A CTO's hard-won lessons on implementing sustainable technology practices, from lifecycle assessments to frontend optimization and the AI sustainability challenge.
Ludi Akue, CTO at Bpifrance Digital, shares seven critical lessons learned while implementing Green IT practices across her organizations. Drawing from her experience transforming Lunii's hardware operations and now leading digital transformation at Bpifrance, Akue reveals how the tech sector's rising emissions—now 6% of global emissions and projected to triple by 2030 due to AI—threaten our ability to meet Paris Agreement climate goals.
Operating in a Destabilized World
The foundation of Green IT begins with recognizing our new design constraints. Natural disasters are becoming more frequent and severe, directly impacting digital infrastructure. The UK's NHS hospitals experienced IT failures during 2022 heat waves, demonstrating how climate instability affects technology systems. This reality means emissions and planetary boundaries are no longer optional considerations—they're fundamental design constraints that software engineers must address.
Lesson 1: Assessment Challenge
Starting with Green IT is overwhelming. The first concept to master is Life Cycle Assessment (LCA), the standard method for evaluating emissions across your entire supply chain—from raw materials through manufacturing, usage, and disposal. You'll need to understand scope frameworks: Scope 1 covers direct emissions from your activities (like company vehicles), Scope 2 covers emissions from energy you consume (office electricity), and Scope 3 covers your entire supply chain including cloud providers.
The key insight is that perfect measurement isn't possible or necessary. Carbon footprint calculations are inherently relative, and your numbers will be wrong—that's acceptable. The goal is establishing direction, not precision. Start with carbon because it's simple and creates momentum, but remember that technology also stresses materials, water usage, and power grids.
Lesson 2: Infrastructure Reality
Cloud infrastructure isn't automatically green. Moving from Platform-as-a-Service to cloud infrastructure revealed that PaaS offers the least flexibility and transparency due to abstraction layers. Everything becomes opaque—scaling, database tuning, and underlying operations are hidden.
Understanding electricity mix becomes crucial. France's nuclear-powered grid is greener than Virginia's, affecting your carbon intensity. Cloud regions have different carbon footprints, so you can shape workloads by moving batch jobs to low-carbon hours or greener regions. Major cloud providers offer sustainability guidance—follow it. For on-premises operations, apply the same principles and seek efficient hardware like ARM processors.
Lesson 3: Software Architecture Paradox
Microservices can become distributed monoliths, creating technical debt that Green IT can help resolve. At Lunii, the architecture supported hardware production, multilingual audiobook publishing across 13 countries, and e-commerce operations. Using Domain-Driven Design, Akue made three strategic moves:
First, simplify by identifying the core domain (audio publishing) versus supporting domains (e-commerce, which could be delegated to platforms like Shopify). Second, migrate from microservices back to modular monoliths to reduce complexity. Third, refocus the entire sociotechnical team on the core domain, bringing simplicity that correlates with Green IT goals.
Lesson 4: Frontend Migration
Frontend optimization is often overlooked in Green IT discussions, yet webpages are energy and emissions sources. Over the past decade, webpage weight has grown fourfold, straining user devices and forcing more frequent charging and device replacements. This creates an equity issue—heavier webpages disadvantage users in low-network regions.
Tools like Lighthouse help optimize page speed and accessibility, but you must also consider page size, weight, and network throttling to understand real-world performance. Compress images, lazy load and defer scripts, and trim bundles. Where possible, use static rendering and edge caching to reduce data transfer—a major emissions contributor. Adopt efficient protocols like HTTP/2 and HTTP/3, and continuously measure frontend impact.
Lesson 5: FinOps Correlation Trap
FinOps doesn't automatically equal Green IT. Cheaper cloud regions are often more carbon-intensive, so moving workloads for cost savings can worsen environmental impact. Services like Instance Spot appear green by using only needed instances, but often rely on old virtual machines, making things worse.
True FinOps requires including impact metrics and measuring electricity consumption. Use cloud calculators and build sustainability considerations into procurement processes. The correlation between cost optimization and environmental optimization is not automatic—you must actively measure and verify.
Lesson 6: Enabling Change
Green IT is fundamentally sociotechnical—technical changes alone won't succeed without organizational transformation. Build external accountability by publicly committing to sustainability goals, similar to telling friends about gym commitments. This creates support when challenges arise.
Train teams on planet boundaries, system thinking, and new reflexes. Unlearn performance optimization habits and learn sufficiency principles. Create feedback loops with milestones, metrics, and celebrations. Integrate sustainability into delivery backlogs as regular user stories, not side quests. Translate company OKRs to include sustainability goals.
Distribute leadership by appointing champions across tech, product, design, and marketing teams. Let these distributed leaders build change collectively rather than centralizing mandates.
Lesson 7: Emerging AI Patterns
AI presents the most urgent sustainability challenge. Current AI governance, including the EU AI Act, barely addresses sustainability concerns while AI adoption accelerates everywhere. Agentic AI and embedded AI systems are proliferating with heavy training costs. Inference has surpassed training as the primary energy consumer, yet nobody discusses the two-to-three-year lifespan of GPU chips or the power grid stress from new data centers.
Product and tech teams build AI tools without considering necessity or environmental telemetry. Bpifrance promotes AI literacy through tools like PromptSage GPT, which helps users build sustainable prompts and traditions. Monitor AI usage with tools like EcoLogits, LiteLLM, and Langfuse to track token consumption and emissions.
Key Takeaways
The journey begins with acknowledging we operate in a destabilized world where tech plays a central role in meeting climate goals. The most important skill is building intuition for sufficiency—asking fundamental questions: Why am I building this? What am I building? Where and how am I running it? For whom?
Three pieces of advice guide the path forward: First, be curious and willing to step outside your comfort zone, as Green IT requires new thinking. Second, practice honesty in assessing impact and owning negative behaviors we've offloaded for years. Third, collectively push cloud providers, SaaS companies, and AI platforms for greater transparency.
Resources are abundant, but success depends on creating space for reflection and building new organizational norms. The goal isn't to treat Green IT as a separate initiative but to make it the normal way of working—a transformation that requires courage, education, and distributed leadership to achieve.

Comments
Please log in or register to join the discussion