Dream Labs
General-purpose robot foundation model that translates camera feeds and English instructions into motor commands.
Website: https://dreamlabs.ai/
Cover Block
Open sources
| Field | Detail |
|---|---|
| Name | Dream Labs |
| Tagline | General-purpose robot foundation model that translates camera feeds and English instructions into motor commands. [Dream Labs, retrieved 2026] |
| Headquarters | San Francisco, United States [Yahoo Finance, October 2026] |
| Founded | 2026 [Korea Post, October 2026] |
| Stage | Seed [Forbes, October 2026] |
| Business model | API / Developer Platform [Dream Labs, retrieved 2026] |
| Industry | Deeptech |
| Technology | AI / Machine Learning |
| Geography | North America |
| Growth profile | Venture Scale |
| Founding team | Co-Founders (2): Joel Jang, You Liang Tan [RuntimeWire, October 2026] |
| Funding label | Seed |
| Total disclosed | Approximately $90,000,000 [Forbes, October 2026] |
Links
Open sources
- Website: https://dreamlabs.ai/
- X / Twitter: https://x.com/jang_yoel?lang=en
What an Investor Needs First
PUBLIC Dream Labs is building a general-purpose robot foundation model and API that turn camera feeds plus plain-English instructions into motor commands, and it merits attention now because it emerged from stealth with an unusually large $90 million seed round for a 2026 robotics software company [Forbes, October 2026] [Dream Labs, retrieved 2026]. The company was founded in San Francisco in 2026 by Joel Jang and You Liang Tan, with launch coverage consistently tying the team back to NVIDIA robotics research and, in Jang’s case, GR00T-related work [RuntimeWire, October 2026] [Business Insider, September 2026] [NVIDIA Learning and Perception Research, retrieved 2026].
The product claim is straightforward but ambitious: Dream-1 is presented as a model that can generalize across robots, tasks, and environments, while the Dream API gives developers a commercial interface starting at $3 per hour according to company materials and launch coverage [Dream Labs, retrieved 2026] [RuntimeWire, October 2026]. If that cross-hardware abstraction holds up outside controlled demos, the differentiation would sit less in a single robot SKU and more in reducing task-specific integration work across low-cost robotic arms and bimanual systems [Dream Labs, retrieved 2026].
Team quality is a central part of the initial underwriting case because public operating history is still thin. Jang’s public record shows prior NVIDIA research roles tied to world models and robot foundation models, while Tan’s published background includes NVIDIA research and earlier work at Berkeley AI Research’s RAIL Lab [LinkedIn, retrieved 2026] [NVIDIA Learning and Perception Research, retrieved 2026] [You Liang's Personal Site, retrieved 2026].
Capital formation is already a signal. Forbes reported a $90 million seed round in October 2026 led by B Capital, with Kindred Ventures and other named firms participating, and the company appears positioned as an API and developer-platform business rather than a vertically integrated robot manufacturer [Forbes, October 2026] [Axios, October 2026].
The next 12 to 18 months should clarify whether Dream Labs can convert technical promise into durable commercial proof. The public evidence to watch is expansion beyond launch partners, credible signs of paid usage on the Dream API, and repeatable performance across heterogeneous hardware, since current traction disclosures are still closer to ecosystem interest than to validated deployment scale [Korea Post, October 2026] [Dream Labs, retrieved 2026].
Partially corroborated -- Core financing is corroborated by multiple public outlets, but several product and traction claims remain company-sourced.
Taxonomy Snapshot
| Axis | Value |
|---|---|
| Stage | Seed |
| Business Model | API / Developer Platform |
| Industry / Vertical | Deeptech |
| Technology Type | AI / Machine Learning |
| Geography | North America |
| Growth Profile | Venture Scale |
| Founding Team | Co-Founders (2) |
| Funding | Seed, total disclosed approximately $90,000,000 |
Inside the Company
PUBLIC
Dream Labs surfaced publicly in October 2026 with a narrow but ambitious pitch: a San Francisco robotics startup building a general-purpose model that turns camera input and plain-English instructions into robot motor commands [Dream Labs, retrieved 2026]. The company presents Dream-1 as "a general robot brain you can build with today," and its public site pairs that model launch with a developer-facing Dream API, which places the company squarely in the infrastructure layer rather than in robot hardware itself [Dream Labs, retrieved 2026].
