Maxinsights
Provides multimodal training and evaluation data for physical-AI and robotics foundation models.
Website: https://www.maxinsights.ai/
Cover Block
From the public record
| Field | Detail |
|---|---|
| Name | Maxinsights |
| Tagline | Provides multimodal training and evaluation data for physical-AI and robotics foundation models. |
| Headquarters | San Francisco Bay Area [robottoday.com] |
| Founded | 2024 [robottoday.com] |
| Stage | Seed [Crunchbase] |
| Business model | B2B [Crunchbase] |
| Industry | Deeptech [Crunchbase] |
| Technology | AI / Machine Learning [Crunchbase] |
| Geography | Global / Remote-First [Crunchbase] |
| Growth profile | Venture Scale [Crunchbase] |
| Founding team | Co-Founders (2): Kun Wang, Yuguang Yuan [Crunchbase, Retrieved 2026] [outx.ai] |
| Funding label | $10M+ [Crunchbase] |
| Total disclosed funding | ~$30,000,000 [Yahoo Finance, October 2026] |
Links
From the public record
- Website: https://www.maxinsights.ai/
- Crunchbase: https://www.crunchbase.com/organization/maxinsights
The Short Version
PUBLIC Maxinsights provides multimodal training and evaluation data for physical-AI and robotics foundation models, and it merits investor attention now because the company is trying to solve one of the more stubborn bottlenecks in robotics, namely the supply of large-scale, real-world human demonstration data [robottoday.com] [Forbes, October 2026]. Founded in 2024 in the San Francisco Bay Area, the company describes its wedge as collecting first-person recordings of everyday tasks and enriching them with hand and body tracking, language labels, tactile signals, and quality evaluation so the resulting dataset can be used to train physical-AI systems [robottoday.com] [maxinsights.ai].
The product surface appears broader than simple video capture: public materials point to multimodal data collection, motion capture, tactile and force sensing, simulation, and model or data evaluation, while the October 2026 financing announcement added MaxLoop, an agent-in-the-loop quality-control system, and MaxGlove, a sensor glove for touch and force capture [robottoday.com] [Yahoo Finance, October 2026] [01net.it]. That combination is the main differentiator visible in public sources, although most operating claims remain company-supplied and should be treated as directional rather than independently settled facts [Forbes, October 2026] [maxinsights.ai].
On team, public records identify Yuguang Yuan as co-founder and CEO, with reported prior experience at Waymo, Google, and Microsoft, while Crunchbase and RocketReach identify Kun Wang as co-founder and CTO with prior experience including QCraft, Waymo, and Google [Yahoo Finance, October 2026] [Crunchbase] [rocketreach.co]. The financing picture is clearer than the operating picture: a November 2025 seed round was listed with participating investors including HSG, Alumni Ventures, Top Harvest Capital, South Park Commons, Tola Capital, and Plug and Play, and Reuters and Yahoo Finance reported a $30 million seed round in October 2026 led by Peak XV, with participation from Tola Capital, South Park Commons, INCE Capital, and Alumni Ventures [robottoday.com, November 2025] [Reuters, October 2026] [Yahoo Finance, October 2026].
The business model is straightforward B2B infrastructure, selling data collection and evaluation capability to AI labs and robotics model builders that would otherwise need to stand up their own contractor, sensor, and annotation operations [robottoday.com] [Yahoo Finance, October 2026]. Over the next 12 to 18 months, the central questions are whether Maxinsights can convert publicity around scale into durable customer concentration across leading labs, prove that products such as MaxLoop and MaxGlove improve data quality rather than just expand product breadth, and show that its contractor-heavy collection engine can maintain consistency as demand grows [Forbes, October 2026] [AI Weekly, October 2026].
Unconfirmed -- This section relies materially on company statements and reprinted company claims, with funding corroborated by Reuters and Yahoo Finance but most product and operating metrics unconfirmed outside company-linked sources.
Taxonomy Snapshot
| Axis | Value |
|---|---|
| Stage | Seed |
| Business Model | B2B |
| Industry / Vertical | Deeptech |
| Technology Type | AI / Machine Learning |
| Geography | Global / Remote-First |
| Growth Profile | Venture Scale |
| Founding Team | Co-Founders (2) |
| Funding | $10M+ total disclosed, approximately $30,000,000 [Yahoo Finance, October 2026] [Crunchbase] |
The Company in Brief
PUBLIC
Maxinsights appears to have formed around a narrow but timely problem: robotics and physical-AI labs need large volumes of real-world human demonstration data, and the company is positioning itself as the supplier of that input rather than as a model builder itself. The company describes its work as providing production-grade egocentric and multimodal data for physical AI, while Crunchbase lists it as a San Francisco Bay Area company founded in 2024 [maxinsights.ai] [Crunchbase].
