Fourier Intelligence

1.1 Identity, Footprint, and Business Model

Fourier presents itself as a full-stack robotics company whose public identity now spans both a general-robotics brand (“Fourier”) and a rehabilitation-focused sub-brand (“Fourier Rehab”). The official About and homepage materials say the company was founded in 2015, started from exoskeleton rehabilitation robots, and has since expanded into a broader portfolio that includes the GRx humanoid series, RehabHub, the MetaMotus Galileo system, and open-source tooling. The company’s own contact page anchors the operating footprint in Shanghai, Singapore, and Malaysia, while multiple official pages state that Fourier serves more than 2,000 institutions across more than 40 countries and regions. That combination matters strategically: unlike a pure humanoid startup, Fourier already has an installed base in rehabilitation and clinical settings, which it can use to validate hardware, collect data, and cross-sell newer humanoid products. Publicly, the company still frames the business as product-led rather than through detailed commercial metrics; no reviewed source disclosed revenue, gross margin, or ARR figures.[CO001, CO002, CO003, CO004, CO005, CO006]

1.2 Founders, Leadership, and Institutional Network

The public leadership picture is centered on founder Alex Gu, who is repeatedly identified as founder and CEO (or founder and group CEO), with Zen Koh named as co-founder and group deputy CEO in the 2022 Series D coverage. Public evidence does not surface a fully disclosed board or broader governance stack, so the company overview must rely on operating leadership rather than investor-governance detail. What is visible is the international institutional network Fourier has assembled around its rehabilitation franchise: the 2022 funding coverage described more than 1,000 institutional healthcare customers and 17 international research joint labs, while 2025 official announcements show Fourier Rehab signing MOUs with Brooks Rehabilitation in Florida and Kurage in France. Those partnerships reinforce that Fourier is not just a lab-stage robotics company; it already works with hospitals, rehab researchers, and therapy-system partners. However, the reviewed public record does not disclose board seats, debt facilities, or secondary transactions, which remains an important diligence limitation for later chapters.[CO009, CO010, CO011, CO027, CO028, CO029]

1.3 Capital Base, Valuation, and Disclosure

Fourier’s public financing arc is now clear enough to establish it as a growth-stage Chinese robotics company, even though the cap table remains opaque. The strongest confirmed round before the humanoid wave was the January 2022 Series D: RMB400 million led by SoftBank Vision Fund 2 with backing from Prosperity7 Ventures and Yuanjing Capital. Three years later, 36Kr, FL Cube, Sango Automation, and NetEase all reported that Fourier completed a nearly RMB800 million Series E in January 2025 with a syndicate that included Guoxin Investment, Pudong Venture Capital, Zhangjiang-affiliated funds, Prosperity7, and Junshan Capital. In late May 2025, Runyang’s planned strategic investment of up to RMB300 million implied an RMB8 billion pre-money valuation and gave the market a fresh unicorn anchor. The disclosure quality, however, is still venture- style rather than public-company grade: cumulative capital is reported in Chinese media at about RMB1 billion, but no official cap-table breakdown, liquidation stack, debt schedule, or current revenue metrics are publicly available.[CO012, CO013, CO014, CO015, CO016, CO036]

1.4 Product Arc and Commercial Positioning

Fourier’s public milestone arc shows a company deliberately extending a rehabilitation robotics base into general-purpose humanoids. The official About page places the launch of the humanoid project in 2019, the first-generation GR-1 prototype milestone in 2022, and the formal GR-1 introduction in 2023. Official GRx materials and the 2024 GR-2 launch announcement position the GR-1 as the industry’s first mass-produced humanoid robot and then upgrade the platform in GR-2 to 175 cm, 63 kg, and up to 53 degrees of freedom, with optional 12-DoF hands, tactile sensing, and a more developer-friendly control stack. By August 2025, Fourier shifted the narrative again with GR-3, its first explicitly care-centric humanoid, using softer industrial design and multimodal emotional interaction as differentiators. Support-site documentation also shows that Fourier publishes developer-facing robot documentation and the Fourier ActionNet dataset, underscoring the company’s “made for AI” posture. This is a meaningful differentiator versus pure concept-stage humanoid startups, although the reviewed public record still gives only partial visibility into unit economics and customer-level deployment scale.[CO018, CO019, CO020, CO021, CO022, CO023]

1.5 Milestone Snapshot and Adverse Frame

By the report date, Fourier’s public profile supports a strong momentum narrative: it has a ten- year company history, a rehab installed base, a 500-plus employee disclosure from 2024, a fresh 2025 funding round, new rehab partnerships, and a CES 2026 showcase for GR-3. Yet the same record also justifies caution. The only public valuation anchor is the Runyang pre-money disclosure rather than a fully negotiated priced-round post-money number; the oft-cited GR-1 “100-plus delivered” number relies on secondary Chinese reporting rather than an audited company ledger; and The Robot Report’s 2026 China outlook explicitly says competitive pressure and consolidation will persist as investors demand commercially meaningful applications. In other words, Fourier looks more credible than many humanoid startups because it starts from a rehab business and institutional network, but it still operates in a market where commercialization, cost pressure, and disclosure discipline will determine whether a premium robotics narrative turns into durable value creation.[CO015, CO017, CO025, CO026, CO031, CO036]

