Bagel Labs

Cover Block

From the public record

Attribute	Details
Name	Bagel Labs
Tagline	A physical AI research lab developing world-action models for robot control. [Wellfound, retrieved 2026]
Headquarters	San Francisco, California, United States [Crunchbase, retrieved 2026]
Founded	2023 [Perplexity Sonar Pro Brief, retrieved 2026]
Stage	Seed
Business Model	B2B
Industry	Deeptech
Technology	AI / Machine Learning
Growth Profile	Venture Scale
Founding Team	Solo Founder
Funding Label	Seed (total disclosed ~$3,000,000)

Links

From the public record

• Website: https://angel.co/jobs
• LinkedIn: https://www.linkedin.com/company/bagel-labs

The Short Version

From the public record Bagel Labs is a physical AI research lab building world-action models for robot control, a technical bet that has attracted a $3 million seed round from a syndicate of crypto-native and traditional venture funds [Perplexity Sonar Pro Brief, Jan 2024]. The company was founded in 2023 by Bidhan Roy, who also founded the Bagel Network, a decentralized data platform for machine learning [Bloomberg Markets] [LinkedIn]. Its core technical approach involves Distributed Diffusion Models (DDM), a method for training frontier diffusion models across commodity hardware, which it first proved with a public model called Paris before applying the technique to physical AI [Bagel Labs Careers].

The go-to-market strategy targets a technically sophisticated but narrow audience: robotics companies, autonomy teams, simulation platforms, and research groups that need infrastructure to train, evaluate, and deploy models [Perplexity Sonar Pro Brief]. This focus on early-stage AI startups and academic researchers suggests a long-term, research-driven commercialization path rather than a rapid, broad-market product launch [Cerebral Valley, Nov 2024]. The company's current hiring push for a Head of GTM and research staff indicates a dual focus on advancing its core technology while beginning to structure its commercial outreach [Perplexity Sonar Pro Brief].

For investors, the next 12-18 months will be defined by Bagel Labs' ability to translate its DDM research into validated use cases with named partners in robotics or simulation, and to clarify its strategic relationship with the broader Bagel Network ecosystem. Single-source, plausible -- Core company description and funding round corroborated by multiple sources; founder background and product claims are from primary sources but lack independent verification.

Taxonomy Snapshot

Axis	Classification
Stage	Seed
Business Model	B2B
Industry / Vertical	Deeptech
Technology Type	AI / Machine Learning
Growth Profile	Venture Scale
Founding Team	Solo Founder
Funding	Seed (total disclosed ~$3,000,000)

The Company in Brief

From the public record

Bagel Labs was founded in 2023 as a physical AI research laboratory, an identity it has maintained through its public communications [Bagel Labs Careers, retrieved 2026]. The company describes its core mission as developing world-action models for robot control, a focus that has remained consistent since its inception [Bagel Labs Careers, retrieved 2026]. Its sole founder and CEO, Bidhan Roy, has been listed in that role since April 2023 [Perplexity Sonar Pro Brief, retrieved 2026].

The company is headquartered in San Francisco, California [Crunchbase, retrieved 2026]. The primary public milestone for Bagel Labs is a $3 million seed financing round announced in January 2024, led by CoinFund [Perplexity Sonar Pro Brief, Jan 2024]. This capital appears to have supported the initial research phase, including the development and public release of a model named Paris to prove its Distributed Diffusion Models (DDM) approach [Bagel Labs Careers, retrieved 2026]. Current activity is signaled by active hiring for a Head of GTM for Physical AI and several research staff positions, indicating a transition towards commercialization and continued technical development [Perplexity Sonar Pro Brief, retrieved 2026].

Confirmed across multiple sources -- Confirmed by Crunchbase, company careers page, and a dated funding report.

What They Have Built

Mixed sourcing

Bagel Labs frames its core product as a research-driven infrastructure layer for physical AI. The company describes itself as "a physical AI research lab developing world-action models for robot control," a formulation that positions its output as foundational models rather than end-user applications [Bagel Labs Careers, retrieved 2026]. Its go-to-market language targets technically sophisticated buyers, specifically naming robotics companies, model teams, simulation platforms, and research groups as the entities it aims to help train, evaluate, and deploy models [Bagel Labs Careers, retrieved 2026]. This suggests a product surface that could include model training services, evaluation tooling, or deployment frameworks, though the specific commercial packaging is not detailed.

