For banks, the crucial question when deploying artificial intelligence (AI) often begins long before a language model gives its first answer: Which information may actually be entered? As long as customer data or confidential business documents remain excluded, much of the potential can hardly be unlocked.

That is exactly where Carla Bünger, co-founder and vice president of Phoeniqs, sees the business opportunity for her company. It operates cloud and AI infrastructure for firms with high requirements for security and data control. With the rise of generative AI, demand has accelerated sharply, she says in an interview with finews.

Sensitive Data Stays on Site

The principle: Phoeniqs runs publicly available language models on its own infrastructure in Switzerland. According to the company, processing of the requests, known as inferencing, also takes place domestically. This is intended to allow customers to apply AI to confidential information without transferring it to a foreign AI service.

According to Bünger, a range of models is available, including Apertus, which was developed in Switzerland. Companies can therefore decide which model to use depending on the task.

Phoeniqs aims to complement existing cloud offerings. Many customers still use the major international platforms, Bünger explains. Particularly sensitive or strategically important data, by contrast, is processed on the Swiss provider's infrastructure.

The Next Bottleneck is Computing Power

Bünger expects demand to rise further with the next stage of AI development. As digital agents take on multi-step tasks and process larger volumes of information, infrastructure requirements will also grow.

Phoeniqs' supercomputer. (Image: courtesy)

Many banks initially built up their own computing capacity. In her view, such installations could reach their limits in the next twelve to eighteen months. For Phoeniqs, this opens up a market for additional, externally sourced computing power.

To that end, the company relies on high-performance computers and a self-developed operating system. In June, Phoeniqs commissioned a commercial AI supercomputer at the Uptown Basel innovation campus in Arlesheim. According to the announcement on the commissioning, the infrastructure is aimed among other things at companies and public institutions that want to run AI applications under Swiss law.

Sunrise Opens up Further Access to Business Customers

Phoeniqs relies on partners for marketing. Sunrise plays a central role: the telecom group brings access to business customers, while Phoeniqs provides the underlying infrastructure.

The two companies announced their strategic partnership in May. The offering is intended to enable companies to use AI whose infrastructure is developed, hosted and operated in Switzerland.

For Phoeniqs, the collaboration offers the opportunity to reach more customers without having to build up a complete sales operation on its own. Bünger also sees this as a fundamental strategic direction: the company deliberately works with partners that want to change established market structures.

First Structured Financing Round

Now financing is also to be put on a broader footing. Phoeniqs has so far grown mainly out of ongoing operations, Bünger says. In addition, there have been individual investors. However, the company has not gone through a classic sequence of financing rounds.

That is set to change. According to Bünger, Phoeniqs is working with two investment banks on its first structured capital raise. She does not name a target volume in the interview.

The financing comes at a time when she expects consolidation among European cloud and AI infrastructure providers. The decisive factor will be who can handle the growing technical demands and expand capacity economically.

Sovereignty Cannot Simply be Exported

The ambitions extend beyond Switzerland. According to Bünger, Phoeniqs has founded a subsidiary in Spain and is planning further growth opportunities, especially in Germany and the Middle East.

However, the Swiss offering cannot be exported unchanged. "Digital sovereignty is always local," Bünger says. Anyone wanting to offer German customers a similarly local solution must also build infrastructure in Germany and ensure that the data remains there.

Phoeniqs therefore primarily wants to export its technology and operating model. Computing power is to be created where the respective customers need it.

More Efficient Models, More Usage

Bünger does not yet see demand falling as a result of more efficient language models. Admittedly, newer models require fewer resources for certain tasks. At the same time, however, they are becoming so useful that customers are deploying more applications with them.

For Phoeniqs, usage is therefore rising despite technical efficiency gains. Bünger’s growth bet rests on the idea that banks and other companies are increasingly integrating AI into their day-to-day operations - and are therefore looking for infrastructure to which they can entrust their sensitive data.