1.6 Exhibits

2.1 Market Boundary and Scope Definition

Fourier Intelligence's addressable market sits at the intersection of two structurally distinct but increasingly overlapping robotics sectors. The first is rehabilitation robotics: powered exoskeletons, therapeutic robots, and assistive gait devices sold primarily to rehabilitation centers, hospitals, and—via a new Medicare pathway—directly to home users. The second is humanoid and embodied AI robotics: anthropomorphic, full-body robots sold to industrial, healthcare, and consumer end-markets. Fourier participates in both via its GRx humanoid platform and its established Fourier Rehab portfolio. Excluded from Fourier's primary TAM are: surgical robots (separate reimbursement path and buyer base); non-wearable stationary rehabilitation devices (continuous passive motion, ultrasound therapy); and the broader professional service robotics market spanning logistics, hospitality, and cleaning robots (USD 31.11B in 2026 per Fortune, dominated by transport and logistics). Including broad service robotics as an additive TAM would be misleading—Fourier does not currently address logistics or hospitality applications. The International Federation of Robotics formally classifies powered exoskeletons as wearable robots within the medical robot taxonomy, distinct from industrial robots. Rehabilitation and non-invasive therapy robots are tracked as a dedicated sub-category of medical robots. This definitional precision matters: IFR's medical robot category saw a 91% unit-volume surge in 2024, while rehab and non-invasive therapy specifically rose 106%, demonstrating the acute acceleration within Fourier's core segment. Status-quo substitutes for robotic rehabilitation include: licensed physiotherapists and occupational therapists delivering manual therapy sessions; body-weight support treadmill systems (non-robotic); and standard orthotic devices including non-powered braces. The transition from these substitutes to robotic systems depends on evidence, reimbursement coverage, upfront cost acceptance, and therapist-facing workflow integration—all structural barriers that also define the adoption ceiling for Fourier's addressable market.[CM001, CM002, CM003, CM008, CM044]

2.2 Demand Drivers: Epidemiology, Demographics, and Healthcare System Pressure

The primary structural demand driver for rehabilitation robotics is a global epidemic of neurological and musculoskeletal conditions intersecting with an aging population that generates ever-larger volumes of post-acute rehabilitation need. The World Stroke Organization's 2025 Global Stroke Fact Sheet—drawing on GBD 2021 data—reports 93.8 million prevalent stroke cases globally in 2021, making stroke the second leading cause of death and third leading cause of combined death and disability globally. Stroke incident cases rose 70% between 1990 and 2021, with 87% of deaths and 89% of DALYs concentrated in lower- and lower-middle-income countries. The global economic cost of stroke exceeds USD 890 billion per year and is projected to nearly double by 2050. PAHO's estimates of rehabilitation need in the Americas show 366 million people in 2019 had health conditions amenable to rehabilitation—a 58% increase from 231 million in 1990—driven primarily by population growth and aging. The likelihood of rehabilitation need rises exponentially with age. This creates a secular demand tailwind for any device category that can scale stroke and neurorehabilitation throughput. The International Federation of Robotics identifies labor shortages in healthcare and the growing elderly population as the two dominant demand drivers for service robots in the professional medical space. Oliver Wyman frames the macro case at system level: without productivity intervention, global healthcare spending could rise from USD 11.8 trillion today to USD 23.1 trillion by 2040; AI and specialized robotics could remove up to USD 5.1 trillion of that projected cost increase annually under a breakthrough adoption scenario. A global caregiver deficit estimated at 15 million workers reinforces why hospital systems are evaluating robotic augmentation of therapy staff capacity. These epidemiological and workforce dynamics together define a structurally expanding demand base for rehabilitation robotics and adjacent care robotics.[CM004, CM005, CM006, CM007, CM009, CM010]

2.3 Core Market: Rehabilitation Robotics Sizing and Segmentation

Four independent analyst firms published rehabilitation robotics market estimates for 2026, yielding a wide but informative range. TBRC places the market at USD 2.22 billion in 2026 growing at 22.7% CAGR to USD 4.87 billion by 2030. Mordor Intelligence estimates USD 1.77 billion in 2026 at 17.1% CAGR to USD 3.89 billion by 2031. IMARC Group bases its estimate on a USD 1.82 billion 2025 market expanding at 16.53% CAGR. Fortune Business Insights offers a materially lower estimate of USD 579.5 million in 2026 at 16.36% CAGR, the result of a narrower scope (exoskeleton and therapeutic devices only, excluding some service and lab automation categories). These estimates are not reconcilable without reviewing the underlying methodologies; the USD 579.5M–USD 2.22B spread (3.8×) reflects definitional scope differences in whether assistive robots, prosthetics, and related services are bundled. The mid-range consensus—Mordor at USD 1.77B to TBRC at USD 2.22B—represents the most defensible TAM for Fourier's rehabilitation segment for 2026 planning purposes. Within this market, exoskeleton robots account for approximately 47.35% of type-based revenue (Mordor 2025 data), confirming their dominance relative to therapeutic and assistive device sub-categories. Geriatric users account for approximately 61.4% of demand. Rehabilitation centers hold 53.2% of end-user revenue; hospitals represent the second-largest institutional buyer. Homecare is the fastest-growing adoption channel at 27.1% CAGR, driven partly by the 2024 CMS exoskeleton reclassification. Asia-Pacific is the fastest growing geography at 21.2% CAGR, driven by rising stroke incidence and aging demographics—an important dynamic for a Shanghai-headquartered company like Fourier. North America retains the largest regional share at approximately 74.76% of 2025 revenue by Fortune's estimate. Fourier Intelligence is explicitly named as a key competitor in TBRC's rehabilitation robotics competitive landscape, confirming the company's positioning within the commercial market rather than as a research-stage participant. IFR data shows the 2024 unit-level surge (106% growth in rehab robot sales) is driven by professional end-users in clinical settings, which aligns with Fourier's 2,000-plus institutional customer base from chapter 1. The adjacent elderly care robotics market—including companion robots, mobile service robots, and robotic exoskeletons specifically for senior-care facilities—is sized at USD 0.39 billion in 2026 growing to USD 0.64 billion by 2031 at a 10.4% CAGR (Knowledge Sourcing). The broader AI-powered elderly care solutions market (including non-robotic AI) is projected to reach USD 7.38 billion by 2031 at 31.6% CAGR. These adjacent markets are accessible to Fourier's GR-3 care humanoid but represent distinct buyer communities (eldercare facilities, home health agencies) rather than the rehabilitation center and hospital buyers who anchor the core rehab robotics addressable market.[CM013, CM014, CM015, CM016, CM017, CM018]

