Asimov

Internet-scale marketplace for robot training data

Winter 2026ActiveB2BSupply Chain and LogisticsRoboticsData LabelingAISan Francisco, CA, USA; Remote

Company

Asimov is scaling the most diverse dataset of human motion data to teach robots how to interact with the physical world. We deploy our hardware across a network of 5,000+ contributors in households, restaurants, hotels, and factories to supply frontier robotics labs with thousands of hours of organic human data.

Verdict

High Signal

Market Opportunity

Robot training data is a genuine and rapidly growing market as humanoid robotics labs (Figure, Physical Intelligence, Boston Dynamics, 1X, etc.) are all data-hungry and paying real money for high-quality human motion data. B2B sales to frontier robotics labs is a clear ICP with real budget. TAM is plausibly $1B+ as the robotics wave accelerates through 2026-2030.

Medium Signal

Founder Signal

Both founders are current UC Berkeley undergrads (graduating 2026), making them effectively fresh grads with internship-level experience. Lyem has the stronger profile — 1.5 years at FLIP doing ML for Air Force fleet analytics and 6 months as a founding engineer at Blume (YC W24), which shows early-stage execution experience. Anshul has a 4-month SWE internship at Scale AI (relevant for data pipelines), a 4-month Amazon internship, and ML research at BAIR — all internships, no full-time shipping experience. Neither has robotics hardware deployment experience, which is core to the product thesis.

Low Signal

Competition

This space has real competitors including Scale AI (already doing robotics data), Physical Intelligence's internal data collection, Hugging Face's LeRobot dataset efforts, and specialized players like Open-X Embodiment. Scale AI in particular is a direct threat given its existing relationships with robotics labs and Anshul himself interned there. No meaningful differentiation has been articulated beyond '5,000+ contributors' which is an unverified claim.

Low Signal

Product

No press coverage, no named customers, no revenue metrics, no live demo or API docs. The website (tryasimov.ai) is described as a marketplace with 5,000+ contributors, but this claim is unverifiable and there is zero third-party evidence of any real deployments or paying customers. Pure description-stage vaporware at this point.

OverallC Tier

Strong market thesis riding the humanoid robotics wave, but execution evidence is near-zero — no customers, no press, no verifiable traction. Both founders are current undergrads with internship-level experience; Lyem's founding engineer stint at Blume (YC W24) and ML work for the Air Force is the strongest signal, but neither has shipped hardware at scale or managed a data contributor network. The claim of 5,000+ household/restaurant/factory contributors is extraordinary and completely unsupported. Scale AI, a massive incumbent, is already in this space and Anshul literally just interned there — which cuts both ways as domain knowledge but also signals the threat. This looks like a smart team with a real market insight but no product proof yet; needs customer logos and revenue before it can be taken seriously.

Active Founders

Lyem Ningthou

Founder

Defense tech robotics researcher → building the data layer for humanoid robotics

Anshul Verma

Founder

Previous software engineering at Scale AI, Amazon and ML research at Berkeley. Building a marketplace for robot data.

Asimov

TierC Tier

BatchWinter 2026

Team Size3

StatusActive

LocationSan Francisco, CA, USA; Remote