By Virginie Corraze, Head of Engineering Centers of Excellence, Amadeus.
Carbon measurement in digital systems is growing in importance. Just as data fragmentation before it moved from a niche concern to a central topic of conversation with the growth of AI, the process of tracking and quantifying the greenhouse gas emissions generated by digital infrastructure, software, and data usage is taking center stage.
Yet, even as effective measurement becomes increasingly prevalent, this information can often fail to translate into meaningful action. The issue is no longer awareness or intent. Instead, it is clear that many of the foundations of how we measure carbon emissions today were designed for a different generation of systems with different objectives.
Here we look at why, without stronger, open-source foundations, attempts to optimise, govern, and ultimately reduce emissions risk remaining limited in their effectiveness.
A challenge shaped by digital complexity
Most carbon accounting tools were built for a different era. They work reasonably well for high-level reporting: infrastructure totals, energy bills, cloud invoices, and average carbon factors. However, modern digital systems have evolved rapidly. Applications scale dynamically, share infrastructure, evolve continuously, and rely on complex data and AI pipelines.
In these environments, high-level aggregated measurement can limit visibility into what is driving impact.
Organisations may find it hard to see which applications, architectural decisions, or usage patterns need closer attention when emissions are reported at provider or data-center level. In complex portfolios, improvements in one area can also influence emissions elsewhere, reinforcing the need for a more complete system-level view.
The result is a familiar pattern: sustainability teams can report numbers, engineering teams focus on performance and delivery, finance teams manage cost. It’s quite common to see each of these teams operating with different views of carbon at the level where real decisions are made, rather than working from a shared, integrated perspective.
Why we need to go beyond efficiency gains
Cloud providers have made significant progress on efficiency gains, including better utilisation, elastic scaling, and lower average Power Usage Effectiveness (PUE). These advances matter. But efficiency does not automatically lead to lower emissions at application or portfolio level.
Demand is growing faster than efficiency improves. More services, more data, more traffic, and now AI workloads are overwhelming per-unit gains.
Software design choices – from polling frequency and data duplication to model size and over-engineering – silently drive compute, storage, and energy use. Yet software’s role in environmental impact remains underestimated and largely invisible in today’s reporting.
This is why so many sustainability initiatives plateau. Organisations optimise what they can see, not necessarily what actually matters. Without granular, application-level insight across the entire portfolio, it is impossible to understand whether emissions are truly falling or simply being redistributed.
From measurement to governance
The next phase of carbon measurement must move beyond reporting and into governance. Measurement alone does not change outcomes unless it is embedded into how systems are designed, funded, and operated.
This requires three shifts.
- First, carbon data must be traceable and consistent at the application portfolio level. Teams need a shared view that connects emissions to real workloads, not abstract infrastructure totals. Tools like Carmen, built around open standards such as the Green Software Foundation’s Software Carbon Intensity specification and Impact Framework, point toward what this looks like in practice: portfolio-level visibility that can scale across heterogeneous, multi-cloud environments without relying on black-box assumptions.
- Second, accountability should follow on from insight. It’s essential to have clear ownership at product and application levels, including responsibility for technical debt and decommissioning.
- Third, organisations need to start thinking in terms of carbon budgets. Just as financial discipline emerged when costs became visible, planned, and constrained, sustainability will only scale when trade-offs are explicit. Scaling one workload should force decisions elsewhere. Without constraints, growth will always erase efficiency gains.
Why openness matters
Trust and comparability are becoming non-negotiable. As regulatory scrutiny increases and disclosures mature, organisations must be able to defend how their numbers are produced. Opaque, vendor-specific calculations undermine credibility and slow industry learning.
This is where open approaches matter. Open standards and open-source tools allow shared assumptions, transparent methodologies, and collective improvement. They reflect the reality that software sustainability is a challenge for the whole industry. No single organisation controls the full stack, and no single tool will solve it alone.
The real value of initiatives like Carmen moving into open ecosystems is not the tool itself, but the signal it sends and the subsequent actions it enables. Sustainability needs shared foundations if it is to scale, and it needs many of us to collaborate to achieve scalability.
Keeping up with reality
As digital systems continue to scale, carbon measurement faces an inflection point: whether it evolves into a disciplined operating practice or remains a reporting exercise.
Furthermore, it’s interesting to consider how the measurement challenge is evolving. AI introduces new layers of complexity and industry discussions are ongoing about how to define robust, practical standards. This uncertainty is not a reason to delay action, but a reminder that collaboration and transparency are essential.
And carbon is only part of the picture. The water consumption of AI infrastructure and its broader environmental impacts, including biodiversity, demand a wider lens.
Carbon must be treated like cost: visible, allocated, governed, and constrained. Engineering, finance, and sustainability need to converge around common signals and shared accountability.
This is how measurement will stop being an end in itself and start becoming the mechanism through which organisations make better, harder, and ultimately more sustainable decisions about the software they build and run.
