Runware Launches Serverless GPU Platform for AI with Prices from $0.63 per GPU-Hour

Runware’s managed GPU platform delivers AI compute at up to half the cost of hyperscalers, running on compact, modular data centres it builds in-house.

SAN FRANCISCO, CA, UNITED STATES, October 1, 2026 /EINPresswire.com/ -- Oct 1st 2026 — Runware today announced the public availability of Runware Serverless, a serverless GPU platform for running custom AI models, code and containers. GPU compute starts from $0.63 per GPU-hour, up to 50% less than typical hyperscaler rates.

The launch opens up the infrastructure behind Runware’s Models API to teams building their own applications. Developers choose their runtime, GPU configuration and scaling limits. Runware handles scheduling, autoscaling, worker lifecycle, execution and monitoring.
The platform supports all types of AI workloads, including image, video, audio, language and 3D models. It also runs fine-tuned and private models, multimodal pipelines and long-running GPU jobs.

A modular alternative to traditional data centres

Runware Serverless runs on Sonic Pods, compact modular data centres that Runware designs and operates for AI workloads. Each pod fits around 1,200 GPUs into a 20-foot shipping container, with servers, networking and cooling built in.

A Sonic Pod takes an average of three weeks to manufacture and can be deployed in days at a prepared site. Conventional data-centre construction takes years. Because the design is modular, Runware can add new GPU architectures without rebuilding the facility around them.

At large scale, Runware estimates this approach could cut the cost of building and deploying data-centre infrastructure to around 1% of a traditional facility, excluding the GPUs themselves.

Lower infrastructure overhead, combined with Runware’s inference software, means lower compute prices for customers. Serverless gives them that infrastructure without having to build the systems to run their workloads on it.

“Most GPU pricing carries the cost of the data centre around it,” said Flaviu Radulescu, [Title] at Runware. “We build our own in weeks, not years, so we can offer serverless GPU compute from $0.63 per GPU-hour and still run the whole platform for you. Developers bring their code and pick their GPUs. We handle everything underneath.”

NVIDIA GPUs matched to the workload

Runware’s hardware range covers NVIDIA L40S, RTX PRO 6000, H100, H200, B200, B300 and GB300 NVL72 configurations, with availability depending on deployment and capacity. Vera Rubin NVL72 allocations are planned from 2027.

Customers pick GPUs based on model size, memory, throughput and budget. Smaller inference workloads can run on a single GPU. Larger models and pipelines can use multi-GPU workers or rack-scale configurations. Runware’s NVL72 cluster storage is built to NVIDIA’s reference architecture.
Teams can scale on demand with per-second billing, or reserve capacity for an agreed production baseline. Current hardware options and rates are on the Serverless pricing page.

From custom code to managed production workloads

Developers bring Python code or a containerised workload, set its compute requirements and deploy through Runware’s SDK, CLI or API. Runware manages GPU placement, queues, workers and scaling around the application.

The platform is built for fast cold starts, efficient GPU utilisation and scale-to-zero. Startup optimisations cut the time it takes to bring workers online. Scheduling and batching keep allocated GPUs busy as demand changes. Teams can keep warm workers for latency-sensitive applications or release idle workers entirely. When an on-demand app scales to zero workers, idle GPU compute charges stop.

Observability covers deployments, worker state, GPU allocation, queue depth, requests, errors, logs and cost. Teams keep control of their code and runtime while Runware runs the managed execution platform underneath.

Additional options for infrastructure teams and model developers

Runware Compute offers dedicated bare-metal deployments and reserved capacity for organisations that want direct access to GPU infrastructure and control over their own software stack. Customers can run their own workloads on the hardware or use Runware’s managed Serverless platform, depending on how much of the operations they want to own.

Teams developing their own model can use Runware API Gateway to put it into production as a dedicated API. Runware works with them to define the API, select hardware and optimise serving, then handles hosting, scaling and operations. Models can be made public through Runware’s Models API and Playground, or kept private on a dedicated endpoint, with per-inference pricing and observability.

Runware Serverless is available now. Developers can create a Runware account and request Serverless access from their dashboard, sharing their workload and GPU requirements to get capacity allocated. Organisations that need larger GPU configurations, reserved capacity or dedicated infrastructure can talk to the team through Runware Compute.

About Runware
Runware provides AI infrastructure through a unified Models API offering access to more than 400,000 models, serverless GPUs and dedicated compute. Its platform is used by over one million developers, with customers including Higgsfield AI, Together AI, Envato and Wix. Runware supports image, video, audio, language and 3D workloads, alongside customers’ own models and applications. The company designs, builds and operates Sonic Pods, modular data centres purpose-built for AI inference, and the software that runs and scales workloads on them.

Press Team
Runware LTD
email us here
Visit us on social media:
LinkedIn
YouTube
X

Runware Serverless: Run AI Workloads at Half the Cost