Publicly available source material is still thin on the formal company record. The website presents Maxinsights as a provider of annotated egocentric video, motion capture, tactile force sensing, simulation, and model evaluation for physical-AI systems [maxinsights.ai]. Crunchbase identifies Kun Wang as Co-Founder and CTO and records the company as venture-scale and seed-stage, but neither source in the record set here provides a legal entity name or state filing detail [Crunchbase].
The visible milestone sequence is short but clear. Crunchbase dates the company to 2024 [Crunchbase]. The company website now shows a broader product surface that extends beyond raw data capture into evaluation and simulation tooling [maxinsights.ai], which suggests the business is framing itself as infrastructure for model development rather than a single-service annotation shop, though that remains an interpretation from positioning rather than a separately confirmed operating history.
Single-source, plausible -- Confirmed by Crunchbase and the company website; legal entity details and state filings were not established in the cited public record.
What They Have Built
Mixed sourcing
Maxinsights is selling a data pipeline rather than a robot. Public descriptions are consistent on the core motion: the company collects first-person recordings of people performing real-world tasks, then layers on hand and body tracking, language labels, tactile signals, and quality evaluation so the resulting dataset can be used to train physical-AI and robotics foundation models [robottoday.com] [maxinsights.ai]. Forbes reported that contractors are equipped with head-, wrist-, and chest-mounted cameras while performing routine activities such as loading dishwashers, making smoothies, and washing cars, which gives some concrete shape to the otherwise broad category label of multimodal robotics data [Forbes, October 2026].
The product surface appears to have expanded in public at the time of the October 2026 financing. Yahoo Finance, carrying a company announcement, said Maxinsights introduced MaxLoop, described as an "agent-in-the-loop" system that scores and corrects data collection while it is happening, and MaxGlove, a sensor glove intended to capture tactile touch and force signals that camera footage cannot capture [Yahoo Finance, October 2026]. A separate Italian-language pickup adds a narrower operational detail: MaxLoop is meant to address difficult annotation conditions such as occluded hands, multiple hands entering frame, gloves or rings, and motion blur [01net.it].
The technical claim worth watching is breadth. On its website and in directory coverage, Maxinsights says it offers production-grade annotated egocentric video, hand tracking, upper-body and whole-body motion capture, tactile and force sensing, simulation, and model or data evaluation, positioning that suite as infrastructure for robotics, vision-language-action, and world-action models [maxinsights.ai] [robottoday.com]. That reads less like a single labeling tool and more like an outsourced data operations stack for labs that need real-world demonstrations at scale, although most of the performance claims remain company-sourced rather than independently benchmarked [Yahoo Finance, October 2026] [Forbes, October 2026].
Unconfirmed -- This section relies materially on company website statements and company-announcement coverage, with limited independent technical verification beyond descriptive reporting in Forbes and partial corroboration in 01net.it.
Market Size and Demand
From the public record
The market matters now because robotics labs and physical-AI model builders appear to be hitting a familiar constraint from earlier generative-AI cycles: model ambition is rising faster than the supply of usable training data [Forbes, October 2026] [Yahoo Finance, October 2026].
Public source material on Maxinsights is notably stronger on demand signals than on formal market sizing. No named third-party TAM, SAM, or SOM estimate appears in the supplied research, so the safer read is qualitative: the company sits at the intersection of robotics data infrastructure, multimodal AI tooling, and outsourced data operations for model development [maxinsights.ai] [robottoday.com]. That intersection is narrow enough to be specialized, but broad enough to pull budget from several adjacent lines, including foundation-model training, robotics simulation and evaluation, and annotation workflows that were previously handled in-house or through generalist data-labeling vendors [maxinsights.ai] [Yahoo Finance, October 2026].
The demand case rests on the difficulty of gathering real-world demonstrations at scale. Forbes reported in October 2026 that Maxinsights equips contractors with head-, wrist-, and chest-mounted cameras to capture first-person task data, and separately reported that the company claims more than 2 million hours of footage and more than 30 customers [Forbes, October 2026]. Even allowing for the fact that these operating metrics are company claims carried by media coverage, they point to a market need that is operational rather than theoretical: robotics teams need synchronized, labeled, quality-checked data from physical environments, and many do not appear set up to build that collection pipeline themselves [Forbes, October 2026] [Yahoo Finance, October 2026].