2.4 Adjacent Market: Humanoid and Embodied AI Robotics

The humanoid robot market represents Fourier's longer-range growth vector and its most ambitious market-entry claim. Two large-sample analyst estimates place the global humanoid robot market at USD 10.69 billion (FMI) and USD 8.32 billion (TBRC) in 2026, with five-year CAGRs of 37% and 47–53% respectively. Healthcare is the leading application segment, with FMI assigning 28% of 2026 humanoid robot revenue to healthcare-related use cases—implying roughly USD 3.0 billion of addressable healthcare humanoid spend. Roland Berger's April 2026 report paints a longer horizon: projected operating cost of USD 2 per hour makes humanoids economically compelling versus human labor in regions facing structural labor shortages, with the OEM-level market potentially reaching USD 300 billion (baseline) to USD 750 billion (optimistic) by 2035. TrendForce projects Chinese humanoid output will rise 94% in 2026, with Unitree and AgiBot accounting for approximately 80% of shipments as the industry shifts from demonstration to delivery. BusinessWire cites Goldman Sachs data showing humanoid manufacturing costs fell 40% year-over-year—far ahead of the 15–20% decline previously modeled—with current robot costs ranging from USD 30,000 to USD 150,000. Unitree's R1 launched at USD 5,900 in July 2025, establishing a sub-USD 10,000 consumer price point. The FMI and TBRC estimates diverge by approximately USD 2.37 billion, reflecting different scope boundaries, geographic coverage, and vintage assumptions. Both are ahead of some independent industry views—Roland Berger notes very few productive use cases for humanoids exist today outside narrow industrial pilots, and the ecosystem (software, integration, safety certification) trails hardware maturity by three to five years. For Fourier specifically, the relevant SAM within the humanoid market is not the full USD 8–11 billion but rather the healthcare, rehabilitation, and care-assistance sub-segment where Fourier's GR-3 humanoid is positioned. On FMI's 28% healthcare share applied to the USD 10.69B total, that implies a USD ~3B healthcare humanoid TAM in 2026—but Fourier would be competing against global platform leaders (Figure, Agility, Unitree, Boston Dynamics) with substantially more capital and manufacturing scale. The GR-3's differentiation via care-centric design and a clinical installed base is a distinct angle, but market-share penetration in the near term will be limited. A conservative view of Fourier's serviceable share of the humanoid market is well below 5% through 2027.[CM022, CM023, CM024, CM025, CM026, CM027]

2.5 Payer, Buyer, and Budget Owner Landscape

Rehabilitation robotics procurement divides into three distinct buyer and payer archetypes, each with a different budget owner, decision timeline, and adoption trigger. Institutional buyers—rehabilitation centers and hospital outpatient departments—are the dominant channel (53.2% of market by revenue, per Mordor), funded by facility capital budgets with procurement decisions typically made by clinical operations leaders and finance committees. Reimbursement from government payers and private insurers indirectly enables these purchases by anchoring therapy session billing codes that fund device ROI. Home-use buyers are the fastest-growing but smallest segment. Medicare's 2024 Home Health Rule (effective January 1, 2024) reclassified powered exoskeletons within the brace benefit category, creating a reimbursement pathway for personal exoskeleton devices. CMS proposed payment at approximately USD 94,617 per device, with the Medicare beneficiary responsible for the standard 20% copay. Completion of training at a certified clinical site is a prerequisite for home use. Lifeward reports that when Medicare, VA benefits, and workers' compensation are combined, approximately 40% of U.S. spinal cord injury patients are now covered by payers that have reimbursed personal exoskeleton claims, up from effectively zero pre-2024. The CY2026 OPPS/ASC final rule updated outpatient hospital payment rates by +2.6%, affecting approximately 4,000 hospital outpatient departments and 6,000 ambulatory surgical centers. This incremental rate update supports overall institutional capacity to fund rehabilitation technology procurement, but the marginal impact on robot-specific budgets is indirect. Research and academic institutions represent a niche buyer channel for developmental platforms and data-generation partnerships—relevant to Fourier's open-source and developer-ecosystem strategy but not material to near-term revenue. Humanoid robotics procurement in 2026 is overwhelmingly pilot-driven and enterprise-funded. Deployments at BMW (Figure), Amazon and GXO (Agility Digit), and UBTECH's automotive OEM partnerships are enterprise capex decisions with one-to-three-year evaluation cycles. Healthcare humanoid pilots are even earlier stage; care humanoids like GR-3 have no established reimbursement pathway and must generate ROI through direct cost displacement of labor hours. This limits near-term institutional health system adoption to innovation or productivity budget lines rather than mainstream capex.[CM030, CM031, CM032, CM033]

2.6 Constraining Lens: Headwinds and Adoption Barriers

The bull case for rehabilitation and humanoid robotics rests on demographic tailwinds, unit cost declines, and policy catalysts. A responsible market analysis must weight these against material constraining forces. On clinical evidence: the 2026 JMIR umbrella review of robot-assisted therapy for upper-limb rehabilitation after stroke found improvements in motor function, but activities of daily living (ADL) gains were generally comparable to conventional therapy. This finding directly limits the strongest payer-approval argument—that robot-assisted therapy produces meaningfully superior functional outcomes. A 2026 PubMed RCT did find exoskeleton-assisted rehabilitation improved Fugl-Meyer lower-extremity scores and Barthel Index versus conventional training for post-stroke gait, supporting the lower-limb/gait-rehab case more clearly than upper-limb ADL. The evidentiary picture is thus mixed and segmented: stronger for lower-limb/gait exoskeleton use cases and weaker for broader ADL outcome claims. On commercial readiness for humanoids: the IFR's 2026 trends report explicitly states humanoid robots must match high industrial requirements in cycle times, energy consumption, and maintenance costs before they can compete with traditional automation—and that key gaps remain as of 2026. Robotics & Automation News identifies manipulation, power and endurance, and general-purpose autonomy as the three dominant technical bottlenecks; most 2026 deployments remain supervised pilot or co-pilot applications rather than autonomous operations. Roland Berger confirms the ecosystem (software integration, safety certification frameworks, regulatory clarity) trails hardware by three to five years. On cost barriers: rehabilitation robotics devices priced at USD 30,000–USD 150,000 face a procurement ceiling determined by reimbursement rates (roughly USD 94,617 per Medicare pathway), facility capital budgets, and institutional risk tolerance. Mixed long-term outcome data and high upfront cost together temper procurement in emerging markets and pediatric settings. Fortune Business Insights and Mordor Intelligence both flag high initial capex and integration costs as the primary market restraints. Privacy and data security concerns, plus anxiety about replacing human connection in care settings, represent soft barriers for AI-robotic eldercare adoption. Collectively, these constraints suggest that while the market size estimates are large and the growth rates compelling, the proportion of the stated TAM that is realistically addressable in the 2026–2028 window—particularly for a mid-size player like Fourier—is substantially narrower than headline figures imply.[CM034, CM035, CM036, CM037, CM038, CM039]

