主权人工智能：构建战略韧性和影响力的生态系统（英）

Technology, Media & Telecommunications PracticeSovereign AI: Building ecosystems for strategic Sovereign AI is achievable only through an ecosystem effort thatconnects energy, compute, data, models, platforms, and applications Disclaimer: This article is descriptive and analytical. It does not provide policy, regulatory, or national security advice, andit does not recommend specific national strategies. Examples are included to illustrate ecosystem patterns Sovereign AI is moving from a policy debate to an economic and strategic imperative.Acrossgovernments, enterprises, and investors, leaders increasingly view the ownership of AI capabilities ascentral to economic competitiveness, strategic resilience, and societal trust. Yet despite this urgency, many sovereign AI initiatives are stalling and failing to deliver their expectedresults. In this article, we analyze how sovereign AI efforts are being pursued and what differentiatessovereign ecosystems that successfully translate intent into scaled adoption and durable advantage. Drawing on a global survey of enterprises, providers, governments, and investors,1we then examine theroles different actors must play, the challenges actors face, the partnership models that consistently Sovereign AI refers to a nation’s or organization’s ability to develop and control its own AI capabilities toensure strategic independence and alignment with domestic values and laws. That said, sovereign AI does —territorial:where data and compute physically reside —operational:who manages and secures data and compute —technological:who owns the underlying stack and intellectual property —legal:which jurisdiction governs access and compliance Viewed this way, sovereign AI is best thought of as a spectrum of potential solutions distributed acrossdifferent tiers of sovereignty, depending on stakeholder and local circumstances (Exhibit 1). As a result, sovereign AI represents one of the largest opportunities within AI. McKinsey estimates that30 to 40 percent of AI spending could be influenced by sovereignty requirements. This would represent a But seizing that opportunity means confronting a very specific execution challenge: Success in sovereignAI is not achieved through a single policy decision, cloud contract, or “national model” announcement.Instead, sovereignty is best thought of as an ecosystem effort that connects multiple layers—energy,compute, data, models, cloud platforms, and applications—into one coherent system, managing McKinsey & Company McKinsey & Company The sovereign AI ecosystem: Moving from ‘sovereign assets’ to For business leaders and policymakers, a useful starting point is to change the way they think aboutsovereign AI. Many initiatives focus on inputs—such as GPUs, data centers, cloud regions, and nationalmodel announcements—and while those inputs matter, the prize is in the long-term outcomes, such as An effective sovereign ecosystem is not necessarily one in which everything is built domestically. Instead,it is one in which key control points are sovereign by design, even if other elements of the stack mayremain open to partnerships, interoperability, and competition. Our analysis shows that the most effectiveecosystems operationalize “minimum sufficient sovereignty” with a repeatable decision rule: Classify Different jurisdictions are pursuing sovereign AI through distinct ecosystem archetypes. Even jurisdictionswith advanced capabilities are rarely self-sufficient across all layers and often rely on external providersin at least part of the stack (particularly in hardware and advanced compute). The following are among the —End-to-end hub and frontier AI powerhouses.In this model, private data center operators build massiveAI-ready capacity to attract hyperscalers and frontier AI labs, enabling large-scale training and —State-led with data center or cloud execution.This approach is led by a state to keep compute, data, andmodel intellectual property under national control for strategic and public sector workloads. Local cloud —Model development led by research and policy.Led by research institutions and policymakers,regulatory and state incentives steer domestic model development and compliant data access,with cloud and data center ecosystems providing trusted environments for research compute to —Industry-led and compute-hardware-driven adoption.In this approach, data center providers andcloud players partner with local enterprises and chip ecosystem and semiconductor leaders to create —Policy-enabled regional hubs with strong local or regional demand.In this model, policy enables fastpermitting, power access, and investment incentives, and data center operators build and aggregate Even amid these differences, effective sovereign AI ecosystems tend to share a common set of —A demand-led anchor and sectoral adoption.Leaders start by clustering demand sources (such ascitizen services, health outcomes, financial integrity, critical in