ANALYSIS Two government-backed funding moves this week expose the divergent strategies nations are adopting to secure their positions in the AI supply chain: Canada is writing checks directly into a foundation-model company, while South Korea is financing the power plants that AI infrastructure demands. Together they represent a new phase in which sovereign AI ambitions are denominated not in policy papers but in tens of billions of dollars.
Why it matters
Cohere is in advanced talks to raise between $2 billion and $3 billion at a $20 billion valuation, with financing from the Canadian government and existing backers1. Separately, the South Korean government is preparing to announce an investment of over $100 billion for eight nuclear power plants and a natural-gas project in Texas, part of a broader $350 billion commitment to the United States3. ◆ The two deals sit at opposite ends of the AI stack, one funding model development, the other funding the energy layer, but both are propelled by the same logic: governments treating AI capacity as a matter of national interest worth underwriting at scale.
The big picture
South Korea's energy commitment traces back to an October 2025 trade agreement in which Seoul pledged $350 billion in U.S. investment in exchange for the U.S. reducing its tariff rate from 25 percent to 15 percent on most South Korean imports. That package included $200 billion for strategic investments, $100 billion of which was earmarked for energy projects, and $150 billion for the U.S. shipbuilding sector. The nuclear plants will likely be built on federal land and use the AP1000 design by Westinghouse Electric, though both sides have discussed having at least some reactors be the APR1400, built by South Korean manufacturer Kepco. A memorandum of understanding for the energy deal could be finalized as early as September 18, according to the Seoul Economic Daily. Japan has also agreed to play a role in the Trump administration's $80 billion buildout of Westinghouse AP1000 reactors, and last October Japan's trade ministry announced it would invest $550 billion in the U.S. in return for lower tariffs.
Canada's approach is more direct. By participating in Cohere's raise, the Canadian government is placing capital into a company that has built its brand around AI sovereignty for regulated industries2. Cohere this week released North Small Translate 1.0, a 218-billion-parameter mixture-of-experts model built for machine translation across over 50 languages and locale variants, with 25 billion active parameters and a 16,000-token context window.
Between the lines
Cohere's sovereignty pitch, however, carries a tension that its own licensing choices make visible. North Small Translate's FP8 weights are available on Hugging Face under the CC BY-NC 4.0 license for non-commercial use, but enterprises that want to put the model into production must purchase a commercial license and deploy it through Model Vault. In June, Cohere pitched its North Mini Code release under an Apache 2.0 license without comparable restrictions, framing it as a response to developers demanding the same sovereignty guarantees that regulated industries have long required. ANALYSIS The shift from Apache 2.0 on one model to CC BY-NC on another suggests Cohere is calibrating openness model by model, granting sovereignty-style on-premises control only to paying customers for its translation capability.
South Korea's calculus operates on a different plane. President Trump threatened to reimpose a 25 percent tariff rate in January because the South Korean National Assembly had not yet approved the deal. The South Korean government created a state-run corporation earlier this year to select, finance, and manage investments in the U.S. and make recommendations to Trump. ◆ The institutional machinery Seoul is building around these commitments indicates the investment is as much a trade-diplomacy instrument as an energy strategy, with AI infrastructure serving as the justification that aligns both governments' interests.
Financing eight nuclear power plants and a natural-gas project will likely exceed the $100 billion South Korea earmarked for American energy projects last October. The eight plants alone are expected to total $120 billion, and expected costs for a natural-gas plant in Encinal, Texas, are around $22 billion. For reference, it cost $35 billion to build two AP1000s in Georgia. ◆ Those cost benchmarks suggest the timeline and budget risk on the nuclear side are substantial, making the gap between pledged capital and delivered megawatts the variable that will determine whether AI-driven energy demand actually gets met.
What's next
The Seoul Economic Daily reported that a memorandum of understanding for the energy deal could be finalized as early as September 18. Cohere's raise, if closed at the reported range, would make it one of the largest single rounds for a Canadian AI company, with the government's participation formalizing Ottawa's stake in the sovereign-model layer. ◆ Whether these commitments produce competitive AI infrastructure or remain diplomatic gestures will depend on execution: reactors that take a decade to build and model licenses that gate commercial use both test the premise that sovereignty can be purchased at speed.