For a long time, when a company chose an artificial intelligence tool, the number one question was: “Where does my data go?”

Today, another question is increasingly appearing in specifications and procurement requirements: “What is the environmental impact of this AI?”

This is an important and positive shift. Because useful AI should not only be efficient and secure, it should 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 more frugal approaches, less unnecessary computation, fewer network exchanges, and better-optimized infrastructure.

Green IT: 5 Simple (and Highly Effective) Criteria to Add to Your Checklist

Here are Green IT criteria that we are seeing more and more often in companies.

1) Limit unnecessary API calls
Every request has a cost: server-side computation, network traffic, latency, logs, retries, and more.
Reducing unnecessary calls (polling, permanent autopilot, automatic refreshes) is often one of the most immediate levers.
Capgemini is also promoting more frugal usage approaches. For example, a recent article explains how more efficient prompting practices can help reduce consumption and emissions while maintaining performance.

2) Avoid “lift-and-shift” without eco-design
Moving to the cloud is not enough if the application remains inefficient. Capgemini points out that a “lift-and-shift” approach can cancel out part of the benefits if the architecture is not redesigned with sustainability in mind (green cloud architecture, better resource usage).

3) Optimize exchanges between layers (front end, back end, AI, storage)
Less network chatter means lower consumption and often better speed.
Smart caching, batching, compression, and appropriate formats may seem like technical details, but at scale, they make all the difference.
4) Choose transparent and committed infrastructure providers
Cloud providers and data centers are not equal when it comes to environmental traceability. Clear commitments (targets, low-carbon energy factors, reporting) make governance easier.

5) Keep humans at the center of “costly” actions
One of the best Green IT principles is simple: only trigger computation when it is actually useful.

Our Approach at Specgen: Resource-Responsible (and Controlled) AI

1) No permanent “autopilot”: every action is triggered by a human
Every AI action is triggered and controlled by the user: “human in the loop.”
This avoids running GPU/CPU power when no one actually needs it.
We optimized Specgen by removing unnecessary automatic exchanges: no continuous background requests running “just in case.” The result: fewer calls, less computation, and more control.
This is a good Green IT practice… and good business practice as well: AI actions become useful, explainable, and aligned with the actual need.

2) No external API: why this is a plus (for both sustainability and security)
When a solution depends on external APIs, it adds:
• additional network transfers,
• processing layers outside your control,
• latency and retries,
• dependencies that make governance more complex.
By keeping the chain under control, both risk and waste (traffic + computation) are reduced, while compliance is simplified.

3) On-premises: GPU optimization for more efficient hosting
For on-premises deployments, we have optimized execution to make better use of GPU resources and avoid oversizing.
The objective is clear: enable less resource-intensive and less costly hosting, without sacrificing quality.
A specialized and properly sized AI often avoids redundant calls and unnecessary processing, whereas uncontrolled use of “general-purpose” AI can create overconsumption. The major lever remains design and usage.

4) Cloud: French hosting with OVHcloud, with published (and comparable) indicators
For our cloud clients, 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 notably reports:
• Group PUE: 1.24 (energy efficiency). As a benchmark, the industry average is around 1.56.
• Group WUE: 0.34 L/kWh (water efficiency). A commonly cited industry range 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 is changing today is that digital frugality 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 it is possible to achieve all three. Useful, high-performance, and secure AI… without automatic waste, and with infrastructure that is consistent with sovereignty and environmental challenges.