2.7 Exhibits

3.1 Fourier's competitor map is bifurcated, not singular

Fourier's public product surface makes clear that it is trying to bridge two markets rather than defend one narrow niche. The company still sells a dense rehabilitation stack through Fourier Rehab—more than 2,000 institutions in 40+ countries, a RehabHub layer integrating 30+ robot types, and multiple limb-specific systems—but it also markets GR-1 as a mass-produced humanoid assistant and GR-3 as a care-oriented social robot. That dual identity means the direct competitive set splits into two groups with different proof requirements. On the rehab side, the relevant public peers are Wandercraft, Ekso Bionics, CYBERDYNE, and Lifeward, all of which compete on clinical evidence, gait-training workflow, and in some cases reimbursement access. On the humanoid side, Fourier runs into UBTECH, Unitree, AgiBot, Figure, and 1X, where public discussion centers on shipment scale, price floors, production capacity, and pilot deployments. The practical implication is that Fourier cannot be benchmarked against a single "robotics" average. Hospitals and rehab buyers care about therapist efficiency, regulatory posture, and whether a device fits reimbursement or care protocols. Factory, logistics, research, and early household buyers care more about autonomy, manipulation, hardware reliability, and headline price. That makes Fourier's care-humanoid strategy interesting but also fragile: it inherits stronger care adjacency than industrial-first humanoid peers, yet it also inherits slower healthcare-style proof standards than consumer or factory robotics vendors.[CP001, CP002, CP003, CP004, CP006, CP007]

3.2 Rehabilitation competitors win on evidence, reimbursement, or specialized gait workflows

Fourier's strongest direct competition today remains in rehabilitation robotics, not general humanoids. Wandercraft is the closest strategic overlap because it now spans hospital gait training, home exoskeleton ambition through Eve, and an industrial humanoid extension via Calvin-40 and Renault. Ekso Bionics and Lifeward are more straightforward rehab rivals: both have explicit U.S. reimbursement signals around personal exoskeleton access, and both frame their offerings across clinical and home or community settings. CYBERDYNE remains more differentiated technologically, using bio-electrical signal control and offering medical, well-being, and labor-support variants, but it is less transparent in English-language public materials about recent global commercial scale. Fourier's advantage in this set is breadth of institutional rehab distribution. Its official materials show a larger multi-robot rehabilitation platform than any single-exoskeleton peer, which should help in therapist relationships, bundled sales, and cross-selling. The weakness is that U.S.-visible reimbursement and public payer proof are currently clearer for Lifeward and Ekso than for Fourier. Public price transparency is also thin: outside Medicare-linked personal exoskeleton reimbursement references, most vendors remain quote-driven. That opacity reduces the confidence of pure price-performance comparisons and pushes buyers toward vendor trust, evidence, and service footprint instead. Switching costs exist—clinician training, protocol redesign, software workflows, and payer familiarity—but they are not hard platform lock-in of the kind seen in surgical robotics, so no rehab incumbent is unassailable.[CP008, CP009, CP010, CP011, CP012, CP013]

3.3 Humanoid rivals reset the market on price, volume, or industrial proof—not on clinical fit

In humanoids, Fourier is not the scale leader and it is not the price leader. UBTECH is already presenting itself as a listed full-stack humanoid company with mass production, automotive partnerships, and explicit exposure to elderly care and business service. Unitree is the public price disruptor: G1 starts at US$13.5K on its own site, while external market summaries point to R1 at US$5.9K, far below the pricing logic implied by institutional rehabilitation hardware. TrendForce's 2026 view that Unitree and AgiBot together could capture nearly 80% of China shipments underscores how quickly Chinese vendors are industrializing the category. AgiBot's own materials and CES press claims show a broad product family and thousands of robots shipped, even if independent verification remains uneven. Western peers have a different posture. Figure's homepage now sells a home-help narrative for Figure 03, but the strongest public proof on record is still BMW: daily shifts, 90,000+ parts loaded, and 30,000+ vehicles affected in an automotive setting. 1X is even more clearly home first, emphasizing internal home testing, quieter hardware, passive safety, and teleoperation. That means Fourier's GR-1 and GR-3 are differentiated less by raw autonomy or manufacturing scale than by care adjacency. GR-3's soft-shell, touch-forward design and GR-1's human-like assistant framing are better suited to care or social-presence narratives than most industrial peers. But that is also Fourier's weakness: public evidence for scaled GR-series deployments, public pricing, and quantified ROI is materially thinner than the evidence available for Unitree's price floor, UBTECH's order book, or Figure's partner-validated production learning.[CP020, CP021, CP022, CP023, CP024, CP025]

