- A new IBM study reports only 9% of executives say they fully understand their AI dependencies, while 71% say switching vendors would be difficult
- Multi-vendor strategies are often accidental, not strategic
- Telcos are emerging as key players in AI sovereignty due to local infrastructure and regulatory positioning
AI sovereignty is quickly shifting from an abstract policy concept into a practical network and infrastructure problem — and one that enterprises aren’t prepared to manage.
According to a new IBM study, The calculus of AI sovereignty, “only 9% of executives say they have an excellent understanding of their dependencies on AI vendors, models and infrastructure,” even as AI becomes embedded in operational decision-making.
At the same time, “71% say switching their primary AI vendor or model would be difficult,” underscoring how deeply enterprises are locked into specific ecosystems. (See image below.) That lock-in is already showing up in operational risk, noted the reports authors.
The report also highlights the volatility enterprises are dealing with: significant price increases, changes to terms of service or usage restrictions and model or service deprecation/discontinuation rank among the most common disruptions over the past 24 months. These three map directly to cost spikes, performance issues and compliance exposure.
The report’s authors note: “AI introduces new forms of dependency…that are more volatile, more opaque, and harder to unwind.”
That dependency extends beyond infrastructure into the model layer, where behavior can change “without formal release cycles.”
How telcos can help
For telecom operators, this is where the report strikes a positive tone. The report calls out the sector’s role as AI sovereignty moves into execution.
“Telcos are emerging as critical enablers…inside national borders, under strict regulatory regimes, and at the intersection of data, compute, and critical infrastructure.” These points all give telcos a clearer role in enterprise AI architectures — particularly as companies look to localize workloads and reduce cross-border dependency risk.
But enterprises aren’t making those decisions cleanly or easily. In fact, 28% of organizations use four or more AI vendors, yet the report finds this fragmentation is rarely strategic. Instead, “most AI ecosystems are not the result of deliberate design.”
The financial implications are growing. Running AI far from data can drive costs up sharply. Enterprises can pay “2.8× more in token processing” due to poor data placement. And those without strong control models risk far more: the report finds companies with higher AI sovereignty “protect 55% more operating profit from AI-driven disruption.”
The bottom line: AI sovereignty is no longer about ownership. It’s about maintaining control in a system where vendors, models and rules are constantly shifting — and where losing control increasingly shows up on the balance sheet.