A company calling itself a vendor of "sovereign AI" is, by construction, not sovereign at all. It is a counterparty. That tension sits at the center of a $260 million funding round announced on 2026-06-18 by Dream, a Tel Aviv and Vienna-based startup positioning itself as a builder of government-controlled artificial intelligence infrastructure. The company says the round values it at $3 billion and will scale a sales pitch that asks governments to trade one form of dependency for another.
In the company's release announcing the round, "sovereign AI" means systems a government can run inside its own borders, on infrastructure it controls, with training data and model weights it can audit and exit with. The pitch lands in a market that genuinely wants this. European Union member states, Gulf monarchies, and parts of Southeast Asia have all said, in policy documents and procurement language, that dependence on a small number of American and Chinese frontier model providers is a strategic liability. Dream is one of a handful of vendors positioning to fill that gap.
The two co-founders most associated with the company are Shalev Hulio and Sebastian Kurz. Hulio built and ran NSO Group, the Israeli spyware maker whose Pegasus tool was deployed by foreign governments to surveil journalists, human rights workers, and at least one murdered Saudi columnist. The U.S. Commerce Department placed NSO on an export control list in 2021 over the abuse cases. Hulio has not been personally charged in connection with Pegasus, and NSO has disputed the abuse findings. Kurz resigned the Austrian chancellorship in 2021 after prosecutors opened corruption investigations into testimony he gave about a tabloid publisher friendly to his party. Austrian courts have since delivered split outcomes, with Kurz convicted on a charge of making false statements and acquitted on a separate corruption count, and additional proceedings resolved through settlements or dropped.
What that background does to Dream's pitch depends on what the contracts actually say. A sovereign AI deal in this market typically commits a government to a vendor for model access, deployment, and ongoing maintenance, with optional clauses for data residency, audit rights, and exit. The data residency question is where the label tends to give way to the contract. A model can run inside a country's borders and still be governed by an export-controlled foreign supplier, updated on a foreign schedule, and inspectable only through the vendor's own tools. The Dream release names no government customer, no deployed country, and no technical contents of the deals it says it has signed.
The company says it has secured nearly $300 million in total contract value since beginning commercial operations in late 2024, and that the new round, co-led by Bicycle Capital and Group 11 with participation from Antler, Bain Capital Ventures, Tru Arrow Partners, and unnamed global investors, will fund expansion into Europe, the Middle East, Asia, and the Americas. Both the contract total and the $3 billion valuation are self-reported in a press release. No filing, named investor confirmation, or independent press report has corroborated either number as of the announcement.
The interesting question is not whether Dream can raise the money. Investors fund many companies whose products never quite deliver. The interesting question is whether a government that signs with Dream ends up with the capability Dream's marketing claims, or with a different kind of lock-in wrapped in different language. The answer, for any specific deal, will live in the audit rights clause, the exit terms, and whether the trained model can be inspected, modified, and taken elsewhere without the vendor's permission. Those details are not in the release. The first government contract a journalist gets to read will tell a lot more than the funding round.