Compute efficiency — Less energy per useful outcome.
Examine workload optimisation, model selection, equipment utilisation and measurement. Prompt: Can a useful task be completed with fewer resources without reducing needed quality?
CHALLENGE 02 / 05 · ENERGY
What Does an Answer Cost?
Every AI answer begins long before a response appears on a screen. It depends on electricity generation, transmission, computing hardware, cooling and human decisions about how resources are shared.
Explore the question
As artificial intelligence expands, how can we power useful intelligence without treating energy, grid capacity or local communities as unlimited?
Mars is the challenge. Earth is the mission.
MARS CHALLENGE 2027
How Much Intelligence Can the Planet Sustain?
Mars is the challenge. Earth is the mission.
02 / ENERGY
AI Has a Power Supply
AI is often discussed as code or computational performance. But every training run, inference workload and network transaction takes place inside a wider energy system.
Electricity must be generated, transported and delivered when computing facilities need it. Cooling and supporting equipment also use power. Local grid constraints, price volatility and environmental impacts shape what can responsibly be deployed.
Five connected lenses:
The goal is not to claim that all AI use has the same footprint. It is to understand the conditions that make particular uses more or less demanding.
Connected system · conceptual diagram
- Generation
- Grid
- Data centre
- AI service
- Cooling & heat
- Community
| Lens | The question to investigate |
|---|---|
| Electricity demand | When, where and how much power does a workload require? |
| Grid capacity | Can the existing network support new demand without undermining other needs? |
| Flexibility | Which tasks can shift in time or place without sacrificing essential service? |
| Resilience | How does a system operate when generation or supply is interrupted? |
| Useful heat | Can rejected heat become a local resource rather than an unmanaged by-product? |
03 / ENERGY
More Intelligence. At What Energy Cost?
An efficient server does not automatically produce an efficient system. A low-carbon electricity supply is not automatically available at every hour. A facility that secures power can still affect grid expansion, land use, water, investment priorities and the communities around it.
In this challenge, teams investigate trade-offs instead of searching for a single magical technology. They ask which uses are worthwhile, how demand might be managed and what counts as a fair and credible energy decision.
Guiding question: How can AI services deliver meaningful value while respecting physical energy limits, local priorities and long-term resilience?
04 / ENERGY
Find a Problem Worth Solving
Flexible demand — Intelligence that works with the grid.
Investigate scheduling, location and demand response for non-urgent workloads. Prompt: Which tasks can move to periods of greater clean-power availability?
Resilient systems — Service under interruption.
Explore storage, microgrids, prioritisation and graceful degradation. Prompt: What must keep working when electricity is scarce?
Heat recovery — Rethinking waste.
Explore whether data-centre heat can meet an actual local demand, subject to temperature, distance, infrastructure and economics. Prompt: When is recovery useful, and when is it not?
Energy justice — Who gets capacity and benefits?
Examine siting, affordability, transparency and trade-offs with other users. Prompt: Who decides which demands deserve scarce network capacity?
05 / ENERGY
On Mars, Energy Cannot Be an Afterthought.
A hypothetical Martian settlement operates with finite generation, storage, replaceable components and maintenance capacity. Critical systems compete for energy; failure has direct consequences.
Earth-Back Thinking translates that constraint into a design discipline:
Mars is a thought experiment and an extreme learning environment, not a claim that Martian equipment can simply be transferred to terrestrial grids.
Understand Earth What service needs energy, and who depends on it?
Apply Mars constraints What if capacity, storage, spares and supply reliability were limited?
Design the response What can be reduced, shifted, recovered, repaired or prioritised?
Return to Earth What would be technically, socially and economically useful here?
06 / ENERGY
What Could You Build?
Illustrative starting points — not mandatory projects or validated solutions:
A worthwhile response might be a hardware concept, algorithm, service, policy framework, analytical model or combined proposal. Teams must state assumptions and show how evidence could test them.
01 / Illustrative concept · not a validated solution
Energy-aware AI scheduler: a conceptual tool that aligns deferrable computing tasks with available electricity, tracking service quality and system constraints.
02 / Illustrative concept · not a validated solution
Resilient microgrid scenario: a simulation of how a critical learning or community service could prioritise loads during supply disruptions.
03 / Illustrative concept · not a validated solution
Heat-use feasibility map: an assessment of where computing waste heat might match a nearby heat demand and where constraints make it impractical.
04 / Illustrative concept · not a validated solution
Energy transparency dashboard: a prototype showing understandable resource-use and uncertainty indicators for a selected AI service.
05 / Illustrative concept · not a validated solution
Community trade-off model: a participatory framework comparing the benefits and costs of alternative infrastructure decisions.
Illustrative concept · not a validated solution
07 / ENERGY
Energy Is Everyone's Problem
Power engineering studies grids and generation; computer science investigates computation; economics addresses prices and incentives; environmental science evaluates resource effects; design clarifies how systems are experienced; law and public policy address accountability and rights. Social sciences help teams understand trust, distribution and community interests.
You do not have to build an AI model to contribute to the AI energy challenge.
- Engineering & computing
- Environmental sciences
- Economics & management
- Law & public policy
- Design & architecture
- Social sciences & humanities
08 / ENERGY
Show Your Reasoning
A team can make its work visible through eight recommended evidence areas:
Define the energy problem and its users.
Map supply, demand, dependencies and affected people.
Document evidence, uncertainties and alternative explanations.
Establish design criteria and explicit trade-offs.
Develop a testable proposal or prototype.
Compare scenarios and practical constraints.
Explain potential value and limitations on Earth.
Defend decisions, feedback and next iterations.
These are editorial guidance for investigation, not confirmed mandatory 2027 deliverables, scoring criteria or competition rules.
09 / ENERGY
Connected to the Other Four Challenges
The five challenges define what teams investigate; the four Mars Challenge missions define how participants enter the shared global experience.
Water
Cooling and energy technologies depend on available water and local conditions.
Infrastructure
Location, grid connection and facility design affect energy demand and access.
Materials
Batteries, cables, chips and power equipment depend on physical material supply chains.
Humanity
Energy choices have consequences for affordability, fairness, safety and agency.
10 / ENERGY
A Challenge for Education, Industry and Society
Students and educators can investigate meaningful questions and build interdisciplinary proposals. Energy and technology organisations may contribute expertise and real constraints. Universities, public institutions and communities can help examine feasibility, governance and local consequences.
This page is an educational exploration hub. It is not an official call for applications. Dates, eligibility, participation requirements, selection and partner arrangements for 2027 must be checked against published programme rules.
11 / ENERGY
Intelligence Needs Energy. Responsible Intelligence Needs Judgement.
The question is not only whether we can power more computation. It is whether we can make better decisions about the systems, places and people that power it.
What does an answer cost — and who should decide whether it is worth it?