The verified public timeline is short, which is typical for a company that emerged from stealth this recently. Dream Labs was founded in 2026 and is headquartered in San Francisco, according to the company materials provided here [Dream Labs, retrieved 2026]. Its first clearly documented milestone is the October 2026 launch of Dream-1 and Dream API on the company website, alongside a disclosed $90 million seed financing reported elsewhere in the record, though the financing details sit outside the source set specified for this section and are therefore not relied on here [Dream Labs, retrieved 2026].
No legal entity name or state filing appears in the cited materials for this section, so the public record on corporate formation remains thin. What is visible, and more relevant at this stage, is that Dream Labs has chosen to introduce itself first through product access, research framing, and developer onboarding rather than through a broad corporate narrative, which usually signals an attempt to win early technical adopters before broader commercial proof is available [Dream Labs, retrieved 2026].
Claim stands unchecked -- This section relies primarily on company website materials, with no state filing or second independent public source used here.
Under the Hood
MIXED
The product claim is straightforward, and unusually ambitious for a company this early. Dream Labs says its core model, Dream-1, takes a robot’s camera feed plus a plain-English instruction and returns motor-control commands, with the commercial surface exposed through the Dream API [Dream Labs, retrieved 2026] [RuntimeWire, October 2026]. On the company’s own framing, the point is not a single robot SKU or a single task, but a model that can generalize across robots, tasks, and environments, with support for fine-tuning through uploaded demonstrations on a customer’s specific robot [Dream Labs, retrieved 2026].
The public evidence also sketches the current operating envelope, though some of that detail remains thinly corroborated. Dream Labs says Dream-1 runs on its bimanual YAM setup and on lower-cost robot arms, and that it can handle tasks in scenes it has not seen before, which amounts to a zero-shot generalization claim that investors should treat as company-reported until independent benchmarks surface [Dream Labs, retrieved 2026]. Coverage from RuntimeWire and Forbes ties the launch to a usage-based API, while Dream Labs lists pricing starting at $3 per hour on its site, giving the company a developer-platform posture rather than a pure systems-integrator model [Dream Labs, retrieved 2026] [RuntimeWire, October 2026] [Forbes, October 2026].
A second point worth isolating is where the early wedge appears to sit. Public materials position Dream Labs around manipulation for bimanual robots, and Korean coverage said more than 20 hardware partners, including Unitree and Robotis, were working with the company after launch, which, if sustained, would matter more than a polished model demo because cross-hardware integration is the harder commercial problem in robotics software [Korea Post, October 2026]. Still, the available public record does not yet establish independent performance testing, paid production deployments, or customer outcomes, so the technical story is best read today as a credible founder-market fit paired with company-led evidence on model capability [Dream Labs, retrieved 2026] [Yahoo Finance, October 2026].
Claim stands unchecked -- This section relies materially on company-published product claims, with partial corroboration from launch coverage in RuntimeWire, Forbes, Korea Post, and Yahoo Finance.
Market Research
PUBLIC
The timing matters because general-purpose robotics software is arriving just as labor scarcity, cheaper hardware, and renewed investor appetite are pulling more industrial and developer workflows toward automation, even though the specific market Dream Labs targets is still early and not cleanly measured in public data [Forbes, October 2026] [Yahoo Finance, October 2026].
Public sources around Dream Labs describe a company selling a robot-control model and API rather than a full robot system, which places it at the intersection of several adjacent markets: industrial robotics software, developer infrastructure for AI models, and foundation-model tooling for embodied AI [Dream Labs, retrieved 2026] [RuntimeWire, October 2026]. The difficulty is that none of the cited reporting provides a dedicated third-party TAM for "robot foundation models" as a standalone category. In practice, that means the market case has to be read through analogous public demand signals instead of a clean top-down sizing frame.
The best public evidence of demand in the current source set is operational rather than numerical. Dream Labs launched with a usage-based API, said to start at $3 per hour, and Korean coverage reported more than 20 hardware partners, including Unitree and Robotis, working with the company after launch [Dream Labs, retrieved 2026] [Korea Post, October 2026]. That does not establish paid production revenue, but it does suggest ecosystem interest from robot makers that would benefit from a common control layer across different machines.
A second tailwind comes from the hardware side. Yahoo Finance cited MakerMods co-founder Ryan Chan describing a practical bottleneck: customers can buy relatively low-cost bimanual robot arms, but still struggle to make them do useful work consistently [Yahoo Finance, October 2026]. That is the commercial opening for software vendors like Dream Labs. If robot hardware becomes cheaper faster than robot programming becomes easier, the control stack absorbs more of the value.
