Balancing AI Innovation and Sustainability

The accelerating adoption of artificial intelligence has brought with it an uncomfortable paradox: the same tools promising to solve complex global challenges are themselves voracious consumers of electricity and water. As awareness of AI’s environmental toll grows, so does resistance to its unchecked expansion. Yet for many professionals and individuals, simply abandoning digital assistants is not a realistic choice. The practical question therefore shifts from whether to use AI, to how it can be deployed in a way that satisfies concerns around sustainability and data privacy.

The promise and limits of purpose-built green search

A logical starting point for many is the web. A search for sustainable AI tools often leads quickly to Ecosia, the environmentally conscious search engine that has branched into AI-powered answers. The service is free, operates entirely on renewable energy, and deliberately employs compact, energy-efficient models. In principle, it sounds like an ideal compromise.

In practice, however, the AI’s limitations emerge with more demanding tasks. Straightforward queries are handled well, but more intricate research prompts can trip up the system. It may return inaccurate information or fail to grasp the question entirely. While Ecosia’s mission remains admirable, such performance is difficult to accept in a tool built specifically for web search. The search for a robust alternative continues.

Viro and the shallow end of green marketing

A deeper look at the market reveals a growing number of platforms eager to brand themselves as sustainable. Broadly, they fall into two camps. The first is represented by services like Viro, whose public messaging emphasises green values without offering much detail. The company states that it channels revenue toward supporting renewable forms of energy and claims to display transparently how much electricity was consumed during a user’s chat session. It also reports having financed around 14,000 kWh of renewable energy to date.

While that figure sounds substantial in isolation, it is almost negligible within the context of industrial data centre operations. In a country like Austria, where electricity prices sit at the higher end of the European spectrum, 14,000 kWh of green power costs roughly 2,000 to 3,000 euros. For a company marketing sustainability as a core virtue, this represents a modest outlay. More striking still is the contradiction at the centre of the offering: Viro routes queries through large language models from ChatGPT, Gemini, and Claude, products built by technology giants not known for exemplary environmental performance. On the positive side, the platform is generally free and ad-supported, with an optional subscription priced at $1.

Services that build sustainability into infrastructure

A second group of providers takes a more rigorous approach. GreenPT is a case in point. All its AI models are hosted on European infrastructure running on 100 percent renewable electricity, and the service is fully compliant with the General Data Protection Regulation. The data centres rely primarily on air cooling, which drastically reduces water consumption compared to conventional liquid-cooled facilities. Moreover, GreenPT deploys open-source models, avoiding dependence on the data-collection practices of major US-based corporations. These benefits are tied to a monthly subscription costing between 4.50 and 17.50 euros.

Behind the scenes, GreenPT’s infrastructure runs on Scaleway’s servers. For technically proficient users who do not generate enough queries to justify a flat monthly fee, an even more cost-effective path exists: using Scaleway’s API directly and paying solely for actual usage. This route also opens access to highly capable models such as glm-5.2, whose performance leaves little to be desired within current technological limits. A similar pay-as-you-go model can be achieved through routing tools like Requesty, provided users consciously select server clusters with a high share of green electricity.

The landscape of supposedly sustainable AI is broad and growing, but genuine resource efficiency varies widely. Which option represents a meaningful compromise depends heavily on individual requirements, technical comfort, and how much one is prepared to pay.