Punnam Raju Manthena, Cofounder & CEO at Tekskills Inc. Partnering with clients across the globe in their digital transformation journeys.
If you’re a business leader, your AI sovereignty is your capability to independently build, deploy and regulate your AI using your own infrastructure, data, models and workforce. A nation’s job is to strengthen its digital infrastructure, such as national payment systems, to avoid depending on foreign financial engines. Considering the geopolitics of compute, this may be akin to oil reserves. Soon, chips, GPUs and cloud infrastructure could be the powerful geopolitical assets that decide whether a country is economically dominant and secure, rather than its oil or gold reserves.
In either case, businesses and countries get to work with chosen vendors, based on cost and capability considerations. This approach may be incomplete and “burn a hole in your pocket” in the long run. It may be better that you look at the technical fit, data governance and total spend as well.
What is AI sovereignty?
Generally, you might think that AI sovereignty is all about frontier AI models: the models at the forefront of your capability. That’s a big no. There are aspects, including jurisdictional access, data residency and vendor availability, that can hit deployments.
First, jurisdictional access means you juggle your deployments with legal compliance, complicating them. When you have your software and data across multiple countries, you need to comply with diverse and sometimes conflicting laws about who can access, audit or seize your data.
Data residency, the locations where you store and process your data, impacts the network and the application. Some locations may have regulatory requirements that mandate that some personal data stay within those regions. Vendor unavailability, the biggest factor that can hit you, can stop critical operations because replacing the vendor can be a time-consuming job and cause downtime, delays, costs and business disruption.
You can see that AI sovereignty is managed on three distinct fronts: compute, model access and data/deployment residency. Compute is your ability to possess, control and run your computing infrastructure to train and run AI without banking on foreign entities. You control which chip can run where. You own high-performance and high-economy infrastructure instead of renting it overseas and risk getting your data or access cut off by foreign governments via a remote “kill switch.”
AI sovereignty model access refers to your ability to control, regulate and secure your AI models without depending on overseas or single-vendor cloud providers. You never know when a vendor can become legally unusable in a region overnight, no matter your contract terms. Try to achieve selective sovereignty by having good control over critical models. Also, try to define who (including AI agents) can invoke a model through identity and policy.
Data residency tells you where your drives and servers are located, while sovereignty tells you the laws that govern that data and whether there is any foreign government that may be able to legally access it.
AI sovereignty is beneficial and I believe worth every cent and effort. It gives you control over your AI infrastructure, data and models to ensure that security and alignment comply with the local laws. Your sensitive information (demographic, health and financial) stays securely within your local boundaries. Your data adheres to regional legal frameworks. You don’t have to rely on foreign or global tech giants, and thus you can safeguard your operations from geopolitical shifts or supply chain disruptions.
AI sovereignty can be an uncomfortable trade-off.
AI sovereignty is desirable considering the benefits it can give you, but the pursuit of AI sovereignty is often an uncomfortable trade-off. It’s a choice between autonomy versus cost and efficiency in that the autonomy ensures security and compliance but needs huge capital upfront. You may have to rely on overseas suppliers, and this reliance may quicken your deployment and bring down costs, but it increases exposure to geopolitical shifts, changing regulations and extraneous pricing models.
Sometimes, you may wish to “put your eggs in different baskets” and so diversify your vendors and regions, but this only adds to the complexity. This hedging isn’t free, and a lot of companies are underinvesting in this because the cost is visible now, but the risk is invisible right until it actually hits you.
At the same time, vendor concentration risk isn’t theoretical anymore. As you depend more and more on vendors for software, hardware, cloud and AI systems, your operational resilience increasingly depends on a small number of such suppliers. If one of them fails, backs out, becomes bankrupt or suffers a breakdown, you suffer badly.
According to ThoughtMinds, a case in point is the risk suffered by Toyota, owing to an earthquake, leading to production delays. The paper also recounts the risk suffered by Apple, which depended heavily on Foxconn for the production of iPhones. These illustrate that vendor concentration risk isn’t theoretical.
As a business leader, you have a difficult choice. AI gives you a tremendous chance to automate your workflows and give your business a competitive edge. But when you depend on an external AI provider, you put your proprietary knowledge into a black box with little or no control. If this is the case, what should a leader do? Map the critical workflows that depend on a single AI vendor or region and look at them as known risks. In parallel, build contractual and architectural fallback alternatives in case you run into a disruption. Look at AI infrastructure risk like supply-chain risk and keep reviewing regularly, and don’t decide just once.
Your competitive advantage will not come from just having access to AI, because everyone around you is also going to have it. Rather, it’s your ability to safely access your AI to make your business processes smoother and better. You may be caught flat-footed if you look at AI infrastructure as a single strategic bet. Instead, recognize that resilience is a competitive advantage, not just insurance, because real sovereignty means knowing what your AI does, why it does it and who is accountable when it does not.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?