xAI develops the Grok family of frontier AI models, trained end-to-end on the Colossus supercomputing cluster. Following the 2026 merger with SpaceX, xAI operates as part of a vertically integrated AI and infrastructure segment, with Grok embedded across the X platform, Tesla vehicles, and Optimus robots - one model family, deployed everywhere the group ships hardware or software.
- Valuation
- -
- Change
- +7.3%
- Founded
- 2023
- Headquarters
- Memphis, TN
- Employees
- 1,200+
About xAI
xAI develops the Grok family of frontier AI models, trained end-to-end on the Colossus supercomputing cluster in Memphis - one of the largest GPU clusters ever assembled, built specifically for this purpose rather than leased from a general-purpose cloud provider. Owning the full training pipeline is what lets xAI compress the time between model generations.
Following the 2026 merger with SpaceX, xAI now operates as part of a vertically integrated AI and infrastructure segment, with Grok embedded across the X platform, Tesla vehicles, and Optimus robots - one model family, fine-tuned per surface, deployed everywhere the group ships hardware or software.
Advance human scientific understanding through maximally curious, truth-seeking artificial intelligence.
How it got here
- 2023
Founded
Started to build frontier AI models with a distinct research approach and independent training infrastructure.
- 2023
Grok-1 released
The first model in the Grok family shipped, establishing the initial architecture and training approach.
- 2024
Grok integrated into X
Native AI features became available directly inside the X platform.
- 2025
Colossus online in Memphis
The dedicated training supercomputer came online, replacing reliance on rented third-party GPU capacity.
- 2026
Group-wide deployment
The SpaceX/xAI merger integrated Grok across the group's hardware surfaces - Tesla, Optimus, and X - under one model family.
Core pillars
Purpose-built compute
Colossus was designed and built specifically to train Grok, giving xAI full control over scheduling and scaling rather than sharing rented cloud capacity.
Multi-surface deployment
The same underlying model family runs inside X, Tesla's in-car assistant, and Optimus's decision-making, fine-tuned per surface.
Fast iteration
Cluster scale compresses the training time for each new Grok generation from months down to weeks.
Continuous retraining
Fleet and platform data feeds back into training runs, so the model improves from real-world deployment rather than static datasets alone.
Facilities
Colossus Training Cluster
A dedicated, purpose-built GPU training supercomputer with custom high-bandwidth interconnect and on-site supplemental power generation.
Research Offices
Core model research and engineering, working alongside the Memphis training infrastructure.
Scale
How it's kept safe
- Model releases go through an internal evaluation and red-teaming process ahead of broader deployment.
- API and product usage is subject to rate limits and content policy enforcement across every surface Grok is deployed on.
- Fleet-derived training data is handled under the same data-governance policies as the rest of the group's products.
Environmental note
Colossus supplements grid power with on-site generation, and successive Grok generations are evaluated for training efficiency, not just raw capability, as cluster scale grows.
Frequently asked
What is Grok?+
xAI's family of large language models, trained end-to-end on the Colossus cluster and deployed across X, Tesla vehicles, Optimus, and a public API.
How is xAI different from other AI labs?+
The main structural difference is owning the full training pipeline - a purpose-built cluster rather than rented cloud capacity - and deploying one model family across hardware surfaces most labs don't build themselves.
Is Grok open source?+
Some earlier model weights have been released publicly; the current frontier generation is served primarily through the API and integrated products rather than released as open weights.
How do I access the Grok API?+
API access is available directly from xAI, priced per token - see the product listing below for current rates.
