Running Ray on the heata network isn’t just a technical test – it’s a working glimpse of what cloud computing can become: decentralised, efficient, and climate-positive.
Ray is an open-source framework that powers large-scale Python, machine learning and AI workloads. heata is a distributed compute provider that turns server heat into free hot water for households. Together, they form a powerful distributed computing network that delivers an 80% energy saving whilst helping families with their energy bills.
- We’ve integrated Ray, the powerful distributed computing framework, with heata’s in-home compute network.
- heata units (compute servers in people’s homes) run Ray workloads and simultaneously provide free hot water to households on the network by capturing the waste server heat.
- Ray’s fault tolerance and scalability make it ideal for our decentralised setup.
- Beta workloads include batch jobs, ML tasks, Python tasks, all running smoothly across the network.
- The model proves cloud computing can be local, resilient, and climate-positive.
Why This Matters
Data centres are one of the fastest growing consumers of electricity, and are expected to absorb 20% of all new generation capacity over the next 5 years to reach equivalent consumption levels as Japan (IEA, 2025). This electricity has a significant carbon footprint, and by some estimates, data centres now have a higher footprint than aviation (Climate IQ). Even by conservative estimates, the emissions of the big three hyperscalers are rapidly increasing as they race to build AI infrastructure; Google’s are up 50% since 2019; Microsoft’s 30% since 2020, and Amazon’s 182% since 2023, despite net zero ambitions.1
But with every challenge comes opportunity, and in data centres, that opportunity is heat - the waste heat from one large data centre could provide enough hot water for 100,000 homes. But with a centralised model, there’s no way of transporting that heat to the homes.
That’s why heata moves the compute to where the heat is needed – in people’s homes. In doing so, we use energy twice - once for compute, once for heating.
The Challenge
Whilst distributed computing makes heat capture a reality, it is not without its challenges - including coordination across multiple systems, ensuring fault tolerance, data consistency, latency in communication. These systems often operate across geographically dispersed nodes, which introduces complications like network failures, synchronisation difficulties, and the need for robust load balancing.
Despite these obstacles, a series of successful projects have made remarkable strides in advancing the field. Notable among them are the MapReduce framework by Google, which revolutionised big data processing; Apache Hadoop, which brought scalable and fault-tolerant data storage and analysis to the open-source world; and SETI@home, which pioneered volunteer computing by utilising idle home computers to analyze radio signals from space. More recently, Kubernetes has enabled flexible orchestration of containerised applications, marking a leap forward in operational efficiency and scalability.
Enter Ray
Ray is light, Python-native, and made for dynamic, parallel workloads – perfect for our network. It handles clusters with varying specs, copes with disconnects, and doesn’t need a mountain of setup. Unlike heavyweight orchestrators, Ray spins up with minimal code and scales from a laptop to a multi-node cluster. In our setup, each heata unit connects as a worker node - your job could be running across dozens of airing cupboards.
Ray also brings built-in fault tolerance. If a device drops out (say, due to a network issue), Ray automatically reroutes tasks. It checkpoints jobs, retries tasks, and makes sure work keeps moving. Combined with our own health checks and safeguards, the result is a resilient, self-healing distributed system.
In the beta
We've focused on workloads that benefit from being spread out: Python batch jobs, and machine learning optimisation. Ray abstracts away the complexity, so the workload doesn’t know it’s running on a network of servers in homes – it just sees CPUs and memory.
The Impact
Here’s where it gets exciting. Every CPU cycle on the heata network is doing two jobs: running a workload and heating water that would otherwise rely on the household's primary heating system (gas boiler or electric immersion heater). At the same time, we eliminate the cooling energy and water used by data centres. This results in a net reduction in energy use (at a system level) of 80% - delivering a carbon saving of 1 tonne per home per year - whilst, saving households up to £340 a year on their bills.
Plus, by distributing resources, we build more resilient systems and avoid local grid congestion.
Help test Ray on heata
We’re working with a handful of beta testers to refine the user experience and develop product functionality. If you’re a data scientist or work in machine learning and are passionate about energy use or climate change, we’d love to get you involved.
And of course, we’re already bringing our Ray offering to our customers to help them reduce their cloud based emissions and tell a great story about helping with fuel poverty. If you’re from an organisation with a data science or machine learning need and you’re interested in finding out more, get in touch.
