Dream Labs Scores $90M for Robot AI Model [2026]

Dream Labs stepped out of stealth on October 19, 2026, with $90 million in seed funding and a pitch that reframes robotics as a software problem rather than a hardware one. The San Francisco company says its flagship model, Dream-1, can read a robot’s camera feed, parse a plain-English instruction, and output motor commands, including on robots it has never been trained on before. The launch arrives through a public Dream API, positioning the company less as a robot maker and more as an infrastructure layer that coding teams can call the same way they call any other cloud service.

The round was led by B Capital and Kindred Ventures, with A.Capital, Unusual Ventures, SV Angel, Wing VC, Scribble VC, and angel investor Adrian Aoun also participating, according to the company’s announcement carried by Yahoo Finance. Dream Labs says it already works with more than 20 hardware partners, and that the bulk of the new capital will go toward compute as it expands the range of tasks and robot types Dream-1 can handle.

Google · Preferred Sources

Don't miss new tech stories on Google

Add TrendinTech once in the Google app and our stories appear in your news suggestions.

Add Now

What Dream Labs Actually Announced on October 19

Strip away the funding headline and the core announcement is a developer product: Dream API. The API takes two inputs, a live camera feed from a robot and a natural-language instruction such as “pick up the cup and place it on the shelf,” and returns motor commands that drive the robot’s arms and grippers. Dream Labs says the system works on hardware it has not previously encountered, a claim that, if it holds up at scale, would let a single model serve robots built by different manufacturers instead of requiring custom control software for each machine.

That is the part of the story aimed squarely at software teams rather than robotics specialists. Dream Labs is explicitly selling an API, the same packaging convention that has made large language models usable by any developer with a credit card and an internet connection. The company is betting that robot control can follow the same path: wrap a frontier model behind a simple interface, let developers integrate it the way they would integrate any other AI service, and let the underlying model improve centrally rather than through fragmented, robot-specific engineering.

The announcement was first reported by The AI Insider, which framed Dream-1 as a robot foundation model designed to transfer skills across hardware made by different manufacturers rather than a model tied to one robot line.

Inside the $90 Million Seed Round

Seed rounds of this size are unusual outside of AI infrastructure, and the investor list signals how much venture capital now treats robot foundation models as an extension of the broader AI buildout rather than a separate, slower-moving hardware category. B Capital and Kindred Ventures led the round, with participation from A.Capital, Unusual Ventures, SV Angel, Wing VC, and Scribble VC, alongside Adrian Aoun.

Detail	Figure
Total seed funding	$90 million
Lead investors	B Capital, Kindred Ventures
Other participants	A.Capital, Unusual Ventures, SV Angel, Wing VC, Scribble VC, Adrian Aoun
Flagship model	Dream-1
Public product	Dream API
Hardware partners	20+
Announcement date	October 19, 2026
Stated use of funds	Compute, to expand task and robot coverage

The size of the round puts Dream Labs in the same conversation as the better-funded robot foundation model players, even though it is only now emerging publicly. It also lands in a quarter where AI-adjacent coding and infrastructure startups have pulled in enormous sums even as some regional ecosystems lag behind, a split covered in TrendInTech’s look at the $25 billion European VC quarter that also lifted AI coding startups. Dream Labs’ pitch fits a pattern investors have rewarded all year: sell the model through an API, let usage data improve it, and treat the underlying hardware as replaceable.

How Dream-1 Turns Video Into Motor Commands

According to Dream Labs’ own description of the system, Dream-1 was trained on video of people performing everyday tasks rather than on robot-specific demonstration data alone. The model is designed to predict what should happen next in a scene and translate that prediction into the motor commands a connected robot needs to execute the task. The company says this lets Dream-1 control robots equipped with grippers without additional, robot-specific training, a zero-shot capability that is central to the pitch.

Early demonstrated tasks are deliberately mundane: picking and placing objects, and folding laundry. Neither task is flashy next to humanoid robots doing backflips, but both are useful proxies for the kind of repetitive manipulation that warehouses, light manufacturing, and eventually homes need robots to handle reliably. Unite.AI’s coverage of the launch noted that Dream Labs paired the model release with a fine-tuning service, letting developers submit a small demonstration dataset to adapt Dream-1 to a specific task or robot, and a browser-based Robot Playground that lets anyone send commands to a physical robot in Dream Labs’ San Francisco office without owning hardware themselves.

