Sovereignty has moved from a compliance footnote to a board-level requirement, and the bar has risen well past tracking where data physically sits. At its Think 2026 conference, IBM made Sovereign Core generally available, a software platform that delivers operational, data, technology, and AI sovereignty through customer-operated control planes running on Red Hat OpenShift. The release treats sovereignty as a portable software capability, no longer a property of a particular cloud region, and that distinction is where the competitive contest now rests.

Sovereignty Outgrows Data Residency

Two years ago, sovereignty conversations with data and AI vendors stayed inside the compliance team and rarely went beyond a single question about which country the data physically sat in. Sovereignty has since moved into the boardroom. In IBM’s 2026 CEO Study, conducted by the IBM Institute for Business Value, 83% of CEOs, and 97% of the most AI-mature among them, said that developing and maintaining AI sovereignty is essential to their business strategy. The pressure is sharpest in regulated industries, where banks, insurers, government agencies, and healthcare providers now have to prove where data lives, who controls the AI models acting on it, how those models reason, and whether every action can be reconstructed after the fact. With countries drawing harder lines around their data and new AI rules emerging region by region, sovereignty has gone from a box that companies could check late in the buying process to something they have to solve before AI can go into production.

Most sovereignty offerings to date stop at the residency piece. A regional cloud zone keeps data inside a border, but it says little about the AI systems running on top of that data. When a model makes a pricing change, routes a shipment, or flags a transaction, the location of the servers tells a regulator almost nothing about how that call was made or whether it can be traced back later. The controls that govern those decisions live at the AI layer, well above the hardware. There’s a second problem, too. When those controls only work inside one vendor’s cloud regions, the customer is covered in one place — but has a governance gap everywhere else its data and AI run. That doesn’t fit the mix of clouds and datacenters most regulated companies use.

On top of that, proving compliance is still mostly a manual job. A team pulls the evidence together for an audit, but the environment keeps changing underneath them as workloads launch and settings shift, so that snapshot goes stale almost as soon as it’s filed. Sovereignty is something a company has to prove all the time, and for most, keeping up with that by hand takes far more effort and specialized staff than they can realistically spare.

A credible sovereign platform now has to do more than fence data into a region. It has to put the controls in the customer’s hands so that the company on the hook for following the rules runs the control plane itself instead of leaning on the vendor. The platform has to extend governance to the AI layer, covering which models run, where they run, and how their decisions get logged, because residency guarantees say nothing about model behavior. Governing that layer is an immediate requirement, because AI is already making operational decisions on its own. In the same IBM study, CEOs reported that a quarter of the routine, rules-based decisions a company makes, like repricing a product or rerouting a shipment, are already handled by AI with no human in the loop today, a share they expect to reach nearly half by 2030. As that autonomy scales, the AI doing the deciding becomes part of what a sovereign boundary has to contain. The platform also has to make compliance continuous instead of periodic, and it has to run across the clouds and on-premises systems an enterprise already operates rather than pulling everything into one provider’s region.

Inside IBM Sovereign Core

At Think 2026, IBM moved Sovereign Core to general availability and positioned it as a software platform for standing up AI-ready sovereign environments that give customers verifiable control over their data, their operations, and the AI systems running on top. The product’s capabilities are organized around four pillars the company calls operational, data, technology, and AI sovereignty; its central design choice is to build those controls into the platform itself so they hold up on their own rather than resting on a vendor’s contractual assurance. In practice that means a customer-operated control plane running on Red Hat OpenShift. Identity, encryption keys, logs, and audit evidence all stay inside the sovereign boundary, and compliance is checked continuously at runtime instead of assembled by hand at audit time. IBM ships more than 160 preloaded regulatory frameworks so teams can produce audit-ready evidence on demand, and it governs AI execution so models, inference, and agents run inside the boundary under local oversight. Dinesh Nirmal, who leads IBM Software, framed the intent directly, saying that sovereignty should not be a constraint on innovation and that with the right software foundation it becomes an enabler of it.

