xAI

Full briefing

xAI was founded by Elon Musk in March 2023 with a stated mission to "understand the true nature of the universe." The company assembled a founding team largely poached from DeepMind, OpenAI, and other frontier labs, including Igor Babuschkin (formerly DeepMind), Tony Wu, and Greg Yang. xAI's first public model, Grok-1, was released in November 2023 with 314 billion parameters — at the time one of the largest openly-disclosed transformer architectures — and made available initially as a real-time search-enabled chatbot exclusively to X Premium subscribers. xAI's defining infrastructure asset is the Colossus supercluster, built in Memphis, Tennessee, at a former industrial facility. Phase 1, completed in September 2024 in approximately 122 days from groundbreaking — a speed widely noted in the industry — deployed 100,000 NVIDIA H100 SXM GPUs networked using NVIDIA InfiniBand and Arista Ethernet switches. This places Colossus among the top three largest single-site GPU clusters in the world at launch. xAI subsequently announced Phase 2, targeting an additional 100,000 NVIDIA H200 and Blackwell B200 GPUs, bringing the total cluster to 200,000 high-end GPUs — a scale that would represent roughly $10–15 billion in GPU hardware alone at prevailing list prices. xAI's Grok model series is tightly integrated with X (formerly Twitter), giving it a unique data moat: real-time access to the full firehose of X posts for pre-training and fine-tuning. Grok-1.5 Vision (April 2024) added multimodal capabilities; Grok-2 (August 2024) positioned competitively against GPT-4o and Claude 3.5 Sonnet on standard benchmarks. The company also released Grok-1 weights as open-source under Apache 2.0 in March 2024, unusual for a frontier lab of its scale, though subsequent Grok versions have remained proprietary. xAI raised $6 billion in a Series B round in May 2024 at a $24 billion valuation, and subsequently merged with X Corp in March 2025 at a combined $80 billion valuation. This merger creates a vertically integrated entity where AI inference (Grok API), distribution (X's 500+ million users), and data collection (X's firehose) are controlled by a single organization — a structure with no direct parallel among AI labs. As one of the largest NVIDIA GPU buyers in the world, xAI's procurement decisions have meaningful supply chain implications. The company's H100 cluster consumed a significant share of NVIDIA's HGX H100 SXM allocations in 2024, and its Blackwell B200 orders represent a major portion of early B200 production. xAI's Memphis site uses NVIDIA InfiniBand NDR (400 Gb/s) for GPU-to-GPU communication and Arista 7800-series switches for the Ethernet front-end fabric — making it a reference customer for both NVIDIA's networking and Arista's AI cluster switching portfolio.