3.4 Fourier's moat is distribution and care fit, but commoditization pressure is rising fast

The most defensible Fourier differentiator is not that its humanoids are most capable today; it is that Fourier is one of the few humanoid entrants already embedded in rehabilitation provider workflows. That matters because hospital and rehab procurement still depends on trust, training, service, and clinical comfort more than spectacle. A care-oriented humanoid layered onto an existing rehab channel is a more believable go-to-market wedge than a generic consumer or factory humanoid suddenly pivoting into eldercare. This is why Fourier's dual identity is strategically useful: the rehab business can seed relationships, datasets, and channel access that pure-play humanoid vendors do not yet have. The durability of that edge is still limited. Public competitor evidence suggests the humanoid category remains pilot-stage and safety-constrained, with dexterous manipulation, endurance, tactile sensing, and minimum-risk behavior still unresolved in real environments. Those limits help Fourier because they slow everyone down. But they do not solve the bigger commercial risk: once humanoid hardware becomes cheaper and more modular, care-adjacent software, service, and clinical workflow integration may be easier to copy than a true reimbursement moat. Manual therapy, conventional assistive devices, internal automation projects, and adjacent entrants like Wandercraft's Calvin-40 or China's industrial platforms remain credible substitutes or future threats. The near-term judgment is therefore nuanced: Fourier has a stronger care narrative than most humanoid peers, but weaker publicly demonstrated scale and weaker U.S. reimbursement proof than the best rehab incumbents.[CP036, CP037, CP038, CP039, CP040, CP041]

3.5 Exhibits

4.1 Capital structure is visible enough to anchor valuation, but not enough to underwrite cash generation

Fourier’s public financing trail now gives a credible capital map even though the company still withholds operating financials. The 2022 Series D brought in RMB400 million from SoftBank Vision Fund 2, Prosperity7, and Yuanjing, with management saying the money would accelerate go-to-market expansion, healthcare-robotics innovation, and global growth. The January 2025 Series E was reported at nearly RMB800 million and was explicitly tied to further humanoid development and broader commercial deployment. The most informative document is not a Fourier press release but Runyang Technology’s June 2025 filing, because it adds an RMB8 billion pre-money anchor, shows a planned RMB300 million check for 3.5060% post-dilution ownership, and discloses a simultaneous 3.0% ESOP top-up. That filing also shows an investorized board structure, with founders retaining six of eleven nomination rights. These are strong capital-structure signals, but they do not substitute for a balance sheet: the public record still says far more about who funded Fourier than about what the existing business earns, burns, or retains.[CI001, CI002, CI003, CI005, CI006, CI007]

4.2 The visible business model is quote-led institutional robotics, with only fragmented public pricing

Official materials show a mixed robotics stack rather than a single product SKU. Fourier markets RehabHub as a platform integrating more than 30 robot types, sells rehabilitation devices through contact-led pages, and frames GR-3 as a care-centric humanoid for enterprise settings rather than for retail buyers. That surface implies at least three economic layers—capital equipment, implementation or service, and some software or platform attachment—but the company never discloses the actual revenue mix across those layers. Public price discovery is similarly fragmentary. TMTPost says GR-3 is priced above RMB200,000, while a current Europa Satellite listing exposes structured product metadata at EUR106,344.61. On the rehab side, a Chinese procurement tracker shows specific upper-limb robot bids at RMB280,000 and RMB347,000 and larger rehabilitation-system awards above RMB1 million. Those signals are enough to show that Fourier is selling into institutional budgets, not consumer impulse purchases. They are not enough to infer realized ASP, discounting, attachment, or margins, because each visible price point reflects a different bundle, geography, or channel.[CI013, CI014, CI015, CI016, CI017, CI018]

4.3 Peer analogs imply usable gross margins are possible, but service and commercialization can still consume them

Because Fourier does not publish its own P&L, the best public way to reason about economics is through evidence-constrained analogs. Ekso Bionics reported 2024 revenue of US$17.9 million, gross margin of about 53%, year-end cash plus restricted cash of US$6.5 million, and US$9.8 million of operating cash use. Lifeward reported 2025 revenue of US$22.0 million, gross margin of 38.2%, and operating loss of US$19.7 million, while explicitly blaming lower volume, tariff pressure, freight, and under-absorbed manufacturing overhead for part of the margin drag. CYBERDYNE’s FY2026 revenue was ¥3,846 million with a ¥601 million operating loss despite an established balance sheet. Those comps matter because they show a recurring pattern: rehabilitation robotics can achieve respectable hardware gross margins, yet still require meaningful capital to fund sales cycles, service, reimbursement work, and product development. Fourier’s own public signals point the same way. TMTPost says manufacturing costs remain high, while patient-use exoskeleton analogs require 40 to 50 hours of training and still depend on payer approvals or institutional budgets. The implication is not that Fourier is unprofitable by definition; it is that its category remains operationally heavy enough that gross margin alone would not settle the cash-burn question.[CI017, CI019, CI024, CI025, CI026, CI027]

4.4 The financial verdict is promising demand and pricing power in niches, but too much opacity for clean underwriting

The evidence supports a plausible bullish case: Fourier has repeatedly raised large rounds, still commands institutional price points, has one credible public growth hint in overseas rehab revenue, and operates in adjacent reimbursement and rehabilitation markets where budgets can support high-value hardware. But the adverse case is stronger than management rhetoric admits. There is still no public revenue base, ARR, gross margin, working-capital profile, monthly burn, or cash runway. The latest valuation signal comes from a strategic-investment filing rather than from audited operating performance. The public price set is also internally heterogeneous: some rehab devices clear at a few hundred thousand renminbi while bigger systems and humanoid distributor listings land far higher, which makes bundle-normalized margin inference impossible. Peer analogs show that 38% to 53% gross margins do not prevent multi-million-dollar annual losses or cash burn. For underwriting, that means the right stance is not “Fourier has no economics,” but “Fourier has not disclosed enough economics to prove the current capital base converts into durable, internally financed growth.” Management must still open the books on realized pricing, segment mix, backlog, cash, runway, and customer concentration before a hard financial conviction is warranted.[CI016, CI024, CI029, CI031, CI035, CI040]

