Company directory
Beijing, China
01.AI now combines the open-weight Yi model family with enterprise platforms and vertical agents for sales, investing, and knowledge work.
Palo Alto, US
1X stands out because it is combining humanoid hardware, home deployment ambitions, and a growing in-house AI stack instead of treating robotics as hardware alone.
Tel Aviv, Israel
AI21 Labs is a long-running independent model company whose Jamba releases combine open weights, long context, and enterprise deployment.
Hangzhou, China
Alibaba matters because it combines model development with cloud distribution and consumer-scale surfaces, giving Qwen a path from research into deployment.
Seattle, US
Amazon’s AI effort is best understood as a distribution machine: models, tooling, and agents delivered through AWS rather than a standalone consumer lab story.
Paris, France
AMI launched in 2026 with a distributed team operating from Paris, New York, Montreal, and Singapore and a focus on controllable world-model intelligence.
San Francisco, US
Anthropic has become one of the strongest frontier competitors by pairing Claude model quality with enterprise trust, long-context workflows, and unusually strong coding and agent performance.
Cupertino, US
Apple is a major foundation-model developer because its models ship directly across a vast device ecosystem with a distinct on-device and privacy architecture.
Austin, US
Apptronik is a leading U.S. humanoid company combining robot hardware, data collection, AI training, manufacturing partners, and customer deployments.
San Francisco, US
Autoscience Institute is worth tracking because it pushes toward the idea of AI systems that do substantive model-development work, not just assistant tasks for researchers.
Beijing, China
ERNIE is one of China's most important foundation-model families, with direct distribution through Baidu Search and Baidu AI Cloud.
Beijing, China
ByteDance / Doubao matters because it shows what scaled consumer AI deployment looks like when a major internet platform combines models, distribution, and media products.
Sunnyvale, US
Cerebras belongs on the map because compute and inference architecture increasingly shape who can train and serve important models in practice.
Toronto, Canada
Cohere stands out as an enterprise-first model company, prioritizing retrieval, secure deployment, and business workflows over broad consumer reach.
Livingston, US
CoreWeave is one of the most important AI infrastructure companies, selling a purpose-built cloud for training and inference to model labs and enterprises that need GPU-heavy capacity fast.
Denver, US
Crusoe matters because AI infrastructure is now a product and capital-allocation story, not just a hosting detail, especially when energy and data-center strategy are central constraints.
Hangzhou, China
DeepSeek is one of the most important open-model challengers, pairing aggressive research releases with an API business that keeps pressure on closed-model pricing and benchmark leadership.
London, UK
Generalist AI is a focused embodied-intelligence lab pursuing scaling laws and general-purpose models for physical skill learning.
San Francisco, US
Goodfire matters as a model-understanding company rather than another application wrapper, turning interpretability research into usable infrastructure.
London, UK
Google DeepMind combines world-class research depth with mass distribution through Google products, making it one of the most consequential labs for frontier multimodal systems, scientific AI, and robotics.
San Jose, US
Groq is positioning itself as the speed-first inference stack for production AI, using its custom LPU architecture and GroqCloud to compete on latency and cost rather than training breadth.
Paris, France
H Company stands out because it is trying to bring frontier-style agentic systems directly into enterprise workflows rather than stopping at research demos.
Palo Alto, US
Inception Labs is notable because it is not just another LLM company; it is using diffusion language models to compete on speed, coding, and inference economics.
Hong Kong
Insilico Medicine matters because it is one of the most prominent efforts to turn AI for science into a repeatable drug-discovery business.
London, UK
Isomorphic Labs matters because it represents one of the strongest attempts to turn frontier AI research into a drug-discovery company with real pharma leverage.
San Francisco, US
Lambda sits in the middle of the AI compute boom, packaging GPU cloud, clusters, and AI factory infrastructure for teams that need training and inference capacity without building it from scratch.
Palo Alto, US
Acquired by NVIDIA in 2025, Lepton now operates as the infrastructure layer behind DGX Cloud Lepton.
San Francisco, US
Letta matters because memory and state management are becoming core infrastructure questions for agentic systems, not just product features.
Seoul, South Korea
LG AI Research combines an active open-model program with direct pathways into a large industrial and consumer conglomerate.
San Francisco, US
Logical Intelligence is an early research lab focused on energy-based reasoning systems for settings where verification and correctness matter more than generic chatbot fluency.
The Staticbreaker
Get the signal behind the week's AI changes.
The Staticbreaker turns launches, category shifts, and market noise into a concise editorial briefing.