Why Cross-Embodiment Matters to Developers

Robotics engineers call the problem Dream Labs is targeting “cross-embodiment transfer,” the ability for one model to generalize across robots with different arm lengths, gripper designs, and sensor placements. Historically, every new robot model required its own training pipeline, its own calibration, and often its own team of specialists. If Dream API genuinely works across hardware from different manufacturers out of the box, it removes a huge chunk of the integration work that has kept robotics software teams small and slow compared with web or mobile development teams.

That framing matters for why this story belongs in a software conversation and not just a hardware one. The developer experience Dream Labs describes, point a camera at a robot, send an instruction string, get back motor commands, mirrors how teams already call coding assistants and AI agents for software tasks, a pattern explored in TrendInTech’s reporting on AI agents completing long-horizon tasks inside Minecraft. Robotics is now being pitched with the same API-first vocabulary that reshaped how developers consume language models.

Pricing: What Forbes Reported and What Remains Unconfirmed

Forbes reported that Dream Labs is pricing Dream API access at $3 per hour for its primary model, with a lighter “flash” variant available at roughly 75 cents per hour. Those figures have circulated widely since the launch, but Dream Labs’ own announcement material does not itself spell out per-hour pricing, so the exact rate card should be treated as Forbes’ reporting rather than a confirmed, published price list. Developers evaluating the platform should expect the company to clarify final pricing tiers as the API moves out of its current access stage.

Usage-based, per-hour billing for a physical-world model is itself notable. It treats robot compute the way cloud providers treat GPU instances, charging for time the model spends actively reasoning about a task rather than per API call or per completed action. If that billing model sticks, it would put Dream API pricing conversations alongside the kind of token-based and call-based pricing debates already playing out for language model APIs, including the per-token pricing shift TrendInTech covered around OpenAI’s Decisions API moving to public beta.

From Nvidia’s Robotics Lab to a Cap Table

Multiple outlets covering the launch, including Unite.AI and Forbes, reported that Dream Labs was founded by former Nvidia robotics researchers, with some team members previously associated with Nvidia’s GR00T robotics research effort. Dream Labs’ own materials have not detailed individual founder biographies beyond that lineage, so specifics of titles and tenure remain attributed to those outlets rather than confirmed by the company directly.

That pedigree is relevant context. Nvidia has spent much of the past two years pushing general-purpose humanoid and manipulation research through GR00T, effectively training a generation of researchers in exactly the cross-embodiment problem Dream Labs is now trying to commercialize independently. A startup spinning out of that lineage with a $90 million seed round suggests investors are comfortable betting that the research is mature enough to productize now, rather than waiting for a slower, incumbent-led rollout.

Dream Labs Versus the Rest of the Robot Foundation Model Field

Dream Labs is not the only company chasing a general-purpose robot brain. Physical Intelligence, Figure AI, and Skild AI have each pursued versions of the same idea: a model that generalizes across tasks and, ideally, across hardware. The companies differ mainly in distribution strategy. Figure AI has focused heavily on its own humanoid hardware line. Physical Intelligence has emphasized research publications and partnerships across robot types. Skild AI has pitched a foundation model, Skild Brain, intended to span quadrupeds, humanoids, tabletop arms, and mobile manipulators without prior knowledge of a robot’s exact form, a description that overlaps closely with what Dream Labs claims for Dream-1.

Company	Flagship Product	Primary Distribution Model	Embodiment Approach
Dream Labs	Dream-1 / Dream API	Public API, hourly billing, fine-tuning service	Cross-manufacturer, gripper-equipped robots
Physical Intelligence	Pi-series vision-language-action models	Research partnerships, select API access	Broad manipulation tasks across robot types
Figure AI	Figure humanoid platform	Vertically integrated hardware and software	Primarily its own humanoid hardware
Skild AI	Skild Brain	Licensing to hardware partners	Quadrupeds, humanoids, arms, mobile manipulators

The distinction that separates Dream Labs from Figure AI in particular is that Dream Labs is not trying to sell robots. It is trying to sell the software layer that any robot maker’s hardware can plug into, which is closer in spirit to how Silicon Labs has positioned AI coding agents as a layer that chip teams wire directly into their own development tools, a move detailed in TrendInTech’s report on the Simplicity AI SDK. Both bets assume that developers, not hardware vendors, decide which intelligence layer wins.

