Generalist AI (also referred to as Generalist), a San Mateo/Bay Area-based frontier lab building embodied foundation models (“robot brains”) for general purpose physical intelligence, has raised approximately $200 million in additional funding as an extension of its Series B.
Generalist AI’s recent $200 million round was led by 8VC and included participation from existing investors. This brings the full Series B to roughly $600 million and lifts the company’s valuation to about $3 billion (up from the $2 billion post money valuation set in the initial June 2026 tranche). Total capital raised now exceeds $650 million.
What is Generalist AI?
Founded in 2024 by Pete Florence (CEO, formerly senior scientist at Google DeepMind, key contributor to PaLM-E, RT-2, and related “vision language action” systems), Andy Zeng (Chief Scientist, also ex DeepMind, known for work such as Code as Policies), and Andrew Barry (CTO, formerly a roboticist at Boston Dynamics involved with systems like Atlas, Spot, and Stretch), the company focuses exclusively on software models rather than robot hardware. Team members draw from DeepMind, OpenAI, Boston Dynamics, and other frontier labs. It operates in the Bay Area and Boston.
Earlier rounds included seed and Series A financing (with figures cited around $12.5–128 million ranges in various trackers, including significant Nvidia involvement). The core Series B of $400 million closed and was announced in early June 2026, led by Radical Ventures. Participants included 8VC, Union Square Ventures, Norwest, Hanabi Capital, Nvidia’s NVentures, Boldstart Ventures, Spark Capital, Bezos Expeditions, NFDG, and angels such as Fei-Fei Li, Eric Yuan (Zoom CEO), Bin Lin (Xiaomi co-founder), and Naval Ravikant. That round valued the company at $2 billion and was intended to fund next generation models, physical data collection/scale-up, and compute/training infrastructure.
The rapid follow-on (just ~2 months later, with talks already circulating by late July) reflects intense investor demand in the “physical AI” or embodied intelligence sector.
What is Generalist’s technology?
Generalist develops large multimodal embodied foundation models trained primarily on proprietary high fidelity physical interaction data (hundreds of thousands of hours collected via its own systems, including handheld grippers for human demonstrations). Models process video, language, proprioception, and other sensors to output high frequency action trajectories (e.g., 100 Hz). They aim for cross embodiment generalization (robotic arms, different end-effectors, potential extension to humanoids and industrial systems) rather than hardware specific solutions.
Key milestones:
- GEN-0 (November 2025): Early embodied foundation model scaling with physical interaction data.
- GEN-1 (April 2026): Demonstrated mastery of simple physical tasks (high reliability, e.g., claims of 99% on diverse dexterous tasks in some reports, with execution speeds up to 3x prior benchmarks) after post training. Supported broader end-effector transfer.
- GEN-1.5 (released/announced August 19, 2026, shortly before the funding news): A significant advance in one shot and few shot learning. It can learn new short horizon manipulation tasks from a single 3–12 second demonstration via “physical prompting” (inserting sensorimotor examples into a ~30 second context window) with no gradient updates or fine tuning. Average success across 10 diverse tasks (e.g., twisting jar lids, unzipping pouches, retrieving items from wallets/purses, sweeping with a brush, stacking cups, folding paper) reached ~59% (±10%). With 1–10 gradient steps on 1–5 minutes of data (~10–50 demos), performance rose to ~83% (±9%). Additional capabilities include compositional chaining of prompts into longer behaviors, zero shot sim to real transfer, some human to robot imitation, error recovery, improvisation (novel tool use such as brush/dustpan, alternative strategies, ambidexterity), and continued improvement after >8 months of continuous pretraining.
The models emphasize real world physics intuition emerging from scale rather than pure adaptation of language/vision models. The company works with a small number of industrial customers, iterating based on feedback for manufacturing, logistics, assembly, and related use cases.
Proceeds are directed toward high performance compute acquisition, expansion of core machine learning research and robotics engineering teams, and scaling collaborative enterprise deployments with industrial clients. This aligns with the capital intensive nature of training (GEN-1.5 alone required months of continuous pretraining) and data collection in the physical domain.
Generalist positions itself as building general intelligence for the physical world, enabling robots to handle unscripted, dexterous tasks across varied hardware and environments, with the longer term goal of impacting factories, warehouses, labs, farms, homes, and beyond. Success here could accelerate automation of blue collar/physical labor by reducing the need for extensive task specific programming or data.
Recommended: Princeton Critical Minerals Raises $16M In Series A And Grant Funding
This funding occurs amid a surge of capital into “robot brains” / physical AI / embodied foundation models, as investors seek the robotics equivalent of a ChatGPT moment (generalist capabilities without per task engineering). The category has seen multi billion dollar valuations and large rounds for pure software or software heavy players.
Notable comparables (valuations approximate and evolving as of mid to late 2026 reporting):
- Skild AI: ~$14B+ (large SoftBank-led rounds; “omni-bodied” models).
- Physical Intelligence: Reported figures ranging from ~$5.6B (confirmed earlier) to ~$11B (discussions/rumors); strong DeepMind alumni roots and VLA style models.
- Field AI: Around $2B.
- Genesis AI: In talks near $3B.
- Broader field includes full stack humanoid efforts (e.g., Figure AI at much higher valuations) and others.
Generalist’s rapid capital raises, technical progress on sample efficient learning, and blue chip backers (Nvidia, Bezos, top VCs, Fei-Fei Li) place it firmly in the leading cohort of software focused players, though absolute valuations trail the largest peers. Challenges remain inherent to the domain: physical data is harder/scarcer than internet text/images, real world variability is high, and achieving reliable long horizon generalist performance at scale is still early. The company’s emphasis on proprietary physical interaction data and continuous scaling provides a differentiated path.
The $200 million extension underscores exceptional investor confidence in Generalist’s trajectory, team, and the broader physical AI thesis, enabling accelerated model iteration, infrastructure build-out, and commercial traction at a pivotal moment for the sector.
Please email us your feedback and news tips at hello(at)dailycompanynews.com