A second tailwind is the shift from vision-only training inputs toward richer multimodal datasets. Maxinsights describes its offering as combining egocentric video with hand tracking, upper-body and whole-body motion capture, tactile and force sensing, simulation, and model evaluation [maxinsights.ai] [robottoday.com]. Yahoo Finance reported in October 2026 that the company introduced MaxLoop for in-process scoring and correction, and MaxGlove for capturing tactile touch and force signals, which suggests that customers are no longer buying raw footage alone, but a higher-value data stack designed to improve model readiness and evaluation fidelity [Yahoo Finance, October 2026].
The closest adjacent markets are easier to identify than the market size itself. One adjacent market is traditional data labeling and annotation, where the buyer already understands outsourced workflow economics but may not yet require sensor-rich physical-world capture [robottoday.com]. Another is simulation and synthetic data for robotics, which can substitute for some real-world collection but does not eliminate the need for ground-truth demonstrations and evaluation data, especially where contact dynamics and human task variation matter [maxinsights.ai] [LinkedIn, Retrieved 2026]. A third is embodied-AI infrastructure more broadly, including tooling for evaluation, reinforcement learning environments, and world-action model development, all of which draw on similar budgets and technical priorities [robottoday.com] [maxinsights.ai].
Macro and regulatory considerations are present, but the public record here is still thin. On the macro side, the strongest force is continued capital formation around physical AI and humanoid robotics, reflected indirectly by Maxinsights' 2026 seed financing and by customer references to frontier labs and robotics model developers [Reuters, October 2026] [Forbes, October 2026]. On the regulatory side, the operating model described by Forbes and AI Weekly, which involves 5,000-plus contractors across six countries, implies exposure to cross-border labor, privacy, consent, and data-handling requirements, even though the supplied sources do not document any specific compliance framework or enforcement issue [Forbes, October 2026] [AI Weekly, October 2026].
| Market lens | Public evidence | Implication |
|---|---|---|
| Robotics training data | Maxinsights says it supplies multimodal training and evaluation data for physical-AI and robotics foundation models [robottoday.com] | Indicates a specialized supplier category forming around embodied-AI data pipelines |
| Multimodal model inputs | Product claims include egocentric video, motion capture, tactile sensing, simulation, and evaluation [maxinsights.ai] [robottoday.com] | Suggests spend may expand from annotation into full-stack data infrastructure |
| Outsourced collection operations | Forbes reports contractor-based first-person data capture and 30-plus customers claimed by the company [Forbes, October 2026] | Points to external demand from labs that prefer to buy collection capacity rather than build it |
| Evaluation and QA tooling | Yahoo Finance reports MaxLoop as an agent-in-the-loop scoring and correction system [Yahoo Finance, October 2026] | Implies the category may evolve from data supply toward workflow software and quality control |
The table supports a conservative conclusion: the addressable opportunity is best understood today as a convergence market, not a clean standalone category with widely published benchmarks. That does not weaken the demand signal, but it does mean investors should treat market size claims cautiously until third-party coverage of embodied-AI data infrastructure becomes more developed.
Single-source, plausible -- Section relies on credible public reporting and company materials, but no independent third-party market-sizing report was available in the supplied sources.
Who Else Is Fighting for This
MIXED Maxinsights is positioning itself less as a robotics application company and more as outsourced data infrastructure for labs training physical-AI models, which places it adjacent to foundation-model developers rather than in direct competition with them [Yahoo Finance, October 2026] [maxinsights.ai].
The public record is thin on named direct competitors, so the competitive map is clearer by segment than by company list. On one side are frontier model builders and robotics labs that can source data internally, including organizations named in Maxinsights-related coverage such as OpenAI, Google DeepMind, Meta, Figure, 1X, Skild, Genesis AI, and Dyna Robotics [robottoday.com]. Those groups are not necessarily competitors in a conventional product sense, but they are substitutes when they choose to build their own collection, annotation, and evaluation stack. On another side are adjacent data providers and tooling vendors, including simulation and evaluation tools, though the supplied sources do not name specific firms in those categories. That matters because Maxinsights is selling a bundled service, first-person capture, multimodal labeling, tactile sensing, and evaluation, into a buyer set that could unbundle the workflow over time [robottoday.com] [maxinsights.ai].
