Robotics Funding Hits $4.9B in September as Data, Chips Draw Bets

A month with more deals but fewer dollars

In September 2026, robotics startups closed more funding rounds than in the summer months, but the rounds were smaller. One monthly tracker counted 167 robotics rounds worth about $4.86 billion. That beats July and August on deal count and trails both on dollars.21 The same tracker notes that about a third of the deals had no disclosed amount, and that a few large late-stage rounds tend to distort the totals.21 Treat the headline figure as a rough guide.

The pattern inside the numbers matters more than the total. The biggest September checks mostly did not go to companies building humanoid robots. They went to the companies that supply robot builders: chips, sensors, training data, simulation and "brains" meant to run on many kinds of hardware. Our view is that this is the clearest sign yet of where investors think the lasting value in robotics will sit.

The chip layer: D-Robotics' $400 million

The largest disclosed embodied-AI deal of the month was D-Robotics' $400 million Series C, announced September 17.2920 The Shenzhen company spun out of Horizon Robotics, the Hong Kong-listed maker of autonomous-driving chips, in early 2024. It designs computing chips for humanoid and industrial robots.18 It says more than 8 million of its Sunrise chips have shipped, and that first-half 2026 revenue grew several times over the year before. It did not give a specific figure.20

Reports differ on who invested. D-Robotics' own announcement named no individual backers. It described a strategic investment from an unnamed major global internet company, plus institutional investors and returning shareholders.29 Other reports name South Korea's Mirae Asset as the lead, with Meituan and state-linked Chinese funds taking part.2914 The round brings D-Robotics' disclosed funding to about $770 million across four rounds in roughly 26 months.18 The company says it will spend the money on more chip tiers and on software that covers the whole development process, from training models to real-time control of robot joints.14

Politics complicates the story. In July, the US Federal Communications Commission added humanoid and quadruped robots to its Covered List, which blocks new import approvals for those products from foreign, mainly Chinese, manufacturers.18 That ruling does not ban the chips inside the robots, but analysis of the deal argues it makes a US robot built on D-Robotics silicon much harder to bring to market than one built on Nvidia's Jetson Thor.18 The likely outcome is a split supply chain: D-Robotics wins in China and parts of Asia while US robot makers stay with American chips.

D-Robotics was not alone. SiMa.ai, which makes chips for on-device AI in robots and autonomous systems, closed a $150 million Series C at a valuation of about $1.45 billion.25 Lyte raised a $165 million Series C led by Maverick Silicon for sensors and chips that help robots perceive their surroundings.26 Together, these three rounds show investors paying heavily for the hardware that physical AI runs on.

Data is becoming its own market

Training data also drew real money, and this trend looks set to last. Robot foundation models need large amounts of real-world physical data, and that data is hard to collect. A new group of startups now sells it.

XDOF collects teleoperation data, recorded while humans remotely control robots, to train general-purpose robots. TechCrunch reported in early September that it was in late-stage talks for a Series B at about $1.2 billion, led by 8VC, less than three months after leaving stealth.28 One report puts its annualized revenue near $50 million.15 The company pitches itself as an outsourced data supply chain for AI labs and robotics companies that cannot easily build their own data pipelines.28

Mecka AI followed in early October with a $60 million Series B led by Sequoia. Nvidia, Microsoft's M12, Qualcomm and Samsung also took part.1517 Mecka pays people to record everyday tasks such as making coffee or repairing cars while they wear body sensors, then sells the motion data to robot developers.15 It claims a $100 million revenue run rate. TechCrunch reported a valuation of about $500 million, which Mecka has not confirmed.1513 Nvidia now backs both robot makers and data suppliers, including NEURA Robotics' $1.4 billion Series C in June.15

Smaller rounds fit the same pattern. Antioch raised a $32 million Series A from Greylock for cloud simulation that lets teams test robots virtually before using real hardware.26 Robocurve raised a seed round to build open-source tools that independently measure how capable robots and their AI models are.26 Iceland's Treble raised $18 million to simulate sound for physical AI systems.29 Benchmarks and simulation are what a field builds when it expects to be held to measurable results.

