🇺🇸 Anthropic Mythos, 💰 Baseten and 🌍 Sovereign AI — government, infrastructure and the global response

The U.S. government just banned foreign nationals from using Anthropic's most powerful models. An inference infrastructure startup — Baseten — is raising $1.5 billion. And two sovereign AI startups — Prem and Dream — together capture $360 million to build outside of American control. Three headlines that sound like separate stories, but tell the same one: AI has entered its geopolitical phase, and it's no longer just about who has the best model — it's about who can use it, from where, and under what conditions.

🇺🇸 Anthropic suspends Mythos: U.S. government bans foreign access

On Friday evening, Anthropic suspended all access to its Claude Mythos 5 and Fable 5 models after the U.S. government ordered the company to block access for foreign nationals. It's the first time Washington has directly dictated who can and cannot use a frontier AI model, marking a before and after in AI regulation.

Mythos 5, Anthropic's most powerful model, had been under debate for weeks. Anthropic itself had admitted its ability to find security vulnerabilities was so high it could be considered dangerous in the wrong hands. But the government's decision goes beyond technical safety: it's an act of sovereign technological control. The Trump administration, which last week ordered the military to accelerate AI adoption, is making clear that frontier models are considered national strategic assets, not commercial products.

The move raises uncomfortable questions: what about open-source models that anyone can download? How do you enforce such a restriction when DeepSeek, Kimi, and Qwen advance unhindered from China? And most importantly: are we entering an era of national AI, where every country will have its own models and its own rules?

💰 Baseten raises $1.5B: inference becomes infrastructure

While the government debates who can use the models, the market is solving how to run them. Baseten, a startup offering a cloud platform for AI inference workloads, is raising $1.5 billion in a round that will make it one of the best-capitalized AI infrastructure providers in the world.

Baseten doesn't train models — it runs them. Its platform lets any company deploy open-source and proprietary models with predictable latency, without having to manage GPUs themselves. It's exactly the kind of infrastructure the market needs when hundreds of startups and enterprises want to use AI but can't (and don't want to) build their own NVIDIA clusters.

The round's size — $1.5 billion — tells you where AI money is flowing. It's no longer just "train the best model." It's build the execution layer that lets those models run in production. And Baseten isn't alone: demand for inference-as-a-service is growing faster than model training itself, because at the end of the day, a model without infrastructure to run it is just an academic paper.

🌍 Sovereign AI: Prem ($100M) and Dream ($260M) build outside U.S. control

And here's the other side of the coin. If the U.S. restricts access to frontier models, what do countries that don't want to depend on American infrastructure do? Two startups are answering that question this very week.

Prem, a startup that lets companies and governments deploy AI models in their own environments — on-premise, sovereign, independent of the U.S. cloud — is seeking $100 million in Series A. Its thesis is simple: export restrictions on Anthropic and OpenAI models aren't a temporary problem — they're the new normal. And if you want to control your own AI data, you need your own infrastructure.

Dream, meanwhile, has closed $260 million to build sovereign AI and cybersecurity tools. Dream doesn't just deploy models — it builds the full stack, from hardware to application layer, so governments and large enterprises can operate AI without depending on foreign providers. Together, Prem and Dream raise $360 million in a single week. That's not a coincidence: it's the signal that the sovereign AI market — independent of American control — is moving from niche to category.

🌐 The thread that connects them: AI goes geopolitical

Anthropic suspends Mythos by government order. Baseten raises $1.5B to run models at scale. Prem and Dream raise $360M to build outside U.S. control. Same week. This isn't a coincidence: all three headlines are different faces of the same phenomenon.

When the U.S. government decides who can use frontier models, it triggers a chain reaction: countries that depended on those models start looking for sovereign alternatives, and demand for inference infrastructure — both American and sovereign — skyrockets. The result is a market splitting in two: a U.S. ecosystem of proprietary models with restricted access, and a global ecosystem of open models with local infrastructure.