The company appears to have an early edge in operating scale, though the supporting evidence is still largely company-linked. Forbes reported in October 2026 that Maxinsights had captured more than 2 million hours of footage, with 1.5 million hours considered ready for modeling, and served more than 30 customers [Forbes, October 2026]. AI Weekly separately reported that the company employs 5,000-plus contractors across six countries [AI Weekly, October 2026]. If those figures hold, the advantage is not only dataset volume but operating machinery: recruiting workers, standardizing capture with head-, wrist-, and chest-mounted cameras, and layering hand tracking, motion capture, tactile signals, and quality control into a repeatable pipeline [Forbes, October 2026] [robottoday.com]. The durable part of that edge would be process knowledge and customer integration; the perishable part is that neither cameras nor contractor networks are structurally scarce if well-capitalized labs decide the margin should stay in-house.
The exposure is straightforward. Peak XV's led $30 million seed gives Maxinsights capital to expand, but the best-resourced alternative remains the customer itself, especially if large labs conclude that proprietary data collection is strategic and should not sit with a vendor [Yahoo Finance, October 2026] [Reuters, October 2026]. Genesis AI is a useful named reference point here. Company-linked material says Maxinsights supplied all 200,000 hours of first-person data behind Genesis AI's GENE-26.5, which suggests relevance, but it also underlines dependence on model companies whose own roadmaps could pull capability inward over time [outx.ai]. The same logic applies to Dyna Robotics, where Maxinsights is described as the largest data provider behind Dyna-2 in company-related coverage [outx.ai]. If buyers begin to treat multimodal data as a core moat rather than a procurement line item, vendor bargaining power can narrow quickly.
The most plausible 18-month scenario is a sorting of the category between neutral infrastructure suppliers and vertically integrated labs. Maxinsights is a potential winner if the market continues to reward specialization, meaning labs prefer to buy production-grade egocentric and tactile datasets from an external provider rather than stand up a global contractor operation themselves [maxinsights.ai] [Forbes, October 2026]. Genesis AI is a potential loser if that same condition holds and third-party data suppliers continue to improve quality control through products such as MaxLoop, because the data layer would become less proprietary at the model-builder level [Yahoo Finance, October 2026] [01net.it]. The reverse scenario is also visible in the public evidence: Maxinsights would be the loser if large labs decide that data collection, especially tactile and first-person interaction data, belongs inside the model company boundary. In that case, a neutral supplier can still survive, but more as overflow capacity than as the primary owner of a strategic layer.
Inferred, not confirmed -- Core positioning and funding are corroborated by company materials and named publisher coverage, but key competitive premises, including customer relationships, dataset scale, and relative standing versus named labs, rely materially on company-linked or single-source reporting.
Opportunity
Upside case
PUBLIC The prize here is straightforward: if physical-AI labs and robotics model builders continue to need large volumes of real-world, multimodal human demonstration data, Maxinsights has a path to become a core data infrastructure vendor in that stack, not just a services provider on the edge [Forbes, October 2026] [Yahoo Finance, October 2026].
The headline opportunity is to become a default external data layer for physical AI. That sounds ambitious, but the public evidence at least sketches the ingredients: the company is positioned around multimodal collection and evaluation for robotics foundation models, it says it has already captured more than 2 million hours of footage, and it introduced product surfaces that move it from raw collection toward workflow control and sensor capture, namely MaxLoop and MaxGlove [robottoday.com] [Forbes, October 2026] [Yahoo Finance, October 2026]. The difference matters. A vendor that only supplies footage can be replaced more easily than one that helps define collection quality, annotation standards, tactile capture, and evaluation loops across multiple customers.
The near-term paths to scale are still narrow, but they are visible enough to frame.
| Scenario | What happens | Catalyst | Why its plausible |
|---|---|---|---|
| Multimodal data utility for frontier labs | Maxinsights becomes a repeat supplier to leading AI labs and robotics model teams that need egocentric video, motion capture, and tactile signals at scale | The October 2026 $30 million seed round led by Peak XV gives the company capital to expand collection and productize workflows [Yahoo Finance, October 2026] [Reuters, October 2026] | The company says it serves more than 30 customers, and public descriptions consistently place it in multimodal robotics data rather than generic labeling [Forbes, October 2026] [robottoday.com] |
| Quality-control layer for physical-AI data ops | MaxLoop evolves from a product announcement into a control point for scoring and correcting collection as it happens, making Maxinsights harder to displace even when customers collect some data themselves | Launch of MaxLoop and its framing around real-time correction of edge cases in footage quality [Yahoo Finance, October 2026] [01net.it] | Public reporting already points to annotation and quality problems in real-world footage, including occlusions and motion blur, which supports demand for workflow software around collection, not only labor supply [01net.it] [Yahoo Finance, October 2026] |
| Tactile-data standard setter | MaxGlove or related sensor tooling helps the company win the tactile and force-sensing segment, where camera-only datasets are incomplete for manipulation-heavy robotics models | Product launch of MaxGlove and broader industry push toward richer robot training signals [Yahoo Finance, October 2026] [robottoday.com] | The company already markets tactile and force sensing as part of its offering, and public product messaging suggests it is trying to own a harder-to-source data modality [robottoday.com] [maxinsights.ai] |
The compounding mechanism, if it appears, would come from operations learning turning into product advantage. Every additional customer dataset can help the company refine capture protocols, quality thresholds, edge-case handling, and multimodal synchronization standards, while a contractor network spread across six countries could widen the set of tasks and environments it can collect against [AI Weekly, October 2026]. MaxLoop is the clearest early sign of that transition from labor-heavy data collection to software-mediated quality control, and MaxGlove points to a second layer where proprietary capture hardware could deepen integration with customers and increase switching costs [Yahoo Finance, October 2026]. This is not a classic consumer network effect. It is closer to an operational flywheel: more collection produces more knowledge about failure modes, which improves tooling, which can improve data quality and customer retention.
