Runware Delivers Modular Data Centers Built Specifically for AI Inference at up to 10X Lower Cost

Purpose-built modular hardware delivers 2x inference throughput at 30-90% lower cost than traditional inference providers, with capacity coming online starting with 160 locations in the United States and Europe in H2 2026.

Runware, an AI inference and AI-native compute company, has unveiled the Sonic Inference Pod, a modular data center designed and built to rapidly increase access to AI inference infrastructure for model providers. Unlike traditional data centers, which take years to build and cost $9-19 billion per gigawatt, Runware’s Sonic Inference Pod delivers 1 megawatt of compute in a 20-foot shipping container, with custom servers, PCBs, and a closed-loop cooling system engineered for inference performance. The result is facilities costs that are up to 100x lower than a traditional gigawatt-scale data center build, a saving Runware passes directly to clients through its GPU compute pricing.

Runware is deploying a network of dozens of pods and tens of thousands of GPUs across the United States and Europe in H2 2026 based on customer demand, opening access to its custom-built hardware beyond its own platform for the first time.

Purpose Built for Inference

Most data centers were designed for general computing, not AI. GPUs draw 8-20 times more power and need far more cooling than the servers those facilities were built around, which is why companies running GPUs in standard data centers often end up with a handful of racks in use while the rest of the floor sits empty. Runware took the opposite approach, stripping out every component a data center needs for general workloads but not for inference:

• Dense, custom hardware: Custom servers, PCBs, racks and data center architecture built for GPU workloads pack significantly more compute into a fraction of the footprint of a traditional data center.
• Lower-cost economics: As an industry standard, operations and facilities costs typically account for around 35% of inference costs. Runware’s model cuts that to low single digits, delivering inference at 30-90% lower cost per GPU-hour than any peer, across all standard NVIDIA GPUs.
• Closed-loop, water-free cooling: Traditional data centers rely on evaporative cooling systems that account for more than 560 billion liters of water use globally each year. Sonic Inference Pods recirculate 1.5 cubic meters in a closed loop while holding GPU temperatures within 2°C of target conditions.
• Deploy anywhere power exists: Because each pod is self-contained and doesn’t require any changes to the power grid, Sonic Inference Pods can be sited wherever power exists, starting with renewable energy generation sites such as solar parks.
• Ready for sovereign deployments: Because pods can be deployed within any national border or jurisdiction, organizations with data residency or regulatory requirements can run inference on dedicated, locally sited infrastructure.

The first nodes will support the full breadth of Runware’s model catalog, including frontier open-source LLMs, specialized fine-tuned models, and Runware’s own media generation capabilities across image, video, and audio.

Deployment and Availability

Sonic Inference Pods ship ready to deploy and are live today across the United States and Europe. Each pod joins a distributed inference network managed as a single fabric, automatically rerouting requests if a pod goes offline to optimize for latency, cost, and availability.

Companies can access capacity on Sonic Inference Pods today, deploying their own models and AI workloads through Runware’s serverless infrastructure while Runware handles provisioning, scaling, and operations. Runware is accepting inquiries from frontier labs, AI studios, enterprises, and organizations with sovereign AI or data residency requirements, with capacity reservations open at runware.ai.

Video

Key Quotes

“After 20 years running large-scale infrastructure, I knew regular data centers couldn’t scale inference – the power and cooling simply aren’t there,” said Flaviu Radulescu, co-founder and CEO of Runware. “So we redesigned from first principles: only the components inference needs, liquid-cooled GPUs, and a modular format we can place wherever power already exists. This way we can produce and deploy 1 GW of compute faster than any other infrastructure provider, with built-in redundancy and real-time routing to the most cost-efficient GPU.”

“We started with Runware’s Model API, but quickly expanded into their inference infrastructure because of the scale and efficiency they could deliver,” said Deykhan Ten, VP of Strategic Partnerships at Higgsfield. “Our models reach millions of users every week, so reliable capacity and cost-efficient inference are critical. The lower our inference costs, the more value we can pass on to our users. Runware understands that deeply, which is why we work closely with them on our most important model launches.”

Key Link

More information about the Sonic Inference Pod from Runware