The primary technical approach centers on Distributed Diffusion Models (DDM). The company states it is building DDM to train frontier diffusion models across commodity hardware, having proven the approach with its first public model, named Paris [Bagel Labs Careers, retrieved 2026]. The application of this distributed training architecture is now directed toward "physical AI for robotics, autonomy, and simulation teams" [Bagel Labs Careers, retrieved 2026]. The research focus on "world-action models" implies an ambition to create AI systems that can understand physical environments and generate appropriate control sequences for robots, a significant step beyond pure perception or planning.

Current hiring provides the clearest window into the technical stack and immediate priorities. Open roles for Members of Technical Staff in Research and Distributed Training Systems indicate a continued heavy investment in core model development and the underlying systems to support it (inferred from job postings). The description for the research role mentions work on "frontier diffusion models across commodity hardware rather than one uniform GPU cluster," reinforcing the DDM thesis and a focus on computational efficiency [Bagel Labs Careers, retrieved 2026]. The simultaneous search for a Head of GTM for Physical AI signals an intent to commercialize these research outputs, though the exact product form factor,whether a managed service, a software library, or a model-as-a-service API,remains [PUBLIC] unspecified.

Single-source, plausible -- Core product claims are from the company's own careers page and public description. Technical approach is stated but not independently verified by third-party technical reviews. Hiring inferences are based on public job descriptions.

Market Size and Demand

From the public record The push to imbue machines with a practical understanding of the physical world represents a critical, unsolved bottleneck in robotics and automation, moving beyond perception to actionable control. Bagel Labs targets the infrastructure layer for this transition, aiming to serve the teams building the next generation of autonomous systems.

Quantifying the total addressable market for physical AI model training is challenging, as it sits at the intersection of several larger, adjacent sectors. The broader industrial robotics market, a key downstream application, was valued at $16.8 billion in 2023 and is projected to grow to $35.3 billion by 2030, according to a Precedence Research report [Precedence Research, 2023]. For a more direct analog, the global market for AI in computer vision, which underpins much of the perception side of robotics, was estimated at $17.4 billion in 2023 and is forecast to reach $50.9 billion by 2030 [Grand View Research, 2023]. These figures provide a sense of the scale of the application markets Bagel Labs' technology intends to enable.

Demand is driven by several converging tailwinds. The proliferation of affordable sensor data and simulation environments has created vast datasets for training, while advances in generative AI, particularly diffusion models, offer new architectural approaches for learning complex, sequential tasks. Concurrently, industries from logistics to manufacturing face persistent labor shortages and are actively seeking flexible automation solutions, creating a pull for more capable and general-purpose robotic systems. Bagel Labs' focus on distributed training across commodity hardware, as cited on its careers page, directly addresses a key constraint: the soaring cost and limited availability of concentrated, high-end GPU clusters needed for frontier model development [Bagel Labs Careers].

Key adjacent and substitute markets include traditional robotic control software, bespoke model development services, and proprietary simulation platforms from large incumbents like NVIDIA (Isaac Sim) or startups like Covariant. The regulatory landscape remains nascent but is a watch item, as physical AI systems deployed in safety-critical environments will inevitably attract scrutiny around operational safety, liability, and data privacy standards.

Industrial Robotics (2023) | 16.8 | $B
AI in Computer Vision (2023) | 17.4 | $B
Industrial Robotics (2030 est.) | 35.3 | $B
AI in Computer Vision (2030 est.) | 50.9 | $B

The projected growth in these enabling and application markets suggests a substantial runway for infrastructure providers. However, Bagel Labs' serviceable market is currently a narrow slice, focused on technically sophisticated robotics companies, research groups, and simulation platforms, as described in its hiring materials [Perplexity Sonar Pro Brief].

Single-source, plausible -- Market sizing figures are drawn from third-party analyst reports for analogous sectors, not a dedicated physical AI TAM study. Bagel Labs' specific target customer definition is sourced from its own careers page.

Who Else Is Fighting for This

Mixed sourcing

Bagel Labs is positioned as a deep-tech infrastructure provider for a technically narrow but strategically significant segment, aiming to build a defensible position through a novel approach to model training before larger, more generalized AI labs can fully commoditize the space.

