# Compute Sovereignty **Compute sovereignty** is the capacity of a nation, bloc, or organization to independently train, deploy, and control AI systems without depending on foreign hardware, cloud infrastructure, or model providers. It is the AI-era equivalent of energy independence or nuclear capability — a measure of technological self-determination that increasingly maps onto geopolitical power. The concept gained urgency after the U.S. imposed GPU export controls on China (October 2022, tightened 2023 and 2024), demonstrating that access to AI compute could be weaponized as a geopolitical lever. A nation that cannot source its own AI-grade chips, build its own data centers, and train its own models is dependent on whoever can — and that dependency can be exploited. ## The Stack of Sovereignty Compute sovereignty requires control at multiple layers, each with its own chokepoints: **Chip design.** NVIDIA designs the GPUs that train virtually all frontier AI models. AMD and Intel are secondary players; Google's TPUs serve Google internally. Chip design expertise is concentrated in the U.S. (NVIDIA, AMD, Intel, Google) with an emerging Chinese competitor ecosystem (Huawei Ascend, Biren, Enflame) operating under U.S. sanctions pressure. **Chip fabrication.** TSMC (Taiwan) manufactures the most advanced AI chips for NVIDIA, AMD, Apple, and others. Samsung (South Korea) is the only alternative at leading-edge nodes. The geographic concentration of advanced fabrication in Taiwan — an island China claims — is the single most strategically precarious chokepoint in the global technology supply chain. **Data centers.** Training frontier models requires GPU clusters consuming hundreds of megawatts of power. The [[The Quest for 90 Gigawatts]] documents the scramble to build this infrastructure, concentrated in the U.S., with emerging buildouts in the EU, Middle East (UAE, [[Saudi Arabia]]), and Southeast Asia. Each data center requires land, power, water for cooling, and network connectivity — physical resources that tie digital sovereignty to geography. **Energy.** AI compute is energy-intensive at a scale that is reshaping national energy policy. A single frontier training run can consume as much electricity as a small city for months. Nations without sufficient energy surplus — or the ability to build new generation capacity quickly — cannot achieve compute sovereignty regardless of their chip access. **Models and data.** Even with hardware, training competitive models requires massive datasets and specialized expertise. Nations that rely entirely on U.S. or Chinese models for their AI applications are model-dependent — their AI infrastructure can be disrupted by a terms-of-service change or an export policy revision. ## Who Has It **Full sovereignty:** The United States (NVIDIA + hyperscaler cloud + energy capacity + frontier labs). No other nation has the complete stack. **Partial sovereignty:** China (domestic chip design under sanctions pressure + massive data center buildout + competitive labs like DeepSeek, Alibaba, Baidu — but dependent on older-node fabrication and smuggled/stockpiled NVIDIA chips). The EU (ASML controls lithography, but no leading chip designer, limited hyperscaler presence, and fragmented policy). The UK (DeepMind/Google, but no sovereign chip supply). **Aspiring sovereignty:** UAE and Saudi Arabia (massive capital investment in data centers and AI funds, but no domestic chip design or fabrication, dependent on U.S. export policy for GPU access). India (growing data center presence, strong talent, but infrastructure and energy constraints). Japan, South Korea, and Taiwan have pieces of the stack but lack the integrated sovereignty of the U.S. or China. **No sovereignty:** Most of the world. Nations in Africa, Latin America, and most of Asia are consumers of AI services provided by U.S. and Chinese companies, with no capacity to build, train, or control the underlying systems. This creates a new axis of technological dependency that mirrors and reinforces existing economic hierarchies. ## KB Significance 1. **The new map of power.** Compute sovereignty is becoming as geopolitically significant as nuclear capability or energy independence. The KB's documentation of U.S.-China competition, GPU export controls, and the energy buildout are all manifestations of the compute sovereignty contest. Nations that lack compute sovereignty are dependent on those that have it — and that dependency is a lever of control. 2. **The NVIDIA chokepoint.** NVIDIA's dominance of AI GPU design, combined with TSMC's fabrication monopoly and ASML's lithography lock, creates a chokepoint that the U.S. government has already weaponized through [[GPU Export Controls]]. The KB's tech-oligarchy analysis includes NVIDIA's Jensen Huang as a figure whose company's products are now instruments of foreign policy. 3. **The Middle East play.** Saudi and Emirati AI investment — billions flowing into data centers, model development, and AI funds — is a sovereignty play by petrochemical states seeking to transition their geopolitical leverage from energy to compute. The KB's documentation of Trump-era Gulf relationships and tech-oligarch courtship of Middle Eastern capital intersects directly with compute sovereignty dynamics. ## See Also - [[Frontier Model]] — What compute sovereignty enables you to build - [[Scaling Laws]] — Why sovereignty requires such enormous resources - [[GPU Export Controls]] — The weaponization of compute access - [[Open-Weight AI]] — A partial workaround for nations lacking training compute - [[The Quest for 90 Gigawatts]] — The energy dimension of compute sovereignty