B3N Frontier

How can more computing fit alongside the people who rely on the same power and water?

Frontier is B3N's long-term research and venture programme. Its agenda has three connected workstreams: power for computing, cooling and water, and the data systems that make AI useful and trustworthy. It is at the agenda stage: the questions and methods are set out below, and we are looking for collaborators.

Propose a collaboration Read the research agenda

The problem Frontier studies

Every new data centre, AI service or computer lab needs electricity, cooling and data. Those resources are shared with homes, hospitals, farms and industry, and in many places, including East Africa, supply is limited. Adding computing without accounting for that pushes costs and shortages onto other people.

Adding it with those people in mind could mean running flexible work when power is plentiful, reusing heat, or siting capacity where it strengthens local supply instead of straining it. Which of these ideas hold up in practice, at what cost and for whom, is still an open question, and the one Frontier is set up to study.

Nothing on this page is a finding yet. Each workstream sets out what we would need to learn and how.

Workstream 1

Power and computing capacity

The question: Where and when can additional computing draw power without competing with homes and essential services, and could flexible computing demand help local supply instead?
Why it matters: Computing loads run day and night and can be large compared with a local grid. Where they sit and when they run decides whether they compete with other users or help use up surplus generation.
Evidence we would need: Grid capacity and reliability by location; the generation mix and any unused generation by time of day and season; tariffs; and the power profile of different computing workloads, especially which ones can move in time.
How we would investigate: A review of published research and interviews with operators; analysis of public and partner-provided grid data; a comparison of candidate sites; and simple models of shiftable workloads against supply.
A useful output: A siting and scheduling framework for small and medium computing deployments, published with its assumptions.
Who could contribute: Utilities and grid operators, energy developers, data centre operators and energy researchers.

Workstream 2

Cooling, water and heat

The question: Which cooling approaches suit a particular climate and water situation, and where could the waste heat be put to use?
Why it matters: Cooling choices trade electricity against water, and every method has costs. The right answer depends on local temperature and humidity through the year and on who else relies on the same water.
Evidence we would need: Local climate records across seasons; water availability and competing uses; measured performance of cooling options in similar climates; and nearby demand for low-temperature heat, such as crop drying, greenhouses or water heating.
How we would investigate: A review of engineering literature and published operator data; feasibility comparisons for candidate sites; models of energy and water use for each option; and small, tightly scoped pilots where a partner has a site.
A useful output: A comparison of cooling and heat-reuse options for named site conditions, with energy, water and cost estimates and the uncertainty in each.
Who could contribute: Facility operators, mechanical engineers, water authorities, farms and industries that use heat, and universities.

Workstream 3

Data systems for useful AI

The question: What does it take to build and maintain the data that makes AI reliable in a specific context, such as a local language, a sector or a single organisation?
Why it matters: AI systems are only as useful as the information they draw on. Quality, where data came from, permission to use it, upkeep and testing decide whether people can trust the answers, and these are often weakest where local context matters most.
Evidence we would need: Inventories of available datasets and their licences; how data is collected, permissioned and corrected in practice; and test results for AI systems with and without context-specific data.
How we would investigate: Case studies inside real organisations, including B3N's own products; test sets built with domain experts; and a comparison of upkeep approaches over time.
A useful output: Practical guidance and open test sets for building context-specific data, tried on real systems where partners allow.
Who could contribute: Organisations that hold domain data, language and subject experts, AI researchers and data protection specialists.

How the research will run

The methods below are proposed, not underway. Each step is cheaper than the next, so ideas are tested on paper before anyone builds anything. We plan to publish what each workstream finds, including what does not work.

Step 1ReviewPublished research, operator data and interviews with people who run real facilities.
Step 2AnalysisGrid, climate, water and dataset analysis for specific places and organisations.
Step 3ComparisonSide-by-side feasibility studies of options for a named site or use case.
Step 4ModelsTechnical and cost models that show the assumptions behind every estimate.
Step 5PilotsSmall, carefully scoped prototypes, only where a partner has a real site or system.

From a promising idea to something deployed

A technical answer is not enough. For every idea we also ask who would pay for it, who would run and maintain it, what it costs over its life and which rules apply. Frontier is a venture programme as well as a research one: ideas that pass both tests could become B3N ventures or partnerships.

What you could bring, and what we bring

We expect to agree each collaboration in writing: the question, who contributes what, how data is handled, who owns the results and what gets published.

Collaborators might bring

Data on grids, climate, water or a domain
Experience operating facilities or networks
A site or system where a pilot could run
A defined problem they need solved
Research expertise and supervision
Funding for a specific workstream

B3N brings

Engineers who build models, tools and prototypes
Experience running our own hosting and deployment platform
Commercial and deployment thinking from our ventures
Coordination, documentation and publication

Propose a question, a dataset, a site or a collaboration.

A short description and a link to relevant work is enough to start.

Propose a collaboration