Today, another question is increasingly making its way into specifications: "What is the environmental impact of this AI?"
It’s an important and positive change. Because a useful AI must not only be high-performing and secure, it must also be resource-responsible.
At Specgen, we see this evolution very clearly. In many projects, confidentiality remains essential… but it is no longer the only requirement. Organizations are now looking for leaner approaches, less unnecessary compute, fewer network exchanges, and better-optimized infrastructure.
Green IT: 5 simple (and very effective) criteria to put in your checklist
Here are Green IT criteria that we are seeing more and more in companies.
1) Limit unnecessary API calls
Each request has a cost: server-side compute, network traffic, latency, logs, retries, etc.
Reducing unnecessary calls (polling, permanent autopilot, automatic refreshes) is often one of the most immediate levers.
Capgemini is also promoting frugality approaches on the usage side. For example, a recent article explains how more “effective” prompting practices can help reduce consumption and emissions, while maintaining performance.
2) Avoid “lift-and-shift” without eco-design
Migrating to the cloud is not enough if the application remains inefficient: Capgemini points out that a “lift-and-shift” can cancel out some of the benefits if the architecture is not redesigned with a sustainable approach (green cloud architecture, better use of resources).
3) Optimize exchanges between layers (front end, back end, AI, storage)
Less network chatter = less consumption and often more speed.
Smart caching, batching, compression, suitable formats… these are technical details, but at scale, they change everything.
4) Choose transparent and committed infrastructure
Cloud providers and data centers are not all equal when it comes to environmental traceability. Clear commitments (targets, low-carbon energy factors, reporting) make governance easier.
5) Put people at the center of “costly” actions
One of the best Green IT principles: only trigger computation when it’s useful.
Our approach at Specgen: a resource-responsible (and controlled) AI
1) Zero permanent “autopilot”; every action is triggered by a human
Each AI action is triggered and guided by the user, "human in the loop".
Avoid running GPU/CPU power if nobody needs it.
We optimized Specgen by removing unnecessary automatic exchanges: no “background” requests running continuously “just in case”. Result: fewer calls, less computation, more control.
It’s a good Green IT practice… and a good business practice: AI actions become useful, explainable, and aligned with the real need.
2) No external API: why that’s a plus (for the environment and for security)
When a solution depends on external APIs, you add:
• extra network transfers,
• processing layers outside your control,
• latency and retries,
• dependencies that complicate governance.
By staying on a controlled chain, we reduce both risk and waste (traffic + compute), while simplifying compliance.
3) On-premises: GPU optimization to host “less resource-hungry”
For on-premises deployments, we optimized execution to make better use of GPU resources and avoid oversizing.
The objective is clear: enable hosting that is less resource-hungry and less costly, without sacrificing quality.
A specialized, properly sized AI often avoids redundant calls and unnecessary processing, whereas the uncontrolled use of “general-purpose” AI can lead to overconsumption. The main lever remains design and usage.
4) Cloud: French hosting at OVHcloud, with published (and comparable) indicators
For our clients in Cloud mode, we chose hosting in France, notably with OVHcloud. This choice is also based on published environmental indicators that are easy to compare.
For FY2025, OVHcloud reports in particular:
• Group PUE: 1.24 (energy efficiency). As a benchmark, the industry average is around 1.56.
• Group WUE: 0.34 L/kWh (water efficiency). An order of magnitude often cited for industry is around 1.8–1.9 L/kWh.
• REF: 100% (renewable energy factor, according to a standardized methodology).
Conclusion: “Green AI” = performance + frugality + control
What has changed today is that digital sobriety is no longer a “bonus”: it is a business requirement.
Choosing an AI tool therefore means making trade-offs across 3 pillars:
1. Security & confidentiality
2. Efficiency & operational value
3. Environmental impact & frugality
At Specgen, we believe we can do all three. Useful, high-performing, and secure AI… without automatic waste, and with an infrastructure aligned with sovereignty and environmental challenges.
Specgen