Runware
Serverless AI inference platform with custom hardware
Runware is an inference platform founded in 2023 that provides serverless compute and dedicated capacity for AI models. It designs its own liquid-cooled hardware modules called Sonic Pods to reduce costs and energy use, serving over one million developers globally.
- Founded in 2023 by Flaviu Rădulescu and Ioana Hreninciuc
- Raised $66 million including $50m Series A in Dec 2025
- Uses custom 1MW Sonic Pods with closed-loop cooling
- Serves over 1 million developers and 20 billion requests
- Pods sited at renewable energy sources to cut costs
What it is
Runware operates as an inference platform designed to lower the cost of running AI models. It offers a unified API for image, video, audio, and language models, allowing developers to switch between them easily. The service includes serverless compute for custom code and dedicated capacity options. The company claims to serve more than one million developers and processes billions of requests monthly. Its infrastructure relies on proprietary hardware modules known as Sonic Pods, which are deployed directly at renewable energy sources to minimise transmission losses and operational expenses.
How it came about
The company emerged from PicFinder, an image generation application launched shortly after Stable Diffusion. Co-founders Flaviu Rădulescu and Ioana Hreninciuc built the app to deliver results in under a second. They found that renting GPUs from various providers resulted in inconsistent performance due to unoptimised CPU clock speeds. In 2023, they began assembling their own liquid-cooled servers to improve efficiency. After realising the underlying infrastructure was more valuable than the consumer app, they pivoted to build Runware. The platform officially launched in October 2024, following significant software development throughout the first half of that year.
The compute
Runware designs and builds its own hardware, specifically the Sonic Inference Pod. Each pod is a 20-foot container with a capacity of 1 megawatt. The system uses a closed-loop liquid cooling method that recirculates water without consuming it or connecting to local mains. This approach avoids evaporative cooling, which typically uses large volumes of water. The software layer manages scheduling, scaling to zero, and multi-tenant isolation. Models remain warm on nodes within a distributed network, allowing traffic to reroute around failures automatically. This integration of hardware and software aims to increase GPU performance by approximately fifty percent compared to standard rented instances.
Where the GPUs are
The company sites its Sonic Pods directly at renewable power generation facilities, such as wind farms, solar parks, and hydroelectric plants. This placement allows them to purchase electricity at generation prices and eliminates transmission losses. The modular nature of the containers means they can be deployed quickly, often within a day, without requiring extensive land clearing or long permitting processes. By not drawing on local utility grids or water supplies, the infrastructure can operate in areas where community resources are already constrained. The team operates across ten countries and six time zones, with key engineering roles based in the United States and Romania.
Customers and deals
Runware serves as a backend for over one million developers and enterprise clients. Notable customers include global platforms like Wix and Higgsfield. The infrastructure supports more than 400,000 models and processes over 20 billion AI requests for hundreds of millions of end users. The platform allows users to bring their own containers and weights, running them without manual provisioning. This flexibility appeals to developers who need reliable, low-latency inference for applications involving image, video, and language processing.
Money and ownership
The company has raised a total of $66 million in funding. This includes a $50 million Series A round completed in December 2025. The round was led by Dawn Capital, with participation from Comcast Ventures, Speedinvest, Insight Partners, and a16z Speedrun. An oversubscribed pre-seed round occurred in December 2023. The capital supports the development of their custom hardware, software engine, and the deployment of their inference pods. The founders previously held leadership roles in gaming and infrastructure sectors, focusing on unit economics and margin optimization.
Its place in AI
Runware positions itself as a cost-efficient alternative to traditional cloud providers for AI inference. By controlling the entire stack from hardware to software, it aims to bring inference costs close to the price of electricity. Its focus on renewable energy and water-free cooling addresses environmental concerns associated with large-scale data centres. The company’s modular approach allows for rapid deployment in diverse locations, potentially decentralising AI compute infrastructure. It serves as a critical backbone for developers seeking predictable performance and lower operational costs for generative AI applications.
Questions people ask about Runware
What is Runware?
Runware is an AI inference platform that provides serverless compute and dedicated capacity. It uses custom-built hardware called Sonic Pods to run models efficiently. The service offers a unified API for various AI tasks including image and language processing.
How does Runware reduce costs?
The company designs its own hardware and sites it directly at renewable power sources. This eliminates transmission losses and allows buying power at generation prices. Their closed-loop cooling system also reduces water usage and operational overhead.
Who are the founders of Runware?
Runware was founded by Flaviu Rădulescu and Ioana Hreninciuc in 2023. They previously built PicFinder, an image generation app. Their experience in gaming and infrastructure helped them optimise the unit economics of AI compute.
What funding has Runware raised?
Runware has raised $66 million in total. This includes a $50 million Series A round led by Dawn Capital in December 2025. Other investors include Comcast Ventures, Speedinvest, Insight Partners, and a16z Speedrun.
Good reading and sources
There is no English Wikipedia article on Runware yet, so the facts on this page come from the sources listed above: runware.ai/about. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryNeocloud wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.
It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.