The platform isn’t locked into IBM’s own software either. An extensible catalog lets customers fill the sovereign boundary with pre-vetted IBM, open-source, and third-party tools from a partner ecosystem that spans silicon, data, and security, including AMD, Intel, Cloudera, MongoDB, Elastic, and Palo Alto Networks. For models, IBM turned to Mistral, a European model maker, and did not tie the platform to its own Granite family; this choice shows that Sovereign Core is built to govern whatever models a sovereign environment calls for while leaving the customer’s model decision open. IBM also introduced Sovereign Core alongside the Blueprint for the AI Operating Model, CEO Arvind Krishna’s keynote framing of enterprise AI as four connected systems covering agents, data, automation, and a hybrid foundation. Sovereign Core is cast as the hybrid governance layer, which positions sovereignty as one part of a larger operating model rather than a standalone compliance tool.

The Gap Between IBM’s Proof and Regulators’ Acceptance

The smartest thing about Sovereign Core is that it takes the hardest question in regulated enterprise buying and turns it into something a customer can purchase and operate. Sovereignty has long been the criterion that stalls AI deals in banking, government, healthcare, and defense, and most answers on the table treat it as a hosting arrangement backed by a contract. IBM’s bet is to make sovereignty something the software itself proves, with evidence generated as the system runs, instead of something a customer simply has to trust. Delivering that is much harder than a written promise, and far more convincing if IBM pulls it off. By shipping that control as software the customer runs on its own infrastructure, including bare-metal environments that answer to no hyperscaler, IBM makes it portable across the systems a buyer already operates — a far better fit for hybrid reality than a single regional zone. The four-pillar framing is the clearest case IBM has made for governance as one unified layer, and extending it to AI execution lands exactly where enterprise risk is heading as models take on more decisions. The Mistral inclusion reinforces the logic, with IBM positioning Sovereign Core as the control layer that governs any model, a stronger long-term stance than betting the platform on its own models winning.

The sharper question is whether sovereignty that’s provable in software is provable to a regulator. IBM can generate continuous evidence and automate the audit trail, but sovereignty gets judged by auditors and supervisory authorities, and whether they accept a control plane’s machine-made proof as adequate is a legal call that varies by country and sits outside IBM’s control. A deployment could be technically airtight and still fall short if a regulator asks for proof the software was never built to produce.

Pricing is a separate unknown, since how IBM prices Sovereign Core will shape whether sovereignty reads as a default or a luxury. That choice lands on a buyer that tends to move slowly, because regulated industries are the natural market here and also the least adapted to the speed the AI era now demands. The design also hands real operating responsibility to whoever runs the platform, which by intent is the customer’s own IT function or a partner, whether a regional sovereign-cloud operator or a system integrator, local or global. It is an open question whether that partner bench is deep and capable enough region by region, because where it’s thin the burden falls back on a customer that often can’t carry it. All of this is the normal state of a general-availability launch, where the framing runs ahead of the evidence. In this case, regulators get the deciding vote, and Sovereign Core only works if they accept that software can prove sovereignty in the first place.

Sovereignty in the Age of AI Becomes a Dividing Line

By 2026, the market has split into two kinds of sovereignty stories. On one side are the offerings that still treat it as a location, a region, or a datacenter named in a contract. On the other are the ones that treat it as control that the customer can hold and prove for itself. The first group keeps clearing the bar buyers used to set, while the regulated companies actually running AI in production have moved the bar to the second.

Sovereign Core is IBM’s bet on the second camp, and a clear one. The customer holds the controls, the software proves them as it runs, and the whole platform sits on infrastructure the customer already owns, which is what regulated buyers have been asking for and rarely been offered. By packaging control-based sovereignty as a product instead of a promise, IBM has set a marker the rest of the field now has to measure against. The next year will show whether sovereignty that a customer can operate and prove for itself becomes the baseline buyers expect, or remains the exception.