Why This Is a Coding Story, Not Just a Robotics Story

It is tempting to file Dream Labs under robotics news and move on, but the mechanics of the launch put it squarely in developer-tooling territory. Dream API, the fine-tuning service, and the Robot Playground are all products built for software engineers who may have never touched a robot arm before. The Robot Playground in particular removes the single biggest barrier to robotics experimentation: owning expensive hardware. A developer with a browser tab can now send a natural-language instruction to a physical robot in Dream Labs’ office and watch it execute, the same low-friction, try-before-you-integrate pattern that coding tools like OpenCode have used to get developers running local AI agents in under two hours, as TrendInTech’s step-by-step OpenCode setup guide walked through.

That lowered barrier matters because it changes who can build robotics applications. Instead of requiring a robotics PhD and a six-figure hardware budget, a team building, say, a warehouse picking tool or a small-batch assembly workflow can prototype against Dream-1 through an API key, test logic in the Robot Playground, and only commit to physical hardware once the software behavior is validated. That is a materially different cost structure than robotics development has had for most of its history.

Market Impact: What $90 Million Signals to the Robotics Sector

A $90 million seed round is large enough to function as a market signal on its own. It tells other investors that robot foundation models are no longer treated as a speculative, decade-out research bet, but as a category worth funding at the same intensity as large language model infrastructure. It also raises the bar for what counts as a credible entrant in the space: a founding team without deep robotics research pedigree, or without an API-first distribution plan, will have a harder time raising at this scale going forward.

For hardware manufacturers, the signal is more complicated. If Dream API delivers on cross-manufacturer control, it reduces the competitive moat that comes from proprietary control software, pushing hardware makers toward competing on cost, durability, and sensor quality instead. Pulse2’s coverage of the raise framed this explicitly as Dream Labs positioning itself as infrastructure that sits above any single robot brand, which is precisely the kind of commoditizing pressure that has already played out in cloud computing and, more recently, in AI model pricing wars among language model providers.

The Open Questions Nobody Has Confirmed Yet

None of the public reporting so far includes Dream-1’s exact model architecture, parameter count, training compute budget, or independently verified benchmark scores. Dream Labs has not published a model card or technical paper alongside the funding announcement, which means claims about zero-shot cross-manufacturer performance currently rest on the company’s own demonstrations rather than third-party evaluation. That is a normal stage for a seed-funded startup, but it is worth naming plainly rather than assuming the capability claims have been independently audited.

Availability terms are similarly unsettled. It is not yet clear whether Dream API is in open beta, waitlisted, or fully public, nor whether the reported hourly pricing applies uniformly or varies by task complexity and robot type. Developers interested in testing the platform should expect the terms to firm up over the coming weeks rather than treating early coverage as a final specification.

Independent Benchmarks Are Still Missing

No outside lab or academic group has published its own test of Dream-1’s cross-manufacturer claims as of this writing. That gap is common at the seed stage, but it means enterprise buyers evaluating Dream API today are relying on vendor demonstrations rather than reproducible, third-party results, the same caution enterprise AI buyers have learned to apply to any newly launched model before committing production workloads to it.

What Developers Should Know Before Building on Dream API

Dream API accepts a camera feed and a natural-language instruction, and returns motor commands, according to Dream Labs’ own description of the product.
The model is designed to work on previously unseen, gripper-equipped robots, a zero-shot claim that has not yet been independently benchmarked.
A fine-tuning service lets teams adapt Dream-1 to a specific task using a small demonstration dataset rather than a full retraining pipeline.
The Robot Playground offers a no-hardware-required way to test commands against a physical robot before committing to an integration.
Reported pricing from Forbes puts primary-model access at roughly $3 per hour, with a lighter model at about 75 cents per hour, though Dream Labs has not published an official rate card confirming those figures.
More than 20 hardware partners are already connected, according to the company, though it has not named most of them publicly.

