The announcement came from a summit, not a boardroom. During the Korea-France summit in Paris on September 8 and 9, Samsung Electronics said it would form a strategic partnership with Mistral AI, the French model maker, and take the lead in the company’s Series D round. The deal ties a chip giant to a model company at a moment when the two layers of the AI stack are drawing closer.
The round raised 3 billion euros, about $3.5 billion, at a post-money valuation above 21 billion euros, roughly $24 billion, according to people familiar with the terms. Mistral described it as the largest equity financing in European tech history. EQT’s Scale-Up Europe Fund and PSG Equity, an existing backer, co-led alongside Samsung.
The strategic content is the more interesting half. Samsung plans to run Mistral’s Large models inside its semiconductor design and manufacturing operations, using them for defect detection, equipment optimization, and yield stabilization. The deployment will be entirely on-premise, a choice Samsung framed as necessary to keep its core process data from leaving the building.
The arrangement gives each side something it could not buy with cash alone. Mistral gains a flagship industrial customer that will run its models where the data is most sensitive, a reference case worth more to the company’s enterprise sales than the round itself. Samsung gains a model partner bound to its production line, with an exclusive view into the process knowledge that its competitors guard jealously.
Samsung’s foundry ambitions explain its enthusiasm. The company has trailed TSMC in leading-edge manufacturing for years and has been trying to win credibility with AI workloads. Owning a relationship with a top-tier model company gives it a customer that will consume enormous compute and a story to tell other AI firms about why Samsung’s fabs deserve their business.
Jun Young-hyun, who runs Samsung’s device solutions division, and Arthur Mensch, Mistral’s chief executive, both spoke to the partnership in statements that emphasized the industrial use case over the financial one. The framing is deliberate: this is not a venture investment with a strategic gloss, but a manufacturing integration with a financing attached.
The on-premise detail is the hinge. Model companies have spent years convincing enterprises to send data to the cloud. Samsung is reversing the direction: it will bring the model to the data, inside its own walls. For Mistral, which has positioned itself as the open and deployable alternative to American labs, the deal is proof that its models can run where the biggest industrial customers actually want them.
Analysts said the deal also marks a shift in how chipmakers relate to model companies. Foundries once treated AI firms as customers to be served. Now they are taking equity, shipping hardware, and opening their production lines, all to secure a position in the layer above the silicon. Samsung is not alone; the pattern is spreading as the value in AI consolidates into a few large players.
For Europe, the round is a statement. Mistral has been the continent’s most credible answer to OpenAI and Anthropic, and a record financing led in part by a Korean chaebol is a measure of how far European AI has come and how dependent it remains on capital from elsewhere. The valuation, above $24 billion, puts Mistral in a class with the American leaders, at least on paper.
The risk runs both ways. Samsung is committing process data to a partner whose models will see it, however locally, and whose reliability is still being proven at industrial scale. Mistral is tying a marquee deal to a foundry that has struggled on leading-edge nodes. If Samsung’s fabs underperform, the partnership’s economics tighten for both.
What Samsung gets that money cannot buy is position. A model company embedded in its production line becomes a long-term customer for its chips and a channel to other industrial buyers. The 3 billion euros is the entry fee for that position, and Samsung has decided the price is worth paying.
The summit setting was fitting for a deal built on state-level industrial policy and corporate ambition. Korea wants a stake in the AI layer above its chips. France wants its champion model company funded and deployed. The partnership satisfies both, which is why two chief executives signed it in the shadow of a diplomatic meeting rather than a term-sheet negotiation.
Mistral’s open-weight strategy is part of what makes the on-premise deployment attractive to Samsung. Models that run inside a customer’s own infrastructure, without a cloud round-trip, are the only option for companies that treat process data as a state secret, and Samsung is among the most protective in its industry.
Mistral now has the capital, the valuation, and an anchor customer that will test its models against the hardest industrial problems in semiconductor manufacturing. Whether the models hold up inside a fab, where errors cost millions, will determine whether this round is remembered as the deal that made Mistral or the one that overpromised.