Adjacent markets matter here because Dream Labs is effectively competing for budgets that might otherwise go to custom robotics integration, task-specific automation software, or internal machine learning teams building manipulation models in-house [RuntimeWire, October 2026] [Citybiz, October 2026]. The substitute is not only another startup's model. It is also the status quo of bespoke programming, systems integration, and narrow-task robotics deployments. That tends to expand the theoretical budget pool, but it also lengthens evaluation cycles because buyers compare the API against existing engineering headcount and incumbent workflows, not only against other APIs.
The regulatory and macro backdrop is more supportive than restrictive at this stage, though the support is indirect. The reporting set ties Dream Labs to former NVIDIA robotics researchers and to the broader resurgence of embodied AI investing in 2026, culminating in a $90 million seed round led by B Capital and Kindred Ventures [Forbes, October 2026] [Axios, October 2026] [Business Insider, September 2026]. For early infrastructure companies, capital availability matters because enterprise and robotics adoption curves are slower than software-only categories. At the same time, the public sources reviewed here do not yet surface sector-specific certifications, safety approvals, or procurement constraints that would clarify how quickly Dream Labs can move from pilot usage to production deployment.
| Cited market signal | Figure | Market read |
|---|---|---|
| Seed financing announced by Dream Labs | $90M | Investor appetite for embodied AI infrastructure appears strong at formation stage [Forbes, October 2026] |
| Dream API starting price | $3 per hour | Suggests a usage-based developer wedge rather than large up-front integration contracts [Dream Labs, retrieved 2026] |
| Hardware partners reported after launch | 20+ | Indicates early ecosystem interest, though not verified as paid customers [Korea Post, October 2026] |
| Annual plan savings | Up to 20% | Implies the company is already encouraging recurring usage behavior [Dream Labs, retrieved 2026] |
The table points to an important distinction. The public case for market demand is strongest around ecosystem formation and developer access, and weaker around hard evidence of scaled production spending. For now, the market looks credible in direction, but still thinly measured in public.
Claim stands unchecked -- This section relies materially on company claims and launch-period coverage, with no independent third-party TAM report in the cited source set.
Competition and Substitutes
MIXED Dream Labs is positioning itself less as a robot maker and more as a model and API layer that could sit across many hardware platforms, which places it in competition with both robotics incumbents and the growing set of AI-native control stacks for manipulation tasks [Dream Labs, retrieved 2026] [Forbes, October 2026] [RuntimeWire, October 2026].
The competitive map is clearer at the segment level than at the company level. On one side are hardware-first incumbents such as Unitree and Robotis, which appear in public reporting here as hardware partners rather than direct rivals, but which also represent a structural alternative for buyers that prefer vertically integrated systems over a third-party model layer [Korea Post, October 2026]. On another side are foundation-model efforts inside larger research organizations, especially NVIDIA's GR00T program, which matters because Dream Labs' public founder narrative is tightly linked to prior GR00T work and therefore invites direct comparison on model quality, simulation tooling, and developer mindshare [Business Insider, September 2026] [NVIDIA Learning and Perception Research, retrieved 2026]. A third bucket is the adjacent substitute: bespoke robot programming and task-specific integration, the status quo Dream Labs is explicitly trying to replace with an API that converts camera input and English instructions into motor commands [Dream Labs, retrieved 2026] [Citybiz, October 2026].
The edge visible today is concentrated in talent and capitalization, not yet in proven distribution. Joel Jang's public record ties him to NVIDIA robotics research and GR00T-related world-model work, while co-founder You Liang Tan's profile points to both NVIDIA and Berkeley AI Research experience, which gives Dream Labs a credible technical starting point in a field where the pool of people who have worked on generalist robot-control models remains small [Business Insider, September 2026] [LinkedIn, retrieved 2026] [You Liang's Personal Site, retrieved 2026] [NVIDIA Learning and Perception Research, retrieved 2026]. The $90 million seed round also matters competitively because it should let the company subsidize compute, integrations, and hardware ecosystem development earlier than a typical robotics software startup could afford [Forbes, October 2026] [Axios, October 2026]. Still, that edge looks perishable unless Dream Labs turns early technical credibility into proprietary cross-robot data and sticky developer workflows. Public evidence so far shows a launch, an API, and reported hardware partners, but not paid deployments or production revenue at scale [Dream Labs, retrieved 2026] [Korea Post, October 2026] [Yahoo Finance, October 2026].
