Sovereign AI: Why Nations Are Building Their Own Models

Summary

• Sovereign AI is breaking the cloud-era assumption that global workloads will concentrate in the US and China. The kingdom announced HUMAIN, a local hyperscaler or AI platform intended to run most AI workloads domestically, within an announced cluster buildout somewhere around $100 billion-$250 billion; roughly 500 megawatts appears to be the “atomic unit.” The goal is “infrastructure independence,” including autonomy over models and deployment.
• “AI factories” are technically distinct from conventional data centers. GPUs are the major active-component difference, while high-density clusters require rack-level liquid cooling, an energy supply close to a power plant, and early energy commitments. Enterprises may also bypass elaborate cloud stacks for Kubernetes plus selected Snowflake- or database-type services.
• Models have become cultural and information infrastructure, making foreign dependence a national vulnerability. Training data embeds values, while post-training steers what models answer or refuse; meanwhile, foundation models already touch defense, healthcare, finance, and the daily decisions of ChatGPT’s roughly 500 million monthly active users. As models replace search and grade schoolwork, whoever controls them could shape accepted history and truth.
• AI data centers resemble industrial-era oil reserves—with the crucial difference that countries can construct them. Capital and political will can create the compute base upon which domestic industries, development, and exports are built.
• The US faces a choice between helping allies build sovereign capacity and leaving the field to Chinese models. Midha’s preferred analogy is a “Marshall Plan for AI”: the original reconstruction looked like a capital export but produced a 70-year US-Europe trade corridor and kept China out of that equation. At the model layer, he reduces the diplomatic choice to “DeepSeek or Llama?”
• Appenzeller rejects comprehensive government control while preserving a targeted government role. Government can fund fundamental research and set sound regulation, but competitive companies must supply the detailed innovation; even the Manhattan Project leaked, making total control “a pipe dream.” DeepSeek’s MIT-licensed release—26 days after OpenAI’s frontier release—supports his strategy of building and exporting the best technology, ideally from the US and its allies, while Midha calls the resulting approach “foundation model diplomacy.”

Deep dive

1. Sovereign clusters overturn the cloud’s geography

• Guido Appenzeller reports that the kingdom announced HUMAIN, a local hyperscaler or AI platform intended to run the “vast majority of AI workloads” domestically rather than reproduce a cloud era dominated by US and Chinese infrastructure. The announced cluster buildout is somewhere around $100 billion-$250 billion, with roughly 500 megawatts emerging as its “atomic unit.”

• Midha connects the move to earlier industrial cycles: whoever controls where technology is built and the underlying assets can shape regulation, usage, and the next wave of innovation. Data centers now occupy a strategic position analogous to oil in industrialization.

2. These are factories, not ordinary data centers

• Appenzeller contrasts a branding explanation with the view that this is more than marketing: under the hood, GPUs are the major active-component difference, and high-density AI clusters require rack-level liquid cooling, an energy supply close to a power plant, and early commitments to that supply.

• Enterprise demand is changing too: customers increasingly accept a simple Kubernetes abstraction, then “cherry pick” Snowflake- or database-type services instead of buying an elaborate full-stack cloud platform.

• Appenzeller’s sharper distinction is that models are “cultural infrastructure.” Training embeds values and norms; post-training governs what gets said or refused, creating demand for jurisdictional control over what each factory produces.

3. Model dependency becomes an information-security risk

• Capabilities have moved beyond the “early toy stage”: foundation models operate across defense, healthcare, and financial services, while ChatGPT has roughly 500 million monthly active users making real daily decisions. Depending on another country’s technology therefore looks like a “critical point of failure.”

• Appenzeller broadens sovereignty into control of the information space. As models replace search and grade essays, omitted history could become the reality citizens inherit, while something truthful might be marked wrong because the model’s controller excluded it from the training corpus.

4. Allied compute could become a Marshall Plan for AI

• Appenzeller calls US leadership an opportunity but says complete centralization will not happen; maintaining leadership while equipping strong allies is the more plausible balance.

• Midha’s historical analogy: GE, General Motors, and other leading enterprises helped subsidize Europe’s reconstruction despite criticism that the Marshall Plan exported capital. The result, in his telling, was a US-Europe trade corridor lasting 70 years and keeping China out of that equation.

• The modern choice is whether to help allies build capacity or leave them reliant on exported Chinese models. At the model layer, Midha asks, “What do we want our allies on, DeepSeek or Llama?” Countries able to finance sovereign infrastructure are already moving.

5. Competitive ecosystems beat a national AI master plan

• Appenzeller rejects Leopold Aschenbrenner’s nationalization thesis, citing East and West Germany as an “A/B test” between central planning and a free-market economy. Government can fund fundamental research and establish good regulation—bad regulation can “torpedo AI”—but “there’s no master plan” capable of supplying the details that markets discover.

• Appenzeller calls a fully centralized government approach “a pipe dream”: even the cordoned-off Manhattan Project leaked. Model weights matter less than the infrastructure running them, and “inference is almost more important.” He also says that US proposals a year earlier had sought to regulate model research and development versus misuse, but that the debate has moved on.

• DeepSeek shattered confident testimony that China was five to six years behind by appearing 26 days after OpenAI put out its frontier model. Its MIT license gave other countries immediate access, leading Appenzeller to conclude that the only way to win is to build the best technology and out-export anyone else. He says the US is better off embracing other countries’ ability to serve their own models, ideally with the best models coming from the US and its allies; Midha names this new approach “foundation model diplomacy.”