Portable’s AI and environmental impact statement

Portable will design, build and run AI that helps Australia thrive while driving our Scope 2 and 3 emissions to half by 2025 and achieve net zero carbon emissions by 2030.. This statement guides our practice and will evolve as industry standards mature.

As a certified B-Corp, Portable is built on the principle that business should be a force for good. We treat AI the same way: a powerful tool with enormous public benefit potential, but also a physical footprint that can’t be ignored. Like all powerful tools, it carries a physical footprint. From the megawatts that train large language models we use to the litres of water that cool data-centres, the impacts are real. Stewardship is non-negotiable.

We are optimistic about AI’s public-good potential and clear-eyed about its footprint. All computing, especially in the cloud, has an environmental cost, from electricity demand to the cooling water used in data centres. These impacts are real and growing, particularly as AI use expands across industries.

We believe that responsible AI practice demands environmental stewardship.

Our approach seeks to follow the lead of Microsoft’s pledge to be carbon-negative (remove more CO₂ than it emits), water-positive (return more water than it uses) and zero-waste by 2030; Google’s push for 24/7 carbon-free energy (every data-centre hour powered by renewables) by 2030; and Hugging Face’s habit of adding model-card “nutrition labels” that reveal the CO₂ released when each AI model is trained.

In practice, that means choosing the smallest-viable model for the task, partnering with vendors that publish strong sustainability metrics, and staying transparent about what we don’t yet know. We’ll keep learning, adjusting and sharing so our clients, partners and community can see what our progress towards truly sustainable AI looks like.

To reduce our environmental impact when using AI technologies, Portable commits to the following practices:

• Match the model to the task. We choose the smallest and most energy-efficient model (lowest life-cycle energy intensity covering both one-off training and cumulative inference based on the forecast usage) available that can still deliver the outcome. We do this by consulting industry model energy use leaderboards to evaluate the models we use where this information is available. This requires additional work, but we believe it’s worth it.
• Run on the cleanest power we can. We’ll prioritise cloud regions that show the highest renewable mix (≥90% where possible) while keeping data in Australia. As providers hit their 2030 sustainability milestones, we’ll keep re-evaluating.
• Disclose the numbers. Portable will report the annual energy, carbon and water footprint of AI production workloads we develop by tracking AI system usage and, where feasible, convert this into estimates of CO₂-equivalent emissions and water usage. We aim to provide a short annual summary that can be included in our broader sustainability reporting.
• Become ISO 42001 ready. Embed the ISO/IEC 42001 AI System Impact Assessment process, including environmental factors, into every production deployment by 2027.
• Stay informed. We actively monitor industry research and vendor announcements regarding the energy and water use of AI systems. We recognise that current data is often incomplete or opaque and are committed to advocating for better transparency.
• Ask for transparency. We write to vendors (such as OpenAI and other model providers) to request environmental impact data, and encourage clients to do the same. We believe that consistent demand for transparency can help shift the industry.
• Support impact reporting, offsetting and remediation. For clients who wish to go further, we can support you to include estimates of AI systems energy and water usage to include in ESG reporting and to enable carbon- or water-positive remediation. We will work with clients to identify appropriate carbon or water offsetting options and support them in reporting on these activities as part of their sustainability commitments.
• Publish learnings, tools and policies. We will share our AI sustainability methodology, encouraging sector change and reinforcing leadership as part of the B-Corp community working towards Net Zero by 2030 goals.

The science and accounting of AI’s environmental impact is still emerging. Public benchmarks exist for some models, but provider-level and task-specific data are often hard to access. Energy usage varies widely based on the infrastructure used, model type, use case, inference, and region among other factors. We acknowledge these limitations and where we estimate impact, we do so transparently, using best-available figures and stating our assumptions.

We’ll keep learning, adjusting, and sharing — and we invite our partners to do the same. Let’s set a new standard for sustainable AI together.