- Sovereign AI awareness remains low despite rising concerns over data sovereignty, model control and infrastructure risk
- Telcos see sovereign AI as a privacy and competitive edge play
- AI control is becoming mission-critical, with sovereign AI emerging as a way to keep key workloads closer to owned data and infrastructure
Sure, you’ve heard about sovereign AI. But could you explain its concepts and risks to someone else? If not you’re not alone. A recent study conducted by IDC and AI model builder Cohere found just 13% of enterprises really understand sovereign AI.
The survey, which included over 500 IT and business decision makers across the globe, found familiarity with sovereign AI was lowest in Canada (10%) and U.S. (12%). Germany scored better, with 29% reporting “high awareness” of sovereign AI concepts.
Awareness appears to be correlated to geography. For instance, Europe has structured policies around data sovereignty and is pursuing more comprehensive regulations that would cover AI sovereignty as well. Hence, German enterprises have a better understanding of such issues than those in the U.S., where no such federal policy exists. Canada has just come out with its AI for All strategy, but its policy framework is significantly less mature than that of Europe.
But Germany’s 29% is still a low score, especially for a country widely viewed as being at the forefront of digital policy development. So that raises the question: why is sovereign AI so hard for enterprises to define and understand?
“Because sovereignty is fundamentally an architectural question,” a Cohere spokesperson told Fierce. While organizations often consider data residency and geography, “genuine sovereignty extends to control over models, infrastructure, access and operational continuity,” the rep added.
Indeed, architectural understanding is tricky, but it’s also key to the AI equation.
A recent IBM study found only 9% of enterprise executives said they have an excellent understanding of their AI dependencies. That’s a huge problem given the same executives said these dependencies have resulted in “price increases, usage restrictions, model deprecations, changes to privacy and data-handling terms, performance degradation and new geographic access limitations” over the past 24 months.
What’s driving sovereign AI demand among telcos?
While the IDC/Cohere survey didn’t break out results for telco understanding of sovereign concepts, it did look at the “why” that’s driving interest in sovereign AI solutions.
The report found 75% of telcos are interested in sovereign AI because they’re worried about the risk of data leakage and privacy breaches when using generative and agentic AI tools. Additionally, 37% of telcos believe sovereign AI will give them a competitive advantage.
The survey didn’t capture operator’s reasoning on the latter point, but the Cohere spokesperson said “we know that telecom operators sit on sensitive customer/network data and critical infrastructure, so greater control over models, data and deployment can differentiate them. Sovereign deployment also gives telcos greater ability to customize AI to their own data, domains and workflows, creating differentiated services rather than relying on the same general-purpose models as everyone else.”
All about AI control
In a nutshell, sovereign AI is all about controlling rather than renting AI and the benefits it brings. After all, as recent events have shown, access to models can be revoked at the drop of a hat. That could obviously have huge ramifications for enterprises running critical workloads on proprietary models on infrastructure they don’t control.
For telcos, gaining control means bringing AI to where the data lives and deploying it in their own environments, the Cohere rep said. “Operationally, that means private cloud, on-premises or even air-gapped deployment where required, rather than building mission-critical systems on an external API whose terms or availability can change,” the spokesperson added.
Notably, though, the survey found infrastructure (17%) was cited as the biggest barrier to sovereign AI readiness. It was followed by cost (10%), alignment (9%) and skills and talent (9%).
Read more about sovereign AI here:
U.S. Anthropic order throws a megaton of fuel on sovereign AI fire
Orange creates internal AI cloud for employees, sells to businesses, too
Could ‘data embassies’ solve the AI sovereignty problem? MTN thinks so.
Korean operators ramp up AI investments in AI superpower push
Dell builds a sovereign AI stack for the on-prem era