Sovereign AI: Piece by Piece | NetApp, Radiant, Hydra Host, Coatue & More | RAISE Summit 2026

Definition and Scope of Sovereign AI

• George Kurian (NetApp) defines sovereign AI as the ability of a nation-state or organization to independently control the development, deployment, and governance of AI activities, encompassing technology, data, operations, and talent.
• Mayor Maddy (Radiant) characterizes sovereign AI as a continuous spectrum rather than a binary end state, where the primary variable is the level of control and risk an entity accepts regarding AI components.
• Ben (Polarize) identifies AI as critical infrastructure akin to utilities (gas, electricity), arguing that token production and underlying infrastructure must not rely on foreign hyperscalers.
• Max (Kotu) introduces a secondary layer of sovereignty focused on data ownership and model provenance, questioning whether a company owns its data and weights even if the infrastructure is locally hosted.

Critical Layers of the AI Stack

• Maddy asserts that control over the "control plane" is the most critical element for sovereignty, regardless of physical infrastructure location.
• Ben and Maddy identify data residency and physical infrastructure (chips, data centers) as the most durable trends, driven by the shift of GPU technology which is fundamentally "anti-cloud."
• George Kurian notes that while hyperscalers offer flexibility, the most critical need for sovereignty is the ability to deploy data wherever the client chooses, whether in a hyperscaler or a sovereign environment.
• Max highlights that while the control plane is vital, geopolitical realities (such as chip export controls) have shifted the focus to the physical location of data centers and the national identity of the hardware.

Feasibility of Sovereignty via Hyperscalers

• Ben rejects the possibility of true sovereignty through existing hyperscalers, citing historical precedents where legal entities (e.g., Airbus, OVH, French government) could not guarantee data residency due to parent company structures or US jurisdiction.
• Maddy argues that while hyperscalers build local data centers, they often retain control of the control plane, meaning nations have not yet reached the "execution" phase of policy but remain in the "rhetoric" or "policy" phases.
• George Kurian observes that for commercial enterprises, sovereignty is often a risk management tool for core IP, whereas governments face a trade-off between fostering national security and maintaining competitive access to global innovation.
• Max suggests that "model routing" services (e.g., Amazon Bedrock, Microsoft Foundry) allow some degree of destination control but complicate the definition of sovereignty by abstracting where data actually resides.

Economic Implications and Costs

• Max forecasts that sovereign AI will significantly increase costs due to the necessity of duplicating data center infrastructure and acquiring more chips, potentially accelerating innovation cycles.
• Aaron Gipps (Hydrohost) argues that sovereign AI compute prices may not command a premium because high margins exist within the ecosystem that can be absorbed to remain competitive with hyperscalers.
• Aaron notes that sovereign AI demand is "price inelastic" for governments, as the value of intelligence and national security is unbounded and undefined, leading to continued investment regardless of cost.
• George Kurian posits that constraints and export controls can spur innovation, such as the development of diffusion-based models in China, creating lower-cost alternatives and alternate chip designs.

Global Progress and Regional Strategies

• Aaron identifies El Salvador as a key site for the first Blackwell deployment in Latin America, leveraging cheaper power (3 gigawatts in Paraguay) to build sovereign AI factories.
• Aaron highlights that while Asia's sovereign AI drive is heavily motivated by US export controls, Europe and South America are driven by energy scarcity and the desire to avoid hyperscaler dependency.
• Maddy points to significant progress in Europe, citing a €30 billion commitment in France and Sweden, and a 10-megawatt operational AI industrial cloud in Germany that increases the nation's AI compute capacity by 50%.
• Ben acknowledges that while Europe has rising stars and initiatives, the continent will likely remain heavily dependent on foreign chips, infrastructure, and foundation models for the next five years.

Future Outlook (5-Year Horizon)

• George Kurian predicts a bifurcated global ecosystem where the US maintains full-stack dominance (chips to apps) and China builds a coalition around open-source models and its own chip ecosystem.
• Maddy suggests that for all nations other than the US and China, sovereignty will involve building controls within these two dominant ecosystems, likening the future landscape to the airline industry with a core duopoly (Boeing/Airbus) and local carriers.
• Ben forecasts that Europe will continue to rely on hyperscalers and foreign infrastructure but must aggressively develop local capabilities to "catch up."
• Max anticipates a proliferation of data centers and localized compute nodes, driven by companies like Polarize, to meet the growing demand for distributed, sovereign infrastructure.