Last week, Eclipse VC announced the close of a new $1.3 billion fund dedicated to investing in "Physical AI" startups. Eclipse, an early backer of Cerebras, is now pivoting toward robotics and AI infrastructure, with a clear rationale behind the move.
But Eclipse is just one player in this wave.
The Numbers First
In Q1 2026, 27 Physical AI startups collectively raised over $6.4 billion. The breakdown:
- Robotics companies: $4 billion
- AI chips and semiconductors: $2 billion
- Other physical infrastructure: $400 million
Looking further back, the entire robotics sector raised about $14 billion in 2025, $8.2 billion in 2024, and $13.1 billion in 2021. 2025 already set a record, and Q1 2026 alone hit $6.4 billion — annualized that's $25.6 billion, nearly double the previous high.
This isn't because investors suddenly got smarter; the underlying conditions have changed.
Why Figure AI Is Worth $39 Billion
Figure AI is the most typical example among humanoid robotics companies.
Valued at $2.6 billion in February 2024, it reached a $39 billion valuation after its Series C in September 2025 — a 15x increase in 18 months. It's not an outlier but a leading representative of the sector.
Why such a valuation?
Figure's robots use multimodal models from OpenAI as their "brain," enabling them to perform automotive assembly line tasks (already partnered with BMW). Before the "AI brain + robot body" combination was validated, robotics companies could only sell valuations based on hardware specs, with a low ceiling. Now, large models provide a path to generalization for robots — the same robot can switch tasks simply by changing instructions, without reprogramming.
This shift has led investors to draw parallels with early Tesla: it looks expensive upfront, but if it works, the market size is astronomical.
Physical Intelligence: The Most Expensive Two-Year-Old Company
Physical Intelligence (π), founded by former DeepMind researchers, is in talks to raise $1 billion at an $11 billion valuation, just four months after its last round.
π's direction is to build a general-purpose robot foundation model — not the robot hardware itself, but the software layer that enables robots to "see, understand, and operate." They have backing from Google, Bezos Ventures, and Amazon.
The $11 billion valuation reflects a key judgment: the software layer is the true high-value segment of the robotics industry. Hardware will become commoditized, but control software and general-purpose policy models are hard to replicate.
What Eclipse Is Betting On
Eclipse's $1.3 billion fund portfolio already includes:
- Arc: Electric boats
- Redwood Materials: Battery recycling and materials
- Bedrock Robotics: Autonomous construction machinery
- Wayve: Autonomous driving technology
- Mind Robotics: Industrial robotics labs
They are investing not just in software, but also in manufacturing, energy, defense, and transportation — the "physical world" sectors.
Eclipse's fund structure is notable: $720 million for early-stage startup investments and $591 million for later-stage follow-ons — they incubate and scale, not just act as a pure financial investor.
Why Now
Three conditions have simultaneously matured:
1. Large models are good enough. Open-source multimodal models like Llama 4 and Gemma 4 can directly serve as the visual understanding and instruction execution brain for robots, eliminating the need for years of in-house perception system development.
2. Hardware costs have come down. The prices of core components like joint motors, force sensors, and depth cameras have dropped significantly over the past few years, mainly due to the scale effects of the consumer electronics and electric vehicle supply chains.
3. Labor shortages are real. Manufacturing sectors in the US, Europe, and Japan all face clear labor gaps. Robots are not competing with humans; they are filling real vacancies.
These three factors were not simultaneously present before 2023. Now they are, and investors see a executable business case, not a vision PowerPoint.
Where the Risks Lie
Of course, there are issues.
Seven Series A rounds exceeding $200 million is unprecedented, implying that product validation is incomplete and institutional investors are pricing in a valuation bubble driven by mutual competition — this is a typical early-stage industry signal and a typical bubble signal; both can coexist.
There is also a technical risk: large models as robot brains perform well in controlled lab environments, but long-tail scenarios in real-world settings like factories, warehouses, and outdoors have not been fully validated.
But if you ask, "Will this path ultimately succeed?" most people's answer is already "yes" — it's just a matter of time.
Sources: VC Eclipse has a new $1.3B fund to back — and build physical AI startups (TechCrunch); Cerebras Backer Eclipse Raises $1.3 Billion for Robotics, AI Infrastructure (Bloomberg); CocoLoop; 27 Physical AI Startups That Quietly Raised $50M+ in Q1 2026 (Foundevo); AI Robotics Lab in Talks to Raise $1 Billion at $11 Billion Valuation (Bloomberg)