The company's main exposure is that the control layer can be squeezed from both directions. If NVIDIA continues advancing GR00T and related tooling, it could retain the strongest brand, ecosystem reach, and platform adjacency for developers already building on NVIDIA hardware and software stacks [Business Insider, September 2026] [NVIDIA Learning and Perception Research, retrieved 2026]. At the same time, hardware makers such as Unitree or Robotis could decide that owning more of the intelligence layer is strategically important, especially if customers want tighter performance guarantees than a general-purpose API can offer [Korea Post, October 2026]. Dream Labs is also currently framed around bimanual robots and tabletop manipulation in public materials, which may narrow its effective beachhead even if management's longer-term ambition is broader [Dream Labs, retrieved 2026].
The most plausible 18-month scenario is a sorting market rather than a clear winner-take-most outcome. Winner if ecosystem breadth becomes the key buying criterion is NVIDIA, because its existing robotics research footprint, tooling, and developer familiarity give it the best chance to absorb demand for a general robot foundation layer [Business Insider, September 2026] [NVIDIA Learning and Perception Research, retrieved 2026]. Winner if neutrality across hardware becomes the key criterion is Dream Labs, because its pitch is explicitly cross-manufacturer and API-first [Dream Labs, retrieved 2026] [RuntimeWire, October 2026]. The most exposed player under that same scenario could be any hardware vendor that cannot offer either superior native autonomy or open compatibility; among the names cited here, Robotis or Unitree would be vulnerable if developers come to prefer a shared intelligence layer over model-specific vendor software, though the current evidence does not show either company taking that position publicly [Korea Post, October 2026].
Partially corroborated -- Section relies on two or more independent public sources for Dream Labs' funding, launch, and founder background, but competitive set identification is partly inferred from partner disclosures and founder lineage rather than named direct-competitor reporting.
Opportunity
PUBLIC
If Dream Labs can make robot control portable across low-cost hardware, the prize is not a narrow developer tool but a foundational software layer for general-purpose manipulation, a position that tends to concentrate value when hardware proliferates faster than usable autonomy [Forbes, October 2026] [Dream Labs, retrieved 2026].
The clearest upside case is that Dream Labs becomes the default control and adaptation layer for a broad swath of third-party robots. The public evidence is early but directionally supportive. The company emerged from stealth with a $90 million seed round led by B Capital and Kindred Ventures, an unusually large first check for a robotics software company, alongside a product claim that Dream-1 turns camera feeds and plain-English instructions into motor commands and can be exposed through a public API [Forbes, October 2026] [Axios, October 2026] [Dream Labs, retrieved 2026]. If that interface works across heterogeneous hardware, the company is not selling a single robot or a fixed application. It is selling abstraction, which is usually where software margins and platform power accumulate.
The reachability of that outcome rests on two public facts and one careful inference. First, Dream Labs says Dream-1 is designed to work across different robots, tasks, and environments, and that users can fine-tune the model on their own robot with demonstration data [Dream Labs, retrieved 2026]. Second, Korea Post reported that more than 20 hardware partners, including Unitree and Robotis, were working with the company after launch [Korea Post, October 2026]. The inference is conditional rather than proven: if even a portion of those integrations convert into repeat developer usage, Dream Labs could become embedded before the application layer consolidates.
| Scenario | What happens | Catalyst | Why it's plausible |
|---|---|---|---|
| Cross-hardware API standard | Developers treat Dream API as the fastest way to make bimanual robots perform useful tabletop tasks, and Dream Labs becomes middleware across multiple OEMs | Additional hardware integrations beyond the 20-plus partners cited at launch, especially if major low-cost arm vendors standardize on Dream-compatible workflows [Korea Post, October 2026] | The company launched with a public API, says Dream-1 works across different robots and environments, and was reported to be working with more than 20 hardware partners soon after launch [Dream Labs, retrieved 2026] [Korea Post, October 2026] |
| Fine-tuning platform for robot builders | OEMs and integrators use Dream Labs not only for inference but for model adaptation on their own fleets, making the company part of deployment and maintenance workflows | A visible customer or partner showing that uploaded demonstrations materially reduce deployment time on new robots [Dream Labs, retrieved 2026] | Dream Labs publicly states that users can fine-tune Dream-1 on a specific robot by uploading demonstrations, which gives it a plausible wedge into recurring workflow software rather than one-off model access [Dream Labs, retrieved 2026] |
| Developer-first robotics ecosystem | The company wins early with developers and lower-cost robot platforms, then expands upward into commercial deployments as tooling matures | Broad adoption among makers, labs, and robotics startups that already struggle to turn affordable hardware into useful systems [Yahoo Finance, October 2026] | Yahoo Finance cited MakerMods describing the core pain point as getting purchased bimanual robot arms to perform productive work, which matches Dream Labs' positioning around reducing robot-specific programming [Yahoo Finance, October 2026] [Citybiz, October 2026] |
What compounds here is not simply model quality. It is the possibility that each new robot integration and each fine-tuning workflow improves the company's position with the next developer and the next hardware platform. Dream Labs says users can adapt the model by uploading demonstrations, and says the model is intended to generalize to new tasks and scenes, including zero-shot use cases [Dream Labs, retrieved 2026]. If those claims hold in production, the business could build a feedback loop where more hardware support attracts more experiments, more experiments generate more adaptation data, and better adaptation lowers the cost of onboarding the next robot class.