4.5 Exhibits

5.1 Fourier publishes enough architecture to show a platform, but GR-3 remains less transparent than GR-2 or N1

Public evidence shows Fourier as more than a single demo-shell humanoid company. The retained set covers three visible humanoid generations plus underlying components: GR-1 as the first mass-produced platform in the company narrative, GR-2 with detailed support-center architecture, and N1 as a smaller open-developer humanoid. GR-2 documentation is especially concrete, breaking the robot into head, torso, arms, hands, waist, and legs, specifying up to 53 degrees of freedom, self-developed FSA 2.0 actuators, and optional 12-DOF hands. The DexHand component page adds unusually specific low-level detail for a private robotics vendor, including actuator count, Ethernet control, load specs, and SDK support. By contrast, GR-3 is publicly positioned more through care-oriented interaction and CES demos than through a comparable engineering manual. Its retained public evidence supports a multimodal sensing and interaction architecture, but not the same depth of battery, payload, runtime, reliability, or field-service disclosure that would let an outside buyer treat it as fully spec-proven.[CE001, CE002, CE003, CE005, CE006, CE007]

5.2 The developer surface is real and unusually open, but still reads like an early integrator stack

Fourier’s public GitHub and support-center footprint is substantive. The company explicitly frames itself as contributing software, hardware designs, and research outputs to the open-source community, and the retained repo set maps a fairly complete embodied-AI loop: teleoperation for data capture, ActionNet for demonstration data, imitation-learning frameworks, URDF and MJCF robot models, Isaac Gym training, MuJoCo verification, and client libraries for external control. NVIDIA’s partner blog strengthens this picture by showing Fourier using sim-to-real workflows on GR-2 and reporting a transfer result on a floor-to-stand task after 3,000 iterations. But the stack is not yet frictionless. The teleoperation repo warns that first-time hardware setup is tricky and out of tutorial scope, and the public GRx client is explicitly alpha and unstable. That combination suggests genuine technical maturity inside the company, paired with an external developer experience that is better than marketing-only theater but not yet the polished, release-managed ecosystem associated with mature robotics platforms.[CE013, CE014, CE015, CE016, CE017, CE018]

5.3 Rehabilitation systems have the clearest public deployment proof and workflow fit

If the humanoid story is still partly forward-looking, the rehabilitation story is already commercial. Fourier’s rehab pages claim more than 2,000 institutions across 40-plus countries, and the retained RehabHub brochure plus customer-site pages give the chapter named deployment proof rather than abstract marketing language. Ascot describes a consultant-led multidisciplinary pathway with more than ten devices on site; STEPS details specific upper-limb, ankle, cycle, and exoskeleton modules; Hobbs lists a live installation of ArmMotus EMU, ArmMotus M2 Pro, and ExoMotus M4 in its outpatient setting. The brochure also describes the system logic behind these deployments: linked devices, digitally tracked training, force and position sensing, auto-generated reports, and a staffing model where one therapist can monitor multiple patients. This is the strongest evidence in the chapter that Fourier has moved beyond demo-ware. The public record supports real rehabilitation productization, concrete therapist workflows, and meaningful implementation know-how, even though product-level outcome benchmarks and unit economics remain under-disclosed.[CE030, CE031, CE032, CE033, CE034, CE035]

5.4 Fourier looks commercial in rehab and technically serious in humanoids, but not yet publicly de-risked for care-robot scale

The retained evidence supports a split verdict. On one side, Fourier has a credible technical moat made of rehab channel access, component-level verticalization, and a real developer stack. On the other, its public humanoid commercialization proof is still thin compared with the rehab business. The strongest retained humanoid commercialization source is a 36Kr report about hundreds of shipped GR-series units and a hospital demonstration-base partnership; the strongest GR-3 sources remain CES launch materials and showcase reporting. That is enough to say Fourier is not vaporware, but not enough to say GR-3 is already deployment-proven in hospitals, eldercare, or large paid enterprise fleets. The biggest public gaps are exactly the ones that matter for underwriting: named production customers, detailed compliance or medical-license evidence for care scenarios, and documented security or privacy controls around networked robot operation. The right chapter judgment is therefore asymmetric: Fourier appears commercially real in rehabilitation and technologically credible in humanoids, but still closer to pilot-stage commercialization than scaled operating proof on the care-humanoid side.[CE041, CE042, CE043, CE044, CE045, CE046]

5.5 Exhibits

6.1 Rehabilitation buyers are visible; payer and budget authority still sit with provider-side workflows

Fourier’s public customer map is not a single homogeneous buyer segment. The best-documented users are rehabilitation providers and hospital rehab services that buy technology as part of therapist-led care delivery, not as standalone consumer devices. In those settings, the institution is the buyer, therapists and rehab teams are the daily users, and patients are the beneficiaries inside an existing program of care. Public site pages repeatedly frame value around measurable training, workflow fit, therapist productivity, and data capture rather than headline hardware novelty alone. That matters because payer logic sits one layer above the robot: CMS documentation rules and NIH reimbursement guidance both show that medical-device adoption still has to fit a billable therapy pathway, documented plan of care, and internal provider ROI case. The result is a buyer stack where rehab directors, innovation leads, procurement or finance committees, and clinical documentation owners all matter. On the edges of that stack, Fourier also shows research-lab buyers, distributor-led private clinics, and early embodied-care pilots, but those segments have much weaker public payment and renewal visibility than the rehabilitation core.[CU005, CU006, CU007, CU008, CU009, CU043]

6.2 Named rehab-site proof is real and geographically repeatable, but still sample-based

The most credible customer evidence comes from named rehab sites and hospital clusters that specify devices and use cases. NHG Health says Fourier technologies have already been deployed and evaluated at Tan Tock Seng Hospital and expanded the relationship into a new five-year MOU plus a Joint Rehabilitation Innovation Hub. Brooks Rehabilitation says ArmMotus EMU and ExoMotus M4 are now in place at its Center for Innovation in Florida. In the UK, STEPS, Ascot, and Hobbs each provide site-level proof that goes beyond logo walls: STEPS lists a broad RehabHub suite and outcome reporting, Ascot describes a 10-plus-device consultant-led pathway, and Hobbs names its installed device stack at Laverstock Care Centre. India adds Saifee Hospital’s ExoMotus M4 installation, while Ren Ci in Singapore contributes the strongest quantified site-level outcome evidence through an ExoMotus adjunct pilot study. Repeatability is therefore real across multiple countries and institution types. But the public record still reads as a sample, not a reconciled account list, and several important references—such as Nagoya or Shirley Ryan—sit closer to research and ecosystem proof than to disclosed recurring commercial accounts.[CU010, CU011, CU012, CU013, CU014, CU016]

