TensorWave Raises $100M to Build World’s Largest Liquid-Cooled AMD GPU Deployment
May 14, 2025 -- TensorWave today announced it has raised $100 million in Series A funding to accelerate the deployment of the world’s largest liquid-cooled AMD GPU cluster, consisting of 8,192 MI325X GPUs.
The round was co-led by Magnetar and AMD Ventures, with additional participation from Prosperity7, Maverick Silicon, and Nexus Venture Partners.
Largest Liquid-Cooled AMD GPU Deployment
TensorWave is building the world’s largest direct liquid-cooled cluster of AMD GPUs, featuring 8,192 MI325X units. The system is designed to support high-density, high-throughput training and fine-tuning of large-scale AI models.
Direct liquid cooling enables the infrastructure to operate efficiently at full load, without thermal throttling. By improving thermal management, the system supports sustained performance across long training runs and inference workloads, while also improving energy efficiency and hardware longevity.
“When you deploy thousands of high-bandwidth GPUs, thermals aren’t a footnote, they’re a first-principles problem,” said Piotr Tomasik, President & COO of TensorWave. “We engineered our system from the ground up to make high-density, high-performance clusters viable; and liquid cooling is the unlock.”
Built for Modern AI Workloads
The MI325X, with 256GB of HBM3e memory, allows for larger models, fewer GPUs per workload, and more efficient pipeline design. TensorWave’s infrastructure is optimized for developers training multi-billion parameter models, fine-tuning LLMs, and scaling long-context inference tasks.
The deployment prioritizes raw, consistent performance over abstraction, giving developers direct access to the hardware and software stack. ROCm-based tools are fully supported, providing an open alternative to proprietary environments.
Expanding Ecosystem and Developer Alignment
TensorWave is committed to supporting open ecosystems and developer choice. The platform is designed to meet the needs of teams frustrated with limited access, locked-down environments, and inconsistent performance on traditional clouds.
As ROCm adoption increases and more open-source communities optimize for AMD Instinct GPUs, TensorWave is aligning with those trends to deliver infrastructure that supports experimentation, control, and rapid iteration. Developers working with open models, including Llama 3 and large-scale MoEs, can access white-glove support and predictable infrastructure at scale.
Use of Series A Funding
The $100 million Series A funding will accelerate the rollout of the MI325X cluster, further development of liquid-cooled infrastructure, and expansion of the engineering and support teams. The company is scaling to meet demand from enterprise AI teams and hyperscale customers.
TensorWave is focused on delivering performance infrastructure purpose-built for modern AI workloads—eliminating the limitations of overbooked, abstracted, general-purpose cloud platforms.
“We’re scaling fast because our customers are scaling faster,” said TensorWave CEO Darrick Horton. “We’re not here to offer another cloud—we’re here to build the one that AI actually needs.”
About TensorWave
TensorWave is the AI and HPC cloud purpose-built for performance. Powered exclusively by AMD Instinct Series GPUs, TensorWave delivers high-bandwidth, memory-optimized infrastructure that scales with the most demanding models—training or inference.
Source: TensorWave
