Covariant vs Preferred Networks

Neither company has publicly disclosed a valuation, making a definitive size comparison difficult. Covariant employs 100-500 people, while Preferred Networks has 100-500 employees.

Preferred Networks has raised more in total funding at $350M, compared to Covariant's $222M — a gap of $128M.

Preferred Networks holds the higher Awaira Score at 72/100, compared to Covariant's 70/100. The Awaira Score is a composite metric factoring in valuation, funding, stage, team size, and market presence — a 2-point gap that reflects meaningful differences in scale or traction.

Covariant was founded by Pieter Abbeel in 2017. Preferred Networks was founded by Toru Nishikawa in 2014. Visit each company's profile on Awaira for a full founder biography.

Covariant: Covariant builds an AI robotic picking and automation platform that enables warehouse and fulfillment robots to handle the enormous variety of product shapes, sizes, and packaging types encountered in real-world logistics operations. The platform is built on RFM-1, a foundation model for robotics trained on one of the largest robotics datasets ever assembled, enabling generalized manipulation capabilities across new product types without task-specific retraining.\n\nThe company raised approximately 222 million USD and has deployed its AI in warehouse environments at major retailers and logistics operators in North America and Europe, with robots handling millions of picks per day across diverse SKU catalogs. Covariant was founded by researchers from UC Berkeley with foundational backgrounds in deep reinforcement learning for robotic manipulation.\n\nIntelligent robotic picking remains one of the hardest unsolved problems in warehouse automation, as the combinatorial variety of product types encountered in e-commerce fulfillment exceeds what rule-based vision systems can handle reliably. Covariant approach of training a generalist manipulation model on large-scale real-world robotics data parallels the approach that made large language models broadly capable, and represents one of the most technically credible attempts to bring general robot AI to industrial deployment at scale. Preferred Networks: Preferred Networks develops deep learning technology applied to robotics, autonomous driving, and industrial applications, building neural network architectures for real-time edge inference in robot control, factory automation, and connected vehicle systems. The Tokyo company gained international recognition for early competition victories in autonomous navigation and robot manipulation challenges and developed its own deep learning framework, Chainer, which influenced the design of PyTorch before Chainer was eventually retired.\n\nThe company raised approximately $350 million including a landmark $105 million Series A from Toyota Motor Corporation and other strategic investors, making it one of the most valuable AI startups in Japan at the time of its fundraising. Preferred Networks collaborates with Toyota on autonomous driving AI, with NTT on communications AI, and with Fanuc on factory robot intelligence, creating a portfolio of deep technology industrial partnerships that provide both funding and deployment scale for its AI research.\n\nPreferred Networks operates in Japan industrial AI market where established relationships with major manufacturing and automotive companies provide a defensible position that international AI startups find difficult to penetrate through conventional sales approaches. The company research focus on edge AI inference for robotics aligns with Japan competitive strengths in manufacturing automation and precision robotics, markets where AI-enhanced robot intelligence is being adopted to address labour shortages and quality requirements that purely mechanical automation cannot satisfy.

Preferred Networks was founded first in 2014, giving it 3 years of additional market experience. Covariant was founded later in 2017. In AI, even a year or two of head start can translate into significantly more training data, customer relationships, and institutional knowledge.

Both Covariant and Preferred Networks report similar employee counts of approximately 100-500. Team size is often a proxy for operational scale, though lean AI companies can punch well above their headcount.

Yes, Covariant and Preferred Networks are direct competitors — both operate in the AI Robotics space and likely target overlapping customer segments. This comparison is especially relevant for buyers evaluating both platforms.