The race for robot "brains" gets more expensive

Robot foundation models are AI models meant to control many types of robots across many tasks. They remain the costliest bet in the sector. One 12-month analysis of physical-AI deals, through July 2026, found these companies took about 45% of disclosed capital from just over a quarter of the deals. Humanoid hardware took about 17%.1 That analysis concluded that investors pay a premium for claims of scalable intelligence, not for the humanoid shape.1 September's deals support that reading.

The largest pending deal is FieldAI. Business Insider reported the company is raising $700 million at a $10 billion valuation, about five times its valuation from just over a year earlier.30 That would bring FieldAI closer to Skild AI, valued at more than $14 billion, and Physical Intelligence.30 Physical Intelligence's valuation is reported inconsistently. Business Insider puts it at roughly $11 billion, a company profile lists $11 billion as a March 2026 target, and other coverage still cites the $5.6 billion from its November 2025 round.3082 The gap probably reflects the difference between rumored talks and closed rounds. Readers should take it as a reminder of how loosely valuations are reported in this sector.

Skild AI is the main evidence that this model can earn money. It reportedly had more than 60 paying customers and $50 million in confirmed revenue within ten months of its first commercial deployment, about 90% of it from manipulation tasks.2 Compute spending is rising quickly too. Figure has raised about $1.9 billion in total. In September it signed a deal with Nscale for up to 100,000 Nvidia Vera Rubin GPUs, with an initial $3.5 billion commitment, to train its Helix model.7 Its CEO, Brett Adcock, has said the company is now limited mainly by data and compute.7 On that view, the main constraint has moved from hardware to training resources, which helps explain the September money flowing into chips and data.

New model work keeps coming. Nature Will, a lab with a San Francisco office, announced Natura-1, a 50-billion-parameter model for dexterous robot hands. It was pretrained on more than a million hours of first-person human video and is designed to follow plain-language instructions with objects it has never seen.7 It controls hands only, not the whole body or walking.7

China's embodied-AI pipeline keeps scaling

Chinese deals run on a different scale and cadence. Shenzhen's Kinetix AI raised more than 500 million yuan, about $75 million, across a series of early-stage Angel+ rounds led by Vertex Ventures, Temasek's venture arm.279 Xieyue Intelligence was founded by former Li Auto AI leaders and is building embodied foundation models for the home. It raised hundreds of millions of yuan, its second round in seven months.4 Beijing's Livsyn Robotics closed a Series A above 100 million yuan.19 In October, Astribot announced a 1 billion yuan round at a 15 billion yuan valuation, about $2.2 billion, three months after its previous raise.6

DeepSeek is also entering the field. It agreed to lead a seed round for Winverse AI at a post-money valuation of about $400 million. Winverse was founded by the former head of Tencent's embodied foundation model team.2 The deal had not closed when reported, and its terms could change.2

One analysis of China's sector warns that funding has far outrun revenue. It estimates Galaxy General's valuation at about 1,400 times its expected 2026 revenue.10 That gap is the main risk for the whole category, not just for China.

Humanoids: big totals, smaller September headlines

Humanoid makers raised their biggest rounds earlier in the year. Data cited by Dealroom showed humanoid startups had raised $8.6 billion by late July, 1.8 times the total for all of 2025.15 The largest September checks went elsewhere, and humanoid activity during the month was mostly early-stage. Feather Robotics raised a $7.6 million pre-seed round for modular mobile manipulation robots.24 O-ID raised $1.2 million for a modular industrial humanoid whose parts can be swapped at customer sites.24 Investors increasingly treat the robot body as one component among several, with the main value in the models, data and chips that run it.

Defense was the other large September theme. Mach Industries raised $600 million and Tekever raised $580 million.26 For defense investors, autonomous systems are already a revenue business.

The takeaway

September's activity suggests the robotics boom is turning into an industry with a supply chain. Money is moving down to chips, data and testing tools, and up to cross-platform models. Humanoid makers still draw attention but no longer take most of the capital. Investors are betting that robot bodies will become interchangeable and that whoever controls the computing, data and models will capture most of the value. Two things could undermine that bet: revenue that stays far below valuations, as the Chinese figures suggest, and trade restrictions like the FCC ruling that split the market. Both are worth watching through the fourth quarter.