Historical Context: How We Got From Scripted Arms to API-Driven Robots

Industrial robotics spent decades relying on hand-coded motion paths, where engineers programmed exact coordinates for every movement a robot arm would make. That approach worked for fixed, repetitive factory tasks but broke down the moment a workspace changed or a new object appeared. The shift toward machine learning-based manipulation over the past five years, accelerated by cheaper simulation, better vision models, and large-scale video pretraining, has slowly replaced hand-coded paths with models that infer motion from perception and intent.

Dream Labs’ pitch is the logical next step in that progression: instead of training a model per robot or per task, train one model on broad human-activity video and let it generalize to whatever hardware a developer plugs in. It is the same generalization bet that made large language models useful across industries they were never explicitly trained for, now applied to physical manipulation. Whether that bet pays off as cleanly in the physical world, where mistakes have real consequences and sensors vary wildly between manufacturers, is the open question the next twelve months of Dream API usage will start to answer.

Predictions: Where Robot Foundation Models Go From Here

Expect at least one competing robot foundation model company, likely Skild AI or Physical Intelligence, to announce its own public, hourly-billed API within the next two quarters to match Dream Labs’ developer-first positioning.
Dream Labs will likely publish a technical report or model card within six months of this announcement, both to satisfy enterprise buyers and to counter skepticism about unverified zero-shot claims.
Hardware partners currently working with Dream Labs will push for clearer liability and safety terms before deploying Dream API in commercial warehouse or manufacturing settings, mirroring the caution seen in early industrial AI agent rollouts.
Pricing will likely shift from flat hourly billing toward tiered pricing based on task complexity, following the same path language model APIs took as usage scaled.
Larger cloud and chip vendors will likely move to either partner with or acquire a robot foundation model startup within the next 12 to 18 months, treating physical-world models as the next frontier after language and code.

Competitive Comparison: Robot APIs Versus Coding Model APIs

It is useful to compare Dream Labs’ launch against how coding model APIs have evolved, since Dream Labs is explicitly borrowing that playbook. Coding assistant APIs started with single-task code completion, moved to full agentic coding with tool use, and are now competing heavily on price per unit of output, visible in the pricing gap TrendInTech documented between Mistral Large 4 and GPT-6 Astra. Dream API is launching at the agentic-task stage already, skipping the single-task completion phase entirely because physical manipulation does not really have a useful single-token equivalent. That means Dream Labs will likely face pricing pressure faster than coding APIs did, since enterprise buyers evaluating robotics automation already have a mental model, shaped by cloud and AI API pricing wars, for how quickly per-unit costs tend to fall.

Frequently Asked Questions

What is Dream Labs?
Dream Labs is a San Francisco-based AI company that announced $90 million in seed funding on October 19, 2026, to build foundation models for robots, led by B Capital and Kindred Ventures.

What is Dream-1?
Dream-1 is Dream Labs’ flagship robot foundation model, designed to convert a robot’s camera feed and a plain-English instruction into motor commands, including on robots it has not previously encountered.

What is Dream API?
Dream API is the public product that gives developers access to Dream-1, letting them send camera input and natural-language instructions to control a connected robot.

How much does Dream API cost?
Forbes reported pricing of roughly $3 per hour for the primary model and about 75 cents per hour for a lighter model, though Dream Labs has not published an official rate card confirming those exact figures.

Who invested in Dream Labs’ seed round?
B Capital and Kindred Ventures led the round, with A.Capital, Unusual Ventures, SV Angel, Wing VC, Scribble VC, and Adrian Aoun also participating.

What tasks can Dream-1 currently perform?
Early demonstrated tasks include picking and placing objects and folding laundry, using robots equipped with grippers.

How is Dream Labs different from Figure AI or Physical Intelligence?
Dream Labs is focused on selling software that works across hardware from different manufacturers rather than building or selling its own robot, a more API-first approach than Figure AI’s vertically integrated humanoid platform.

Is Dream-1’s zero-shot capability independently verified?
Not yet publicly. Dream Labs has not released a technical paper, model card, or third-party benchmark results alongside the funding announcement, so current claims rest on the company’s own demonstrations.

Dream Labs Raises $90M for a Universal Robot API [2026]