Tech companies are investing more money in the creation of so-called “sovereign” AI models in an effort to increase their competitiveness by emphasising local infrastructure.
TakeAway Points:
- Many of the biggest language models available today, including as Anthropic’s Claude and OpenAI’s GPT, store data in U.S.-based data centres and handle requests through the cloud.
- Politicians and authorities in Europe are worried about this because they believe that the continent’s competitiveness will suffer from its reliance on American technology.
- This brings us to the concept of “sovereign” AI, which holds that AI services in a particular jurisdiction should be based on data from that area so that the outcomes are rooted in the local language and culture.
Tech companies invest in sovereign AI
Data sovereignty refers to the idea that people’s data should be stored on infrastructure within the country or continent they reside in.
“Sovereign AI is a relatively new term that’s emerged in the last year or so,” Chris Gow, IT networking giant Cisco’s Brussels-based EU public policy lead, said.
Currently, many of the biggest large language models (LLMs), like OpenAI’s ChatGPT and Anthropic’s Claude, use data centers based in the U.S. to store data and process requests via the cloud.
This has led to concern from politicians and regulators in Europe, who see dependence on U.S. technology as harmful to the continent’s competitiveness — and, more worryingly, technological resilience.
AI sovereignty
The notion of data and technological sovereignty is something that has previously been on Europe’s agenda. It came about in part as a result of businesses reacting to new regulations.
The European Union’s General Data Protection Regulation, for example, requires companies to handle user data in a secure, compliant way that respects their right to privacy. High-profile cases in the EU have also raised doubts over whether data on European citizens can be transferred across borders safely.
The European Court of Justice in 2020 invalidated an EU-U.S. data-sharing framework, on the grounds that the pact did not afford the same level of protection as guaranteed within the EU by the General Data Protection Regulation (GDPR). Last year, the EU-U.S. Data Privacy Framework was formed to ensure that data can flow safely between the EU and U.S.
These political developments have ultimately resulted in a push toward localisation of cloud infrastructure, where data is stored and processed for many online services.
Filippo Sanesi, global head of startup marketing and operations at OVHCloud, said the French cloud firm is seeing lots of demand for its European-located infrastructure, as they “understand the value of having their data in Europe, which are subject to European legislation.”
“As this concept of data sovereignty becomes more mature and people understand what it means, we see more and more companies understanding the importance of having your data locally and under a specific jurisdiction and governance,” Sanesisaid. “We host a lot of data,” he added. “This data is sovereign in specific countries, under specific regulations.”
“Now, with this data, you can actually make products and services for AI, and those services should then be sovereign; they should be controlled, deployed, and developed locally by local talent for the local population or businesses.”
The AI sovereignty push hasn’t been driven forward by regulators — at least, not yet, according to Cisco’s Gow. Rather, it’s come from private companies, which are opening more data centers — facilities containing vast amounts of computing equipment to enable cloud-based AI tools — in Europe, he said.
Sovereign AI is “more driven by the industry naming it that than it is from the policymakers’ side,” Gow said. “You don’t see the ‘AI sovereignty’ terminology used on the regulator side yet.”
Countries are pushing the idea of AI sovereignty because they recognize AI is “the future” and a “massively strategic technology,” Gow said.
Governments are focusing on boosting their domestic tech companies and ecosystems, as well as the all-important backend infrastructure that enables AI services.
“The AI workload uses 20 times the bandwidth of a traditional workload,” Gow said. It’s also about enabling the workforce, according to Gow, as firms need skilled workers to be successful.
Most important of all, however, is the data. “What you’re seeing is quite a few attempts from that side to think about training LLMs on localized data, in language,” Gow said.
