AI labs in US

Public benefit corporation building the Claude models, with a focus on AI safety research.

Maker of the GPT models and ChatGPT; its charter states the mission of ensuring AGI benefits all of humanity.

AI answer engine; post-trains its own Sonar models.

Wafer-scale AI chips and fast inference.

Builds the Grok models; states its goal as understanding the true nature of the universe.

Ex-DeepMind founders building open frontier models and autonomous coding agents.

NVIDIA's research arm; publishes the Nemotron and Cosmos model families.

MIT spin-out building efficient non-Transformer Liquid Foundation Models.

Decentralised training of open models (INTELLECT).

Flagship Pioneering company building scientific superintelligence and autonomous labs.

AI cloud and open-model research (RedPajama, FlashAttention authors).

Non-profit releasing fully open models, data and code (OLMo, Tülu, Molmo).

IBM's open Granite model family.

Amazon's AGI organisation, building the Nova foundation models.

Fast inference for open models.

Databricks' model research group (DBRX).

Field foundation models for robots in unstructured outdoor places.

Founded by OpenAI's former CTO; released the Tinker fine-tuning API.

Apollo humanoid robots.

AI-first drug discovery.

AI for law firms and legal teams.

Open hybrid SSM models (Zamba) and multimodal agents.

AI scientists that run real experiments, starting with materials science.

Salesforce's research group (xLAM agent models).

Video models (Ray) and multimodal world models.

Apple's research group behind the Apple Foundation Models.

Snowflake's research team (Arctic models).

Non-profit open research collective (GPT-NeoX, Pythia, the Pile).

Microsoft's consumer AI and model group led by Mustafa Suleyman, building Copilot and MAI models.

Inference platform for AI models.

GPU cloud for AI labs (NASDAQ: CRWV).

Meta's AI research and model teams, behind the Llama open-weight models.

AI code editor; trains its own coding models.

AI data centres powered by stranded and clean energy.

LPU inference chips and cloud.

Small language models and model merging.

Sohu, a chip built only for Transformers.

Photonic interconnects for AI data centres.

Clinical documentation AI.

General-purpose robot foundation models (π0).

Humanoid robots for the home (NEO); started in Norway.

Customer-support AI agents.

General-purpose humanoid robots with its own Helix model.

GPU cloud and servers.

The main hub for open models and datasets; trains SmolLM.

Multimodal models (Reka Core, Flash).

Conversational character platform; founders returned to Google in 2024.

State-space-model research lab; ships real-time voice models.

Interpretability research and tools (Ember).

AI mathematician combining language models and formal proof.

Foundation models for molecular structure and drug design (Chai-1).

Robot foundation models for commercial tasks.

Spatial intelligence: models that generate and understand 3D worlds.

Listed AI drug-discovery company (NASDAQ: RXRX).

Thermodynamic computing hardware for probabilistic AI.

Devin, the autonomous software engineer; bought Windsurf in 2025.

Mathematical superintelligence; its Aristotle model writes formally verified proofs.

Household robots from Cruise's co-founder.

Video generation.

Music generation models.

Open models (Hermes) and decentralised training.

Built Pi; now focused on enterprise AI after its founders moved to Microsoft in 2024.

Active-inference agents (Genius platform); listed on Cboe Canada.

Digit humanoid robots for logistics.

ServiceNow's research team (Apriel models).

Customer-service AI agents.

Independent image and video model lab.

Open video model Mochi.

Emotionally intelligent voice models.

Founded by two Transformer paper authors; builds open models.

Reconfigurable dataflow chips for AI.

Evaluates frontier models for dangerous autonomous capabilities.

Retrieval-augmented language models for enterprises.

Mojo language and the MAX inference stack.

Enterprise platform with its own Palmyra models.

Autonomous software development agents (Droids).

AI search and agents for enterprises.

Non-profit research on AI control and alignment.

Research on AI trends: compute, data and benchmarks.

Says it has one goal and one product: a safe superintelligence.

AGI lab built around program synthesis, from the creators of the ARC Prize.

John Carmack's AGI research company; Richard Sutton is a research partner.

Formerly Generally Intelligent; builds reasoning agents and tools for coding.

Agents that take actions in software; most of the founding team joined Amazon in 2024.

Long-context models aimed at automating software engineering.

Neuroscience-based machine intelligence (Thousand Brains Project).

Category-theory approach to structured, interpretable reasoning models.

Non-profit research on general-purpose reasoning and world models.

Diffusion-based language models (Mercury).

Coding agents with large-codebase context.

Safety-focused healthcare language models and agents.

Protein language models (ESM3).

A general robot brain that works across different robot bodies.

Robot foundation models for warehouses; founders joined Amazon in 2024.

Physics simulation and robot foundation models.

Video generation models (Gen-4) and research on general world models.

Non-profit building open tools to understand AI systems.

Non-profit alignment research and events.

Theoretical alignment research.

Research and field-building; co-created Humanity's Last Exam.

AI security, red-teaming and robustness.