6.3 Public proof quality is highest for rehab workflows and weakest for revenue durability

The chapter’s evidence quality is uneven in a way that matters for underwriting. Rehab-site pages and hospital studies can prove real use, device mix, and some operational fit. NHG and Brooks are stronger than pure press-release logos because they describe deployment or in-situ evaluation. STEPS and Ren Ci go further by describing usage, session reporting, or measured outcomes. Yet even the best public references usually stop short of what investors need for durability: contract length, reorder history, renewal timing, NRR, GRR, churn, attach rates, and site-level utilization curves remain undisclosed. That means the public record supports a verdict of real adoption but not necessarily sticky revenue. The quality gap widens further outside core rehab. Nagoya, Kurage, and HTPG are strategically useful ecosystem signals, but they should not be confused with end-customer proof. They show channel development, R&D collaboration, and bundled-solution ambition more than they show booked demand. Public customer evidence therefore establishes Fourier as commercially real in rehab while still leaving unanswered whether that installed base reliably expands, renews, and cross-sells at attractive unit economics.[CU015, CU018, CU025, CU026, CU032, CU033]

6.4 Adoption friction comes from capital approval, workflow integration, channel opacity, and a missing humanoid customer roster

The main customer risk is not that Fourier lacks any real-world adoption. It is that public adoption data does not yet let an outsider separate durable revenue from an impressive but blended ecosystem story. Rehabilitation buyers still have to clear documentation, coding, staffing, and capex logic, so sales cycles are likely anchored in provider economics and clinical validation rather than pure feature excitement. Channel partners such as Thor, HTPG, Robocare, and Healthlink can accelerate reach, but they also reduce visibility into end-customer identity, service obligations, and margin capture. That opacity becomes more serious when public named-customer disclosure is already thin relative to the 2,000-plus-institutions headline. The biggest proof gap sits on the humanoid and care side. Independent adverse analysis says there are no publicly confirmed paid GR-1 or GR-3 deployments with named customers, KPIs, or ROI, and public care-setting references still look like demonstrations, pre-sales, or innovation partnerships rather than scaled paid production. Until Fourier discloses renewals, customer concentration, and named humanoid pilots with measurable outcomes, the customer story should be treated as rehab-led with material conversion and concentration uncertainty layered on top.[CU007, CU008, CU022, CU037, CU038, CU039]

6.5 Exhibits

7.1 Rehab medtech risk is about reimbursement, compliance, and postmarket discipline—not whether a product exists

Fourier’s rehabilitation business is no longer a zero-to-one proof story. Brooks and Fourier’s own materials support a real installed product set, and the company publicly anchors that story on a 2,000-plus / 40-country footprint. That lowers existential product risk relative to many humanoid startups, but it does not remove medtech-style regulatory and reimbursement risk. In Europe, AI-enabled medical devices still enter through MDR today and then absorb AI Act obligations on data governance, transparency, cybersecurity, human oversight, and post-market monitoring as the regime phases in. In the United States, FDA guidance makes cybersecurity part of device compliance, while recalls, corrections, removals, and vigilance regimes matter if patient-adjacent systems fail in the field. On the demand side, CMS still ties therapy economics to documented medical necessity and correct coding rather than robot novelty. RehabHub’s workflow and data promises may help providers, but investors still need proof that those benefits convert into reimbursable utilization, renewal quality, and durable site economics.[CR002, CR003, CR004, CR005, CR006, CR010]

7.2 The humanoid thesis is constrained more by missing operating proof and privacy/liability exposure than by awareness

The humanoid risk stack is different from the rehab-medtech stack. Public sources can show GR-3’s design intent, CES visibility, care-oriented interaction features, and Fourier’s ambition to move from research and rehab into more human-centered environments. They do not yet show the kind of named paid deployments, disclosed regulatory pathway, or field reliability record that would let an investor underwrite eldercare or hospital-adjacent scale. Independent trackers and adversarial coverage both describe the humanoid line as pre-commercial or unproven, with no publicly confirmed paid GR-1 or GR-3 deployments, no named ROI cases, and limited channel access for the higher-end GR-2. That matters because human-facing care robots blend physical safety, autonomy, privacy, and liability into one package. Face, voice, touch, and emotional-cue processing can create GDPR and biometrics exposure; any move from demo spaces into assisted settings also sharpens responsibility questions across manufacturer, operator, integrator, and cloud/software provider. The right read is not “low risk because no scandal surfaced.” It is “high-consequence risk because the public marketing surface currently outruns the disclosed compliance pack.”[CR007, CR008, CR026, CR027, CR028, CR029]

7.3 Export-control, cost, and competitive intensity could compress the window for a premium healthcare-humanoid strategy

Even if Fourier solves product readiness, it still has to survive a geopolitical and competitive regime that increasingly rewards scale, lower price points, and compliance depth. U.S. export controls keep advanced China-related compute under screening, license, and end-use discipline, and even the 2026 BIS softening for specific chips remains conditional on security controls, compliance procedures, and third-party testing. China’s own humanoid standard system raises the local bar further, extending safety, ethics, application, and data/model governance expectations across the lifecycle. At the same time, the commercial window is getting tighter. TrendForce sees China output growing 94% in 2026, with Unitree and AgiBot approaching 80% share, while TechCrunch and CNBC describe a market moving from demo-driven excitement toward operations-driven adoption under heavy cost, software, safety, and bubble pressure. That is uncomfortable for Fourier because its premium angle depends on healthcare credibility, yet public evidence still shows limited humanoid customer proof and estimated price points far above lower-cost alternatives with broader availability.[CR030, CR031, CR032, CR033, CR034, CR035]

7.4 The correct monitor is not “no controversy yet” but whether Fourier converts rehab credibility into disclosed compliant deployments