Company	Positioning	Stage / Funding	Notable Differentiator	Source
Bagel Labs	Physical AI research lab; world-action models for robot control; infrastructure for training frontier diffusion models on commodity hardware.	Seed ($3M, Jan 2024)	Distributed Diffusion Models (DDM) approach; focus on robotics, autonomy, and simulation teams as primary customers.	[Perplexity Sonar Pro Brief, retrieved 2026]; [Bagel Labs Careers, retrieved 2026]
Physical Intelligence	AI research company focused on developing general-purpose AI for the physical world.	Series A ($70M, 2024)	Backed by major tech investors (OpenAI, Sequoia); broad charter for "embodied AI" across multiple domains.	[Crunchbase, retrieved 2026]
Figure	Humanoid robotics company integrating AI models for autonomous operation.	Series B ($675M, 2024)	Full-stack approach combining proprietary hardware with AI; commercial partnerships (e.g., BMW) for deployment.	[Crunchbase, retrieved 2026]
Skild	AI company building a foundational model for robotics.	Seed ($300M, 2024)	Massive seed funding to train large-scale models; focus on creating a single, general-purpose robotic intelligence.	[Crunchbase, retrieved 2026]

Competition in physical AI and robotics is stratified by ambition and go-to-market. At the most ambitious end, well-funded entities like Skild and Physical Intelligence are pursuing foundational models intended to be broadly applicable across robotic forms, backed by capital reserves that allow for massive compute expenditure. Incumbent robotics companies like Figure are vertically integrated, controlling both the AI stack and the hardware, which creates a high barrier to entry for pure-play software providers but also a potential partnership channel. Adjacent substitutes include simulation platforms (e.g., Nvidia's Isaac Sim) and generalist AI labs (e.g., OpenAI, Google DeepMind) whose research into multi-modal and embodied agents could eventually be productized for robotics, though their primary focus remains elsewhere. Bagel Labs operates in a more focused wedge, targeting the infrastructure layer for teams already building in robotics and simulation, a segment that may be underserved by the giants focused on either end-to-end solutions or pure research.

Bagel Labs's claimed edge today rests on its Distributed Diffusion Models (DDM) technical approach, which is designed to train frontier models across commodity hardware rather than requiring a single, massive GPU cluster [Bagel Labs Careers, retrieved 2026]. This could offer a cost and accessibility advantage for its target customers,early-stage startups, research groups, and simulation platforms,who may lack the capital for concentrated supercomputing. The durability of this edge is uncertain. It is a perishable technical advantage that could be replicated by larger competitors if the methodology proves superior, or rendered obsolete by continued declines in the cost of concentrated compute. A more durable, though unproven, edge could be built through early developer adoption and the creation of a specialized toolchain that becomes the de facto standard for a niche community, as seen in other deep-tech infrastructure plays.

The company's most significant exposure is to competitive crowding from both above and below. From above, a well-funded competitor like Physical Intelligence could decide to open-source a similar training framework or offer it as a loss leader to attract developer mindshare, effectively commoditizing Bagel's core technical proposition. From below, the company faces the risk that its target customer segment,early-stage technical teams,is too small or financially constrained to support a venture-scale business, especially if those teams are themselves competing for capital in a crowded AI hardware and robotics market. Furthermore, the company's brand is complicated by the existence of an older, unrelated entity also named Bagel Labs, which could create market confusion and dilute search presence [PRIVATE].

The most plausible 18-month scenario involves increased segmentation. If demand for specialized robotics AI infrastructure grows faster than expected, Bagel Labs could solidify its position as a preferred tool for academic and startup R&D, potentially becoming an acquisition target for a larger player seeking an advanced training stack. In this scenario, a "winner" like Bagel would be one that successfully converts its technical approach into a sticky, adopted platform with a visible user base. Conversely, if foundational model progress accelerates and large labs begin offering robust robotics APIs, the need for a separate, specialized training infrastructure could diminish. In that case, a "loser" would be any pure-play infrastructure company, including Bagel Labs, that fails to either achieve significant commercial traction or demonstrate a performance advantage so compelling that it cannot be ignored by the broader market.

Single-source, plausible -- Competitor data is sourced from Crunchbase and public announcements; Bagel Labs' positioning is from its own careers page and a research brief. Direct competitive claims (e.g., DDM advantages) are unverified by third-party benchmarks.

Opportunity

From the public record If Bagel Labs can establish its distributed training infrastructure as the default for developing physical AI models, the prize is a foundational layer in a robotics and autonomy market projected to reach tens of billions of dollars within the decade.

The headline opportunity is becoming the essential infrastructure layer for training frontier models in robotics and simulation. The company's focus on Distributed Diffusion Models (DDM) for commodity hardware directly targets a critical bottleneck for labs and startups: the prohibitive cost of uniform, high-end GPU clusters [Bagel Labs Careers]. By proving this approach with its Paris model, Bagel Labs is positioning to serve the growing cohort of technically sophisticated teams who need to train complex world-action models but lack the capital of a Google or OpenAI [Bagel Labs Careers]. This outcome is reachable because the technical approach is already validated in a public release, and the company is actively hiring for commercialization roles, signaling a transition from pure research to a product-driven wedge [Perplexity Sonar Pro Brief].

