Swiss Blueprint: Responsible AI starts at the chip level

As AI becomes ever more deeply woven into the fabric of everyday life, conversations continue to evolve around the importance of ‘responsible AI’. Those discussions have largely focused on the software itself and how it’s managed. They include pushes for model transparency, data governance and bias mitigation. These are necessary, but miss a critical area of responsible AI development.

Chip development

It is becoming increasingly clear that it is time for a discussion around the hardware itself. That hardware, namely chips, impacts every aspect of how AI behaves and relates to the world, including architecture, infrastructure, energy use and data flow. With an increase in energy costs and the potential environmental impacts of the datacentres underpinning our AI infrastructure, the development of chips  that can prioritise high-performance with a low-energy output are essential, both for sustainability and affordability. 

The rapidly increasing power demands of today’s complex AI models are already beginning to make continued AI growth a potentially damaging prospect to the environment. Datacentres and manufacturing require supports that are both highly water- and energy-intensive, resulting in more emissions and more negative environmental impact.

These increases are also resulting in rising costs, leading to more potential contributors being priced out of joining the market. Smaller regions, with less natural resources or investment muscle than the global leaders such as the US and China, are not able to keep up in the race. 

That’s why a different approach, where more sustainable, high-performance, precise and low-energy output chip production is a potential open door, will reduce emissions and provide a route for smaller regions around the world to enter into the market. Switzerland has spearheaded that model, approaching semiconductor development with these realities in mind. 

Energy efficiency 

As AI adoption expands globally, energy consumption is becoming one of the defining challenges of the industry. Training and operating large models require vast amounts of electricity, raising concerns not only about cost, but about sustainability and continued access to AI technologies. Recent reports have predicted that in the US alone, datacentres could consume upwards of 68 billion gallons of water a year by 2028 (AI Environment Statistics 2026: How AI Consumes 2% of Global Power and 17B Gallons of Water), with an estimated 3% of all electricity consumption around the world being tied to AI demands by just 2030. 

With skyrocketing costs of energy and water, the model of building increasingly larger and more powerful AI models, with the subsequent datacentres to support them, is an unsustainable approach. This is increasingly true for regions around the world without the access to capital to match the billions being spent to match the increase. 

Additionally a reduction in energy demands is not only more sustainable but can improve production resilience. By developing more high-performance and low-energy chips, production and models will be less vulnerable to volatile energy prices, infrastructure constraints and geopolitical disruptions. 

As global supply chains become more complex, energy resources more contested and digital sovereignty thrusted front and centre, efficiency could increasingly become a cornerstone of effective technology research and development.

Sustainability and chip functionality

Energy consumption is not just a byproduct of chip production, it is a constraint that influences deployment feasibility, environmental impact and system longevity. Producing chips with more precision, designed to power systems that have more specific functions, rather than wide-reaching ones, can alleviate these issues. 

On the environmental side, having a more precise function can reduce the need for more data exchanges. aIt can also cut down on datacentres and power demands, which will reduce costs and the environmental impact of a system. 

This low-power, high-performance approach can also improve functionality. For instance, systems that consume excessive resources limit where and how they can be deployed and operated. In the case of wearable sensors, the mobile aspect of the devices limits just how much energy any AI system can draw. Focusing instead of refining chips to the point where they are more precise and consume less energy can improve which functions those wearables can offer. Low-power AI systems often enable broader adoption and longer lifespans along with the reduced environmental impact.

Low-power AI architectures that are also designed to process data directly at the sensor level can dramatically reduce data exposure as well. When raw data never leaves the device, risks related to data interception, misuse or regulatory non-compliance are minimised by design — not by policy.

Specialized intelligence 

A wider array of developers can leverage this chip development approach than can follow the current market’s trajectory.  Large chip productions have attracted massive investments around the world, as companies and regions try to keep up the pace with supporting larger models and to get ahead of competitors. In fact, according to the Semiconductor Industry Association, over half a trillion US dollars  have been committed to be invested in the US alone in new semiconductor development and production projects.

Naturally investments of these sizes have left smaller nations priced out of the market. Switzerland, as an example, is in a situation that would hamstring more traditional pathways to global chip production relevance but is still finding a way to carve its own path. For one, it’s a smaller country, with a lack of natural resources essential to widescale production, limiting its capacity from the outset. On top of that, it’s not a member of the EU and while it has strong trade relationships with Europe, it again limits potential growth. 

However, with its approach of boiling chip production down to its finest point and creating low-energy, high-performance outputs, developers in the region are able to generate at a meaningful level while still operating within their means. These approaches aren’t meant to be a comprehensive solution, but rather are laser-focused on meeting clearly defined criteria at a high level. 

This carves out a role for developers that cannot generate chips at that same pace or scale, giving them a valuable secondary approach to focus on.

Blueprint for the future

The future of AI cannot be defined or assessed solely by the scale of models or computing power, the world just doesn’t have the energy availability to keep making that happen. Instead, it will be shaped by how well highly efficient and low-power models are able to augment what has been developed already, which will lead to a more sustainable technological future. Switzerland has provided a potential blueprint for a sustainable future of chip production. This simultaneously addresses the need for reducing energy consumption while providing an approach for smaller regions to support the global innovation landscape.

As the industry moves forward, the challenge will not be to build the largest possible AI systems, but to build the right ones and the right way. What drives those systems are the right chips that are developed and manufactured sustainably and that perform reliably.

Switzerland’s experience shows that this path is not only viable, but increasingly necessary for the next phase of AI and semiconductor innovation.

Alain-Serge Porret is vice president, Integrated & Wireless Systems, CSEM

See: Top 10 countries for AI adoption