Prem is raising money into a market that suddenly has a much sharper question: if Washington can switch off access to a model overnight, how much of your AI stack do you really control?
Prem's timing is hard to ignore. The Lugano-based startup is raising a $100 million Series A, Bloomberg reported, after a week that made on-premise AI look less like a conservative infrastructure choice and more like basic risk control. If you're a bank, a hospital, or a government agency in Europe, the lesson from Anthropic's export-control fight is simple. Your most useful AI tool can become unavailable for reasons that have nothing to do with your contracts, your data, or your users.
On June 12, the Trump administration ordered Anthropic to block foreign nationals from using its newest Fable 5 and Mythos 5 models, according to The Verge and other reports. Anthropic said the only practical way to comply was to disable access broadly. Business Insider reported that a person close to Anthropic said the White House gave the company 90 minutes to take the models down. That is not an abstract sovereignty debate. That's a vendor dependency becoming an operational problem before anyone in procurement has time to convene a meeting.
Prem, founded in 2023, has been building directly at that pressure point. The company says its platform lets enterprises fine-tune models, analyze documents, and run API inference inside their own infrastructure, without sending customer data back to Prem's servers. Its earlier $14 million seed round included investors such as Fan Zhang, co-founder of Sequoia Capital China, and David Maisel, founder of Marvel Studios. The company also talks up post-quantum encryption, stateless architecture, and Swiss jurisdiction. You don't have to buy every line of the pitch to see why the pitch has become easier to make.
Before this month, sovereign AI was already moving from conference language into budgets. Gartner has forecast global sovereign cloud infrastructure-as-a-service spending of about $80 billion in 2026, with Europe expected to more than triple from $6.7 billion in 2025 to $23.1 billion in 2027, according to reporting on the firm's projections. That was before Anthropic's most advanced models became a live example of geopolitical access risk.
Frankly, that is what changed. Not the software. Companies have been able to run models on their own hardware for years, and regulated industries have always had good reasons to keep sensitive data close. The difference now is that every compliance officer can point to a real incident, with real model names and a real date, and ask why the organization is still treating cloud AI access as if it were guaranteed.
The same point showed up from another direction this week. The Financial Times reported that JPMorgan Chase cut off access to Anthropic's Claude models for staff in Hong Kong after deciding that Anthropic's licensing terms exclude use across Greater China, including Hong Kong. Goldman Sachs had made a similar move earlier in June. For a global bank, that is more than a technical nuisance. If one office can use a model and another cannot, the risk sits inside the workflow.
This is the market Prem wants. Not the founder trying to shave a few cents off inference costs, but the enterprise that cannot afford to learn, six months into deployment, that its chosen model is restricted in Singapore, Hong Kong, or Frankfurt because a licensing term or export-control order shifted somewhere else. A system where the model runs on your hardware, under your policies, has a different value after that.
Switzerland is part of the product
Lugano is not a decorative detail here. Switzerland's revised Federal Act on Data Protection came fully into force in September 2023, and the country sits outside the European Union while still benefiting from an EU adequacy decision for data transfers. That does not make a Swiss AI company immune to every foreign rule, and Prem would be foolish to pretend otherwise. But for an enterprise trying to reduce dependence on American cloud vendors, Swiss incorporation is a concrete fact it can put in front of auditors.
The competitive field is already crowded. Mistral has pushed hard on European AI infrastructure, including an $830 million debt financing for a data center near Paris, according to Reuters. Aleph Alpha has built its name around sovereign deployments for governments and regulated companies. The hyperscalers are not sitting still either. AWS has been marketing its European Sovereign Cloud, and Google, Microsoft, IBM and others know exactly why customers are asking harder questions about control, locality, and legal exposure.
Prem's advantage, if it has one, is focus. It isn't trying to be the most famous chatbot brand in the world. It is selling the less glamorous thing that suddenly matters: local deployment, private data flows, and a stack the customer can explain to a regulator without waving vaguely at an American cloud contract.
That is a narrower business than frontier-model fame, but it may be a better one in 2026. If Bloomberg's reported $100 million Series A closes, Prem will have the capital to test whether the new demand for sovereign AI is deep enough to support another serious European infrastructure company. The answer won't come from slogans about digital independence. It will come from banks, insurers, hospitals, and public agencies deciding whether the next outage should be someone else's political decision or their own technical choice.
Also read: Dream raises $260 million at a $3 billion valuation as sovereign cybersecurity AI hardens into its own asset class • Apple's iPhone 18 Pro will cost more because AI hyperscalers are hoarding memory chips • Blackstone is betting billions of dollars of borrowed money that AI infrastructure demand in Asia-Pacific cannot wait
