Sovereign cloud spending puts a price on business control

Sovereignty needs an owner

The European Union has moved strict provisions of its AI Act to December 2027, but buyers may move faster than regulators. Sovereignty is already becoming a condition for permission to operate, production use of AI agents and access to new markets.

Responsibility remains split across existing roles: the CISO owns the security perimeter, the chief data officer owns data, the CIO owns infrastructure, and the general counsel handles regulation. Sovereignty spans all four, but belongs to none. While 72% of executives say it matters more, only 30% have governance for it built in by default.

The forecast is that a chief sovereignty officer could emerge in 2027, with the topic taking a place on board agendas once held by ESG: a metric, a report and an accountable executive. That is a plausible organizational response, but a new title alone would not resolve the underlying question of what a company must actually control.

Consolidate the platform and the data

A worldwide ThinkerMedia survey conducted in September 2026 found a marked shift in platform preferences. Asked what they would choose next, 63% of executives picked Postgres, 19 percentage points ahead of its nearest competitor; Oracle drew 19%.

The next argument, the forecast says, is less about choosing a database than deciding what belongs inside it. Vector search, governance, agent context and sovereignty could be built into the database or left to external components. The article’s analogy is a Ferrari assembled with parts bolted on from the outside.

The same consolidation logic applies to data. Keeping transactional, analytical and AI workloads in three places means maintaining three homes and three refrigerators. Seven in ten companies say their data, AI and agents are fragmented or being consolidated; almost all expect that to change over the next three years. Inference has become 280 times cheaper, but moving data to it has not.

The proposed answer is to bring intelligence to the data, clean it where it lives and manage systems through a single control plane. The open question is whether consolidation can deliver that without creating a new point of dependence.

Agents need an operating system, not just tools

The forecast compares managing AI agents to coaching in the NFL: 25 to 40 seconds to make each decision, most plays failing, and the team learning continuously. A collection of disconnected tools is unlikely to be enough. Only 44% now consider an AI operating system critical, but 86% expect it to become so within three years — the fastest-rising idea in the survey.

Executives draw a distinction between the data platform and the operating system. Data capabilities belong to the platform; the ability to run anywhere and remain open belongs to the operating system.

Nor is switching agents off after an incident a sufficient control strategy. A billion agents running around the clock, the article argues, need a “switch-on switch”: contextual rules embedded in the data layer and applied at the right moment. A third of companies say leaving their platform would require architectural rework, while governance remains at the planning stage.

Choose governance and portability together, the argument goes, or get neither. When 80% of databases are created by agents, engineers will become editors.

The bill comes due in capital and skills

The industry’s expansion is being financed with debt: capital spending stands at a trillion dollars, a third of it financed by credit, while yields on ten-year bonds approach 5%. Memory, not models, is becoming the scarce resource. One rack consumes as much power as 65 homes, and power constraints affect two out of five AI data centers.

AI spending is approaching $4 trillion, yet 40% of AI-agent projects fail. The diagnosis here is not a shortage of capital but a failure to move systems into production: companies adopt AI where they can measure returns. SaaS spending is falling because agents need data, not applications; sovereignty is emerging as a new budget line.

The labor picture is just as unsettled. Entry-level roles are disappearing, senior AI specialists are leaving before strategy arrives, and the most valuable human skill may be building the skills agents use. Prompts are the typewriter; skills are the printing press. Models can be rented, but a company’s skills library belongs to it and sits alongside its data.

I think the central tension is that sovereignty is being treated as both a strategic asset and a bill companies may soon have to pay. But the forecast offers no common measure of what “real sovereignty” means — or how much control is enough. Without that definition, a new executive title could make the responsibility visible without making the underlying systems any more portable or governable.

The forecast is by Max Romanenko, CTO of EnterpriseDB (EDB). Its sources include MIT’s “Sovereignty Matters” (May 2025), ThinkerMedia’s corporate data and AI survey, n = 600 (September 2026), Gartner, Goldman Sachs and the IEA.