The monitor set should focus on whether Fourier converts rehab credibility into disclosed, compliant operating proof. A named paid humanoid deployment with clear task metrics would matter more than another conference appearance. A published model-by-model FDA, CE, or NMPA strategy would matter more than another lifestyle care-bot demo. A disclosed security packet, DPA, incident-response process, and post-market plan would matter more than generic AI language. And on the rehab side, hard utilization and reimbursement evidence from flagship sites would matter more than top-of-funnel institution counts. That framing also preserves an important nuance around public controversies: this chapter did not retain a smoking-gun enforcement action or public recall naming Fourier, but the absence of a surfaced controversy is not the same thing as a de-risked safety or compliance profile. Public surveillance systems and recall rules exist; the missing piece is Fourier-specific disclosure showing how its quality, cyber, privacy, and regulatory controls would behave at scale. Until that pack exists, risk should be monitored through concrete triggers rather than comfort taken from silence.[CR020, CR021, CR025, CR027, CR041, CR042]

7.5 Exhibits

8.1 The public anchor is useful, but it is not strong enough to underwrite a conviction entry on its own

The right read on Fourier is not that the RMB8 billion pre-money anchor is obviously wrong; it is that public evidence is too thin to know whether it is cheap, fair, or expensive with high conviction. Runyang’s filing gives this chapter something many private robotics stories lack: a real filed price anchor, a check size, and post-dilution ownership math. That matters. The January 2025 Series E also shows there was fresh capital appetite behind a broad investor set, including state-linked Shanghai capital and well-known private backers. But the same public record also shows what is still missing. The valuation anchor arrives without disclosed revenue, gross margin, burn, cash balance, or runway. Even the fully diluted math needs interpretation because the investment sat alongside a simultaneous ESOP top-up and a strategic context rather than a plain-vanilla arm’s-length financing that cleanly prices common equity for outside investors. That combination leads to a disciplined conclusion: treat the filing as a ceiling test for new money, not as proof that the business already earns an RMB8 billion right to exist on operating fundamentals. For strategic insiders with private diligence access, the current level may still be fair. For outside investors limited to public evidence, the recommendation is effectively research-more or track, with medium confidence and a high risk rating.[CV001, CV002, CV003, CV004, CV005, CV006]

8.2 Public rehab peers frame the floor, while private humanoid rounds frame the narrative ceiling

Fourier’s valuation problem is that its public analogs point in two very different directions. On one side, listed rehabilitation and exoskeleton names such as Ekso and Lifeward show how unforgiving the public market can be when revenue is modest, losses persist, and investors can directly see the numbers. Those companies demonstrate that real products, reimbursement progress, and genuine commercialization do not automatically translate into rich valuation multiples. On the other side, CYBERDYNE and especially UBTECH show that robotics businesses can command materially richer multiples once they are listed, financially visible, and perceived as broader platforms rather than narrow rehab-device vendors. Private humanoid rounds then widen the gap even further. Figure’s tracked $39 billion Series C and Apptronik’s roughly $5.3 billion to $5.5 billion February 2026 valuation show how aggressively U.S. embodied-AI stories can be priced when investors are underwriting platform optionality rather than current economics. CNBC’s reporting makes the mechanism explicit: U.S. startups are being valued as AI platforms while Chinese ones are more often valued as industrial hardware businesses, even when Chinese firms are shipping more hardware. Fourier’s public anchor therefore sits between two worlds. It is too high to justify using distressed public rehab comps alone, but far too low to be read as a Figure-style frontier-humanoid narrative valuation. That is why the cleanest interpretation is a blended rehab-commercialization-plus-option-value framework rather than a single-sector comp shortcut.[CV008, CV009, CV010, CV011, CV012, CV013]

8.3 Bull, base, and bear framing puts the current anchor near the top of what public evidence can support today

An evidence-constrained scenario model is more honest here than a spreadsheet-heavy DCF. The bear case assumes the rehabilitation business is real but not large enough to carry the valuation, while the humanoid program remains strategically interesting without converting into named paid deployments or disclosed economic proof. In that world, public-multiple compression and fresh-round discounting could push fair value toward roughly RMB4.5 billion to RMB6.0 billion. The base case assumes public evidence is directionally right about Fourier’s rehab depth and that humanoids retain meaningful option value, but still gives no credit for economics the company refuses to disclose. That yields a range of roughly RMB6.5 billion to RMB8.5 billion, which is notable because the current public anchor already lives near the upper end of that band. The bull case needs more than optimism. It requires disclosed revenue and margin quality, evidence that rehab demand is durable rather than logo-heavy, and named humanoid deployments that narrow the comparability gap to higher-multiple robotics names. Only then does a RMB9.5 billion to RMB13.0 billion range become reasonable. This framing also explains why a 3x venture-style return hurdle is so hard to defend from the current starting point. A roughly RMB24 billion exit is not impossible, but public evidence today does not justify modeling it as the base case.[CV024, CV025, CV026, CV027, CV028, CV029]

8.4 The chapter’s decision rule is simple: do not pay above the anchor until missing operating facts turn into disclosed evidence

The most important valuation discipline here is procedural. Fourier does not need another narrative deck to improve the call; it needs a small set of operating disclosures that would let investors place it inside a real valuation regime. The first gate is financial visibility: revenue, gross margin, burn, and runway. The second is financing quality: whether the Runyang deal actually closed on the publicly described terms and what preference stack or anti-dilution protections sit above common equity. The third is commercialization quality: named paid humanoid deployments, customer concentration, renewal behavior, and evidence that rehabilitation sales are durable rather than merely broad. Until those facts are visible, the recommendation should stay price-sensitive and evidence-sensitive. That means no conviction underwriting at a premium to the public anchor, and no attempt to force a false-precision DCF onto an opaque private company. The upside case is real enough that the company stays investable in principle, especially because the rehab business gives it a floor many humanoid startups do not have. But the public record still supports waiting for either better disclosure or a more compelling entry price rather than treating the current anchor as self-validating.[CV036, CV037, CV038, CV039, CV040, CV041]

8.5 Exhibits