GR00T N2 release within the year and Section 232 negotiation countries and items — Criteria from last edition maintained. Check: NVIDIA announcements and model documentation; Commerce Department and USTR announcements and Federal Register notices. There was no new information on either this week.
• Tracking "capital flow (model layer) versus bottleneck location (physical layer)" — This week: (a) model-layer rounds that disclosed revenue or external verification: zero (no new model-layer rounds at all); (b) overseas volume orders for drive-train components: zero (the C&M item is for domestic quadruped robots, amount undisclosed); (c) data-layer rounds: one (GenRobot AI, US$100 million, valuation undisclosed). Because (a) has been zero for four consecutive weeks, under the 10/19 edition's criterion, model-layer rounds will from now on be recorded separately as "capital signals" only and not used as a basis for judging demand or bottlenecks.

5. Data Reliability Notes

Citable

• Chinese authority's notification prohibiting the acquisition of ABB Robotics' China business (11/4), deal enterprise value of US$5.375 billion, China as the largest market with 90% of China sales produced locally (SoftBank notice, ABB statement; primary sources, multiple outlets).
• Agility Robotics merger closing (11/4) and start of trading (11/5), proceeds of about US$465 million and their composition, 2025 revenue of US$1.78 million, net loss of US$138.1 million, related-party revenue of 64%, Digit's 65,000 cumulative operating hours, 180-day lock-up (company press release, multiple outlets).
• Serve Robotics third-quarter revenue of US$2.8 million, GAAP net loss of US$59.2 million, cash of US$198 million, 917 average daily active robots, guidance changes (results announcement, primary source).
• Doosan Robotics third-quarter revenue of KRW 16.175 billion and operating loss of KRW 13.4 billion (disclosure).
• GenRobot AI Series A+ of US$100 million and investor composition; Zipline US$1.5 billion at a US$21.5 billion valuation (multiple outlets).
• RoboWorld 2026 scale (402 companies from 17 countries), Rainbow Robotics' 12 RB-Y2 sales this year and first-half mass-production plan, the integrated actuator unveilings by four Korean component makers (multiple outlets).
• Crunchbase third-quarter robotics investment of US$9.9 billion, January–September cumulative China US$13.2 billion and US US$12.1 billion (Crunchbase tally, 10/27).

Use with caution

• Conflicting: Agility's first-day close (US$11.51 versus US$10.83), first-half humanoid shipments by China's top three of about 19,000 units (conflicting estimates), Eka Robotics valuation (carried over from last edition).
• Single source: Yurobotics Series A of US$30 million (11/6), Amazon's new home robot product team (CNBC, based on job-posting analysis).
• Industry sourcing (unconfirmed by the companies): Hyundai Mobis supplying all Atlas actuators, Samsung SDI and LG Energy Solution batteries, camera module suppliers (Electronic Times).
• Per the company / company evaluation / company announcement: GenRobot AI's 1 million cumulative data hours and 30 partners; LiberAI's 36% zero-shot rate, glove-data combination score (75.6 versus 72.8) and own-versus-third-party performance (90.4 versus 62.1); C&M's double torque density; Boston Dynamics' 24% unloading speed improvement for the existing Stretch fleet; Agility's Digit 5 orders of more than US$300 million (conditional); ROBOTIS's 10,000-unit annual target; AeiRobot's ALICE target price of US$47,000.
• Unverified: the report that the reason for the Chinese prohibition is concern about robot operating data leaving the country (no official explanation from the authority); Atlas battery specification (70 Samsung SDI 2170 cells, pack of about 1.3 kWh); the China carve-out structure (early-stage discussion); ABB's China business valuation of US$2 billion to US$3 billion (reported at the review stage); whether Amazon's new product is based on Fauna technology; LiberAI weight release and external reproduction; GR00T N2 release; GenRobot AI's cumulative funding of more than US$200 million (as reported); Section 232 negotiation countries and items.
• Simple calculation: Agility's external customer revenue of about US$640,000 is calculated from disclosed revenue of US$1.78 million and the 64% related-party share, and is not a disclosed figure.

This article is industry analysis based on public information and is not a recommendation to buy or sell any security.