Company directory
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.
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.
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.
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.
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.
San Francisco, US
Magic belongs on the map as a model-first coding company, not just another developer tool, because it is trying to push the underlying research frontier.
Menlo Park, US
Meta AI is unusual on the map because it combines consumer-scale product surfaces with one of the strongest open-weight strategies in frontier AI.
Redmond, US
Microsoft’s AI position comes from product and cloud distribution at scale, not just model development, which makes it one of the most durable deployment players on the board.
Shanghai, China
MiniMax belongs on the board because it combines frontier multimodal ambitions with consumer product reach, especially in markets outside the usual U.S.-centric map defaults.
Paris, France
Mistral AI sits at an important intersection of frontier research, enterprise adoption, and European strategic positioning in AI.
Beijing, China
Moonshot AI is one of the highest-visibility Chinese frontier model companies, using the Kimi product line to compete on consumer reach, long context, and increasingly broader reasoning workflows.
Seongnam, South Korea
NAVER is South Korea's most important independent foundation-model platform, combining sovereign-language expertise with consumer and enterprise distribution.
San Francisco, US
OpenAI remains the reference point for much of the market because it combines frontier model development with mass user distribution through ChatGPT, strong API adoption, and fast-moving coding and media products.
Paris, France
Pathway is unusual because it now presents itself as both a frontier-model company and an infrastructure layer for continual, time-aware AI systems rather than a conventional enterprise AI wrapper.
Menlo Park, US
Periodic Labs links model development with automated experiments in its own laboratories. Its reported scientific demonstrations are research results, not evidence of a generally available model API or a consumer robot.
New York, US
Reflection AI is an early frontier lab trying to combine high-end research ambitions with an open-intelligence framing rather than a closed product-only strategy.
San Francisco, US
Reka is one of the more interesting frontier companies because it pushes a modular, real-world framing instead of positioning itself purely around benchmark theater.
Palo Alto, US
SSI matters less as a product company today than as a signal of where frontier research capital and talent are concentrating.
San Francisco, US
Standard Intelligence is a young foundation-model lab focused on learning computer use from internet-scale video and interaction data.
Shanghai, China
StepFun is worth tracking because it is part of the fast-changing Chinese frontier-model field, where multimodal and agentic positioning can shift quickly.
Shenzhen, China
Tencent Hy combines a major Chinese foundation-model program with unusually broad distribution across Tencent's consumer and enterprise products.
San Francisco, US
Thinking Machines Lab is emerging as a frontier lab with a real product angle, not just a talent story, through customizable AI systems and research tooling.
San Francisco, US
TypeSafe AI develops machine-native decision models rather than conversational assistants. Jev is available through an API for classification and related structured decisions; confidence estimates do not guarantee that decisions are correct.
Cambridge, US
Unreasonable Labs is notable because it is aiming squarely at AI for science and knowledge discovery, not just lab copilots or generic RAG systems.
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.