AI Boss Tools | Diagnostic 01 • Back to the Hub
France committed EUR 109 billion. The EU put EUR 20 billion into AI gigafactories. A single gigawatt-scale AI data centre can demand around US$38 billion in capital. Should your government build, pool, or partner? Answer 12 questions and get a research-backed recommendation with a downloadable one-pager.
Sovereign AI means a nation controlling its own AI capability: the data, the models, and often the compute. More than 30 countries now fund national programmes, from South Korea's 260,000-GPU sovereign cloud plans to Canada's CAD $2 billion compute strategy. The costs are just as real: frontier training runs alone have cost between US$78 million and US$192 million, before staff, energy, or data centres.
This diagnostic scores your government across four dimensions: how badly you need sovereign capability, whether you can pay for and power it, whether you have the people to run it, and what alternatives you hold. Your total maps to one of four strategy paths, from Partner First to Sovereign Ready.
It was built for Caribbean and small-state decision makers, where pooled regional compute and open-weight models often beat a solo megaproject. Jamaica's Project Maestro is a working example of the hybrid path.
The Center for a New American Security tracks how nations invest in sovereign AI capability and where compute, data, and talent concentrate.
Estimates frontier training costs at US$78 million (GPT-4) to US$192 million (Gemini Ultra), with hardware taking 47 to 67 percent of development cost.
Independent estimates of AI infrastructure economics, including roughly US$38 billion of upfront capital for a one-gigawatt AI data centre.
A Caribbean case study in the hybrid path: sovereign data and models without solo megaproject spending.