Solmar Insights
Data center infrastructure is at the core of operations for multiple Elon Musk-led firms, including Tesla, X Corp (Twitter), and xAI, all of which are driving large-scale AI and compute deployments in the United States. Tesla alone has installed over 35,000 NVIDIA H100 GPUs, targeting nearly 90,000 H100-equivalents by the end of 2024 for AI training clusters underpinning its Full Self-Driving technology.
Key figures
35,000+ NVIDIA H100 GPUs deployed for Tesla AI
Projected 90,000 H100-equivalent GPUs by end of 2024
Over $1 billion planned investment in Dojo supercomputer
Data center locations and strategy
Elon Musk’s portfolio of companies is establishing and operating data centers across major US technology and industrial hubs. Confirmed sites include Austin, Texas; Sparks, Nevada; multiple California locations; Atlanta, Georgia; Hillsboro, Oregon; and Memphis, Tennessee, with additional footprints abroad in Shanghai, China. The network supports not only traditional data storage and processing, but also the demands of high-density AI workloads and real-time consumer services.
This geographic spread reflects balancing the need for regional resilience, energy sourcing options, and proximity to Tesla vehicle fleets and US consumer bases. The combination of established tech corridors and emerging energy markets is central to supporting high-performance compute and rapid data interchange for Musk’s interconnected ventures.
Each site must address substantial power, cooling, and connectivity requirements. Energy procurement, grid interconnection, and resilience against local constraints are key factors in site selection and operational buildout. The concentration in the United States signals sustained demand on US power grids and data center ecosystems, particularly for AI-driven sectors.
Tesla’s compute and storage backbone
Tesla’s data center architecture is engineered to manage enormous telemetry streams generated by its electric vehicle (EV) fleet worldwide. Data inputs span from vehicle sensors and navigation logs to over-the-air software operations and driving behavior analytics. These data centers serve as the backbone for Tesla’s Full Self-Driving development, as well as cloud-connected in-car features and owner-facing digital services.
Beyond automotive, these systems power Tesla’s broader ecosystem, handling real-time operations for Powerwall home batteries, Supercharger network management, and internal navigation. Rapid scaling of both storage and compute is integral to maintaining the competitive pace of feature deployments and AI model training.
This scale presents distinct challenges for power delivery, cooling systems, and on-premise hardware upgrades, especially given Tesla’s trend toward vertically integrated, own-designed compute components. As demand for high-frequency data exchange grows, network backbone investments and interconnection with hyperscale nodes become increasingly significant.
High-density GPU clusters and shifting allocations
Tesla’s reliance on NVIDIA GPU clusters remains central, with over 35,000 H100s dedicated to self-driving neural network development, rising to a targeted 90,000 GPUs by year end. The scale of this cluster underscores the company’s ambition to accelerate AI research and product iterations through brute-force compute power.
In 2024, a notable shift occurred when Elon Musk reportedly instructed NVIDIA to redirect GPU shipments intended for Tesla, valued at more than $500 million, to X Corp and xAI. Musk stated Tesla did not yet have the infrastructure to deploy the reserved GPUs, so they were better allocated to projects with ready capacity. This type of intra-group asset mobility, while optimizing resource use, also highlights ongoing bottlenecks in physical data center readiness versus cloud reservation rates.
Such GPU procurement and deployment cycles create downstream effects for supporting infrastructure, mainly data center power and cooling demands, utility cooperation, and interconnection lead times. For institutional investors and operators, the pace and flexibility of GPU utilization are as critical as hardware acquisition for maintaining AI development momentum.
Custom silicon: the Dojo supercomputer
Tesla is simultaneously building out its own AI training ecosystem with the Dojo supercomputer, a hardware platform tailored for large-scale machine learning on petabytes of video and sensor data. Central to Dojo are proprietary D1 chips, each designed to outperform commodity GPUs for specific neural net training tasks. Each D1 chip achieves a processing capability of 362 teraflops, with the modular architecture supporting efficient scaling.
Investments in Dojo and its underlying silicon are significant, with Musk noting intentions to spend over $1 billion in related R&D and infrastructure. For 2024 overall, Tesla forecasts capital expenditures surpassing $10 billion, a figure that includes broader company investments but underscores the financial scale and urgency behind its AI ambitions.
This focus on custom silicon aligns with a broader trend among hyperscalers toward proprietary hardware stacks, reducing dependency on commercial GPU supply chains. For digital infrastructure developers and utility-scale energy providers, new categories of workload, with distinct power envelopes and optimization requirements, are coming online rapidly as these purpose-built platforms mature.
Ecosystem integration and operational impact
The combined infrastructure initiatives across Tesla, X Corp, and xAI point to a vertically integrated approach, spanning everything from in-vehicle analytics and mobile apps to world-scale AI chatbot development (notably xAI’s Grok). Data center investments represent not only operational backbone but also competitive positioning for Musk’s entities in artificial intelligence leadership.
Tight integration of data, compute, and product delivery creates supply chain dependencies for power, land, thermal management, and network peering. Strong alignment between digital infrastructure and energy procurement strategies is required to avoid resource shortfalls and grid bottlenecks, especially in regions like Texas and California that feature both hyperscaler concentration and grid constraints.
For developers and investors tracking this sector, these expansion footprints signal ongoing demand for hyperscale-ready sites, grid interconnection, and power contracting. The tempo and technical specificity of Tesla’s and its affiliate’s compute deployments give clear indicators for forward power and fiber demand curves in US digital infrastructure hot spots.
What this means for buyers
Buyers should expect continued strain and competition for US data center real estate, power capacity, and interconnection access, particularly in states with established Musk company footprints. Operators with available grid ties and scalable power in regions such as Texas and Nevada are positioned to benefit from the cascade of GPU and custom silicon deployments. The Musk-linked demand signals intensifying near-term cycles of procurement and upgrade activity among hyperscalers and infrastructure PE sponsors. Early engagement around energy supply contracting and digital asset readiness remains critical to capturing growth from AI tenants with vertically integrated ambitions.