Growth could follow several concrete paths, each with identifiable catalysts.

Scenario	What happens	Catalyst	Why it's plausible
Research-to-Platform	Bagel's DDM framework becomes the de facto open-source standard for academic and industrial labs training physical AI models.	A major research institution or corporate lab (e.g., from automotive or aerospace) adopts and publishes results using the framework.	The company explicitly targets "research groups" and "early-stage researchers from universities" as a core user segment [Cerebral Valley, Nov 2024].
Simulation Vendor Embed	The company's models and training infrastructure are embedded as a core component within leading commercial simulation platforms for robotics and autonomous systems.	A strategic partnership with a simulation software provider is announced, integrating Bagel's models for faster, more accurate synthetic training.	Bagel Labs lists "simulation platforms" as a primary target customer category [Perplexity Sonar Pro Brief].
Robotics OEM Standard	A leading robotics company standardizes its model development pipeline on Bagel's infrastructure, driving adoption across its supplier and partner ecosystem.	A design win with a well-funded robotics startup or a division of a large industrial manufacturer.	The go-to-market motion is defined as working with "robotics companies" and "autonomy teams" to help them train and deploy models [Perplexity Sonar Pro Brief].

Compounding for Bagel Labs would manifest as a data and distribution flywheel. Each new research institution or robotics team that adopts the DDM framework contributes to a growing corpus of training methodologies and, potentially, benchmark results. This public validation attracts more users, which in turn generates more diverse use cases and stress-tests the infrastructure. Success with early adopters in simulation or research creates reference architectures that lower the adoption barrier for subsequent commercial customers. The company's existing focus on serving "early-stage AI startups" suggests a strategy of growing with its customers, embedding its tools early in their development cycles [Cerebral Valley, Nov 2024].

Quantifying the size of the win requires looking at infrastructure providers in adjacent deep-tech fields. For a scenario where Bagel becomes a critical, though not dominant, infrastructure provider for a segment of the robotics AI stack, a credible comparable could be the acquisition multiples for specialized AI tooling companies. A more ambitious but plausible outcome, should the "Research-to-Platform" scenario fully play out, would be to capture a segment of the market for AI developer tools and infrastructure, a category that includes public companies like GitLab (market cap ~$8B as of late 2024) and private entities valued in the hundreds of millions to low billions. If Bagel Labs secures a standard-setting position in physical AI training, its value could approach the lower end of that spectrum (scenario, not a forecast).

Single-source, plausible -- Opportunity framing is extrapolated from stated company goals and target customer segments; growth scenarios are plausible but not yet evidenced by public partnerships or customer announcements.

Sources

From the public record

1. [Wellfound, retrieved 2026] Bagel Labs’ homepage | https://angel.co/jobs

2. [Crunchbase, retrieved 2026] Bagel Labs - Crunchbase Company Profile & Funding | https://www.crunchbase.com/organization/bagel-labs

3. [Perplexity Sonar Pro Brief, Jan 2024] Funding Round Report | https://www.crunchbase.com/organization/bagel-network/news_and_analysis

4. [Bagel Labs Careers, retrieved 2026] Company Careers Page | https://angel.co/jobs

5. [Bloomberg Markets] Bidhan Roy, Bagel Network Inc: Profile and Biography - Bloomberg Markets | https://www.bloomberg.com/profile/person/25082855

6. [LinkedIn] Bidhan Roy on AI and blockchain with NEAR Protocol | https://www.linkedin.com/posts/bagelnet_recently-our-founder-bidhan-roy-joined-activity-7245103297915670528-44KW

7. [Cerebral Valley, Nov 2024] Bagel, Monetizable Open Source AI | https://creators.spotify.com/pod/show/beginners-guide-to-ai/episodes/How-you-can-share-data-and-resources-to-create-your-own-AI-Bidhan-Roy-Interview--REPOST-e2rvq6a

8. [Precedence Research, 2023] Industrial Robotics Market Report | https://www.precedenceresearch.com/industrial-robotics-market

9. [Grand View Research, 2023] AI in Computer Vision Market Report | https://www.grandviewresearch.com/industry-analysis/artificial-intelligence-ai-in-computer-vision-market