There is also a distribution flywheel that matters in robotics more than in pure software. Dream Labs is priced as an API from $3 per hour according to its website and launch coverage, which lowers the threshold for trials relative to a custom enterprise robotics deployment [Dream Labs, retrieved 2026] [RuntimeWire, October 2026]. In practical terms, a metered API can turn fragmented robotics demand into many small starting points. If a subset of those users graduate from experimentation to embedded usage, the company could gain recurring revenue before the market settles on a single hardware winner.
The size of the win is easiest to frame by analogy to infrastructure businesses rather than to any one robot OEM. No credible public market comparable or third-party TAM figure was supplied in the reviewed source set, so precision would be artificial. Still, if the "Cross-hardware API standard" scenario plays out, Dream Labs could plausibly become a multi-billion-dollar software platform for robotics control and adaptation (scenario, not a forecast), because it would sit in the control plane between developers and a growing installed base of third-party robots rather than depending on unit sales from a single hardware line [Forbes, October 2026] [Dream Labs, retrieved 2026] [Korea Post, October 2026]. That is the kind of market position that can produce outsized value even before general-purpose robots are commonplace.
Claim stands unchecked -- Core upside rests heavily on company product claims, with partial corroboration from launch coverage and one third-party report on partner count [Dream Labs, retrieved 2026] [Forbes, October 2026] [Korea Post, October 2026].
Sources
Open sources
[Dream Labs, retrieved 2026] Dream Labs | https://dreamlabs.ai/
[Yahoo Finance, October 2026] Dream Labs Raises $90M Seed to Build the Brain for Everyday Robots | https://finance.yahoo.com/technology/ai/articles/dream-labs-raises-90m-seed-170000245.html
[Korea Post, October 2026] Dream Labs, Founded by Korean Entrepreneur in His 20s, Raises $90 Million in Seed Funding,Why U.S. Investors Are Taking Notice | https://www.kspost.biz/en-us/articles/3061
[Forbes, October 2026] Robot Brains For 75 Cents/Hour? Former Nvidia Researchers Raise $90 Million | https://www.forbes.com/sites/johnkoetsier/2026/10/19/robot-brains-for-75-centshour-former-nvidia-researchers-raise-90-million/
[RuntimeWire, October 2026] Dream Labs raises $90M to sell robot control through a $3-an-hour API | https://runtimewire.com/article/dream-labs-90m-seed-dream-1-robot-api
[Business Insider, September 2026] Nvidia Alumni Are Fueling a New Wave of Robotics and AI Startups | https://www.businessinsider.com/nvidia-alumni-fueling-new-wave-robotics-ai-startups-2026-9
[NVIDIA Learning and Perception Research, retrieved 2026] Joel Jang | https://research.nvidia.com/labs/lpr/person/joel-jang/
[LinkedIn, retrieved 2026] Joel Jang - Co-founder & CEO at Dream Labs | https://www.linkedin.com/in/joel-jang-1289331a5/
[You Liang's Personal Site, retrieved 2026] About | https://youliangtan.com/
[Axios, October 2026] Dream Labs banks $90M, Ovation gets bought, Trivium’s new owners | https://www.axios.com/pro/all-deals/2026/10/19/pro-rata-premium-trivium-dream-labs-parkview-health
[Citybiz, October 2026] Dream Labs Raises $90 Million in Seed Funding to Develop AI Platform for Robotics | https://www.citybiz.co/article/920863/dream-labs-raises-90-million-in-seed-funding-to-develop-ai-platform-for-robotics/
Articles about Dream Labs
- Former NVIDIA GR00T Leaders Raise $90 Million for Robot Foundation Model — The startup, founded by former NVIDIA robotics researchers, is selling robot control as a service, starting at $3 per hour.