The size of the win is best framed through infrastructure analogies rather than direct public comps, because no named competitor or clean public peer is confirmed in the source set. Reuters and Yahoo Finance both confirm a $30 million seed round in October 2026, which suggests investors see a large enough category to fund ahead of proven financial scale [Reuters, October 2026] [Yahoo Finance, October 2026]. If the strongest scenario plays out, namely that Maxinsights becomes a preferred multimodal data and evaluation layer across frontier robotics labs, the company could plausibly grow into a strategic infrastructure asset for larger AI or robotics platforms. That could support venture-scale outcomes in the hundreds of millions to low billions of dollars in enterprise value (scenario, not a forecast), particularly because training-data bottlenecks in robotics are repeatedly identified in public coverage as a core industry constraint [Forbes, October 2026]. The evidence does not yet justify more precision than that, but the upside case is real: solve a scarce input for a fast-rising model category, and the supplier can matter as much as some model builders.
Single-source, plausible -- Section relies on Reuters and Yahoo Finance for funding corroboration, with product positioning and operating metrics partly supported only by company-linked or single-source coverage.
Sources
From the public record
[robottoday.com, November 2025] Maxinsights is a San Francisco Bay Area startup founded in 2024 that supplies multimodal training and evaluation data for physical-AI and robotics foundation models. | https://robottoday.com/suppliers-discovery/maxinsights
[Crunchbase, Retrieved 2026] Maxinsights - Crunchbase Company Profile & Funding | https://www.crunchbase.com/organization/maxinsights
[outx.ai] Funding News | https://www.outx.ai/linkedin-posts/funding-news-9bd027f4
[Yahoo Finance, October 2026] Waymo, Google and Microsoft Alums Raise $30M to Build the Data Layer for Physical AI | https://ca.finance.yahoo.com/news/waymo-google-microsoft-alums-raise-160000050.html
[Forbes, October 2026] There’s Not Enough Robot Training Data. This Startup Plans To Fix That | https://www.forbes.com/sites/johnkoetsier/2026/10/22/theres-not-enough-robot-training-data-this-startup-plans-to-fix-that/
[maxinsights.ai] Maxinsights | Data Foundations for Physical AI & Egocentric Data Vendor | https://www.maxinsights.ai/
[01net.it] Waymo, Google and Microsoft Alums Raise $30M to Build the Data Layer for Physical AI | https://www.01net.it/waymo-google-and-microsoft-alums-raise-30m-to-build-the-data-layer-for-physical-ai/
[rocketreach.co, Retrieved 2026] Kun Wang email address & phone number | RocketReach | https://rocketreach.co
[Reuters, October 2026] Maxinsights Raises $30 Mln Seed Funding Led By PeakXV | https://www.tradingview.com/news/reuters.com,2026:newsml_FWN4681RI:0-maxinsights-raises-30-mln-seed-funding-led-by-peakxv/
[AI Weekly, October 2026] MaxInsights Raises $30M Seed Led by Peak XV, Pays 5,000+ Contractors to Film First-Person Robot Training Data | https://aiweekly.co/alerts/maxinsights-raises-30m-seed-led-by-peak-xv-pays-5000-contractors-to-film-first
[LinkedIn, Retrieved 2026] Choongsoon Bae - General Partner, Top Harvest Capital | https://www.linkedin.com/in/choong-soon-bae-32b3134/
Articles about Maxinsights
- Maxinsights Captures 2 Million Hours of Human Motion for Robot Brains — The YC-backed startup, founded by Waymo and Google alums, is building a global contractor network to film and annotate everyday tasks for AI labs.