Selling Human Movement as Robot Training Data — Mecka AI Raises $60M Using iPhones and Body Sensors

The data that teaches robots "how to use a body" is the next gold mine

Language models like ChatGPT trained on text scattered across the internet. Image models learned from web photos. But robots? "How to use a body" — picking up objects, opening doors, climbing stairs — isn't neatly cataloged anywhere online. A company targeting exactly that gap landed a big round on June 1.

Embodied-AI data startup Mecka AI has raised $60M total, Fortune reported — a $25M Series A closed in November plus a $35M follow-on. Based in New York, what this company does is simple yet shrewd: collect human movement as data and sell it to robots.

The core insight: "real-world data" is the next true bottleneck in AI. Model architecture and compute are advanced enough; for robots to work in the physical world, they need vast data showing "how humans actually move." The company supplying that data becomes the pick-seller of the gold rush.

The players — Mecka AI, CEO Josh Gao, and the investors

Mecka AI was co-founded by CEO Josh Gao. At ~40 employees it's still small, but the ambition is large: capture how people use their hands, walk, and handle objects in daily life, and process it for robot training. The collection method is interesting — not just custom hardware like body sensors, but common devices like iPhones. That opens a path to gathering human movement at scale without special equipment.

The investor list is worth noting. Both rounds were led by Framework Ventures — a VC best known for crypto investing, now betting on robotics data. Menlo Ventures, SV Angel, and Kindred Ventures participated, with angel Ted Xiao joining. Xiao is a former Google DeepMind researcher and a founding member of Jeff Bezos's AI startup "Project Prometheus." Someone at the frontier of robotics/embodied AI putting in personal money signals trust from domain experts.

Framework Ventures co-founder Vance Spencer told Fortune it's "the fastest-growing revenue company we've ever invested in" and "the proof is in the business itself." The data business isn't talk — real money is moving.

What it looks like — numbers and the model

Item	Detail	Note
Total raised	$60M	Disclosed June 1 (Fortune)
Structure	$25M Series A + $35M follow-on	Series A closed November
Lead	Framework Ventures	Both rounds
Participants	Menlo Ventures, SV Angel, Kindred Ventures	+ angel Ted Xiao
Employees	~40	HQ in New York
Data collection	iPhones + custom body sensors	Human movement
Annual run rate	~$100M (on signed contracts)	Run rate
Cumulative data	Thousands of GB	Company-stated

The truly striking figure is the ~$100M annual run rate on signed contracts. A 40-person startup raising $60M has already locked in $100M-scale demand via contracts. Usually startups at this stage raise on "future potential" — Mecka raised on "it's already selling."

The model is clean. Companies building robots and humanoids desperately need "human behavior data" to train their models. Collecting it themselves is expensive and slow. Mecka gathers that data efficiently and at scale via iPhones and sensors, processes it, and supplies it to robot companies. Just as text and image data grew LLMs, Mecka aims to be the raw-material supplier for the era of growing robot AI on "movement data."

Who gains — Mecka, the robotics industry, and the future of data labor

For Mecka, $60M buys "scaling data collection." In a data business, volume is quality and moat. The more people and the more varied environments you capture movement from, the richer the data — and that's the edge. With $100M demand already locked, fresh ammo lets it grow its collection network aggressively.

For robotics and humanoids, the direct gain is easing the training-data bottleneck. A humanoid boom is underway, yet the "human behavior data" to make them smart was scarce. With a specialized supplier like Mecka, robot companies can focus on core robot/model development instead of spending time and money collecting data themselves — speeding up the whole ecosystem.

For the future of data labor, there's meaning too. Collecting human movement via iPhones means anyone could contribute their daily motions as data and be compensated. Just as "data labeling" became a new job in the LLM era, "providing movement data" could become a new form of data labor in the embodied-AI era. Privacy and consent, though, remain problems to solve.

Historical parallels — the wins and shadows of "the company that sells data"

We've seen the light and dark of companies selling not models but data in the AI boom.

Success — Scale AI's labeling empire. In the LLM and self-driving booms, Scale AI grew huge on "data labeling." It didn't build models itself; it processed and sold the data models train on, earning a multi-billion-dollar valuation. Mecka transposes that formula to "robot movement data" — the same "sell the model's fuel, not the model" strategy.

Success — self-driving data companies. In the autonomous-vehicle boom, companies collecting and selling driving data grew into core infrastructure. As cars needed road data to get smart, robots need human-movement data to get smart. Whoever secures that raw material first enjoys the boom steadily.

Shadow — data's ethics and sustainability problems. But the data business has shadows. Collecting people's bodily movement touches sensitive privacy, dragging in disputes over consent, compensation, and ownership. And "human-demonstrated data" is expensive and slow, so it competes with the push toward simulation and synthetic data. How long real-human data's quality edge beats synthetic data's cost edge is the key question.

Competitor counter-plays — the simulation camp and robot giants

The biggest competition is the simulation/synthetic-data camp. Approaches like NVIDIA's robotics simulation platforms, which generate robot-behavior data endlessly in virtual environments, are powerful — cheap and fast, no real humans needed. A "real-world data" company like Mecka must counter on the quality edge that "only genuine human data captures certain subtleties."

Robot and humanoid giants may internalize data. Just as Tesla gathers self-driving data through its vehicle fleet, if humanoid companies stack data through their own robots and workers, the room for an outside supplier like Mecka shrinks. So it's important for Mecka to quickly cement the position of a "neutral, large-scale data supplier not beholden to any one robot company."

Other data startups will pour in too. The moment Mecka proved $100M of demand, "AI's next bottleneck is real-world data" became common knowledge. Ultimately it'll be a race over who gathers more varied, more refined, more ethically clean data faster and at scale.

So what actually changes — by persona

If you follow robotics/embodied AI, read the signal: AI's bottleneck has moved from models to data. However flashy humanoid hardware gets, what makes it smart is data. Mecka's $100M run rate shows this market is already at the stage where real money flows.

If you're weighing a startup, revisit the "pick-seller" lesson. Building the robot body itself has enormous capital and tech barriers, but there's plenty of open space in the data, tools, and infrastructure that make robots smart. While everyone watches the flashy bodies, opportunity for solo and small teams hides in the adjacent "fuel and parts" market.

If you're a general user, know that an era where "your daily motions are traded as data" is coming. Collecting movement via iPhones means the barrier to providing data is dropping. It's an opportunity and a privacy issue at once, so a sense of what data to provide, to whom, and how will matter more and more.

References

• Fortune — Mecka AI raises $60 million to train robots with human data
• Mecka AI Raises $60M to Accelerate Human-Driven Robot Training (citybiz)
• Mecka AI Raises $60M in Funding (FinSMEs)
• Mecka AI Raises $60M to Train Robots With Human Data (Ventureburn)
• Mecka AI Raises $60M Series A (The SaaS News)