30亿欧元。
这不是某个美国AI独角兽又一轮夸张的融资，而是一家成立不到三年的法国公司，在周二宣布拿到的D轮。9月8日，Mistral AI正式对外确认完成30亿欧元（约35亿美元）的D轮融资，投后估值超过210亿欧元（约245亿美元）。三星电子领投，EQT旗下Scaleup Europe Fund和现有投资者PSG Equity联合领投。
这笔融资有三个纪录：欧洲科技史上最大单笔股权融资、开源大模型阵营估值最高的公司之一、今年全球范围内金额最大的AI公司融资。更重要的是，它不是发生在旧金山、不是在西雅图、不是在北京——它发生在巴黎。
为什么是三星，为什么是Mistral
三星领投这件事，比30亿欧元这个数字更值得琢磨。
过去几年，全球AI算力和模型的话语权高度集中在英伟达和少数美国公司手里。OpenAI绑定微软、Anthropic绑定Google、xAI绑定马斯克和自己的数据中心——每一家前沿大模型公司背后，都站着一个算力巨头。这种「软件+硬件」的深度绑定，让后来者很难有机会。
但三星是个例外。这家韩国巨头手里有全球最先进的存储芯片和HBM产能，有自己的Exynos处理器，有手机、电视、家电等海量终端场景，唯独缺一个能打的AI基础模型。它之前也投过AI公司，但更多是财务投资。这次不一样。
Mistral对三星的价值，是补全了三星AI战略的最后一块拼图：从芯片到设备到模型的全链路。想想看——三星的手机、电视、家电、可穿戴设备，全球出货量超过10亿台，这些设备都需要端侧AI能力。Mistral的小模型路线（7B、8x7B MoE）天然适合端侧部署，两者的契合度极高。
而三星对Mistral的价值，则是实实在在的硬件支持和场景落地。有了三星这个战略投资人，Mistral在端侧模型的优化上可以直接拿到最底层的硬件信息，在手机、电视等场景的落地可以走得比其他模型快几个身位。这不是一笔单纯的财务投资，这是一次深度的战略结盟。
"闭源模型越走越贵、越走越封闭的时候，开源阵营需要一个能站在最前排的选手。Mistral正在成为那个选手。"—— 一位AI行业投资人的判断
开源AI的第二极，终于站到了台前
Mistral的崛起，有一个更大的背景：开源大模型阵营正在重新洗牌。
过去两年，开源模型的叙事基本由Meta的Llama系列主导。Llama 2、Llama 3，每一代发布都会带动一波开源创新浪潮。但Llama有一个根本性的问题：它不是真正的开源。商用需要授权、有使用门槛、有各种限制条件，与其说是开源，不如说是「开放权重」。
Mistral走的是另一条路。从Mistral 7B到Mixtral 8x7B，再到后来的Mistral Large和Mistral Small，这家法国公司在开源社区的声望一路走高。它的模型不仅性能强劲，而且对开发者极其友好——权重开放、API便宜、文档清晰、社区活跃。
更关键的是，Mistral证明了一件事：开源模型不仅能在实验室里跑分，还能在商业上赚钱。据业界估算，Mistral的年化收入已经达到数亿欧元级别，客户遍布欧洲和北美。这在开源模型公司里是罕见的——大多数开源模型团队要么靠融资烧钱、要么靠云厂商采购，自身造血能力很弱。
210亿欧元的估值是什么概念？作为对比，去年同期大家热议的Anthropic估值约为600亿美元，OpenAI估值超过1500亿美元。Mistral的估值虽然只有它们的几分之一，但考虑到Mistral走的是开源路线、团队规模小得多、成立时间短得多，这个数字已经相当惊人。
更重要的是位置。在全球AI大模型的版图上，美国有OpenAI、Anthropic、Google、Meta，中国有智谱、月之暗面、DeepSeek、通义千问。欧洲一直是个空白——有研究、有人才、有市场，但缺少一家能打的本土大模型公司。Mistral正在填补这个空白。
三个层面的连锁反应
这笔融资的影响，会在三个层面展开。
第一个层面是市场竞争。有了30亿欧元的弹药，Mistral会在算力投入、模型研发、商业化拓展三个方向同时加速。之前很多人觉得开源模型追不上闭源第一梯队——性能差距确实存在，但Mistral正在用实际行动缩小这个差距。有了充足的算力和人才储备，下一代Mistral模型可能会在更多基准测试上逼近甚至超过闭源模型。这对整个行业都是好事——竞争越激烈，进步就越快，价格就越低。
第二个层面是地缘政治。美国AI出口管制越收越紧，欧洲和亚洲的企业都在寻找「不依赖美国」的替代方案。Mistral作为一家欧洲公司，天然不受美国出口管制的直接约束，这对很多企业来说是巨大的吸引力。三星的投资更是强化了这个定位——一个欧洲模型公司，加一个亚洲硬件巨头，等于一条绕开美国的AI供应链。
第三个层面是开源生态。过去开源模型的最大痛点是「没人愿意持续投入重金研发」——开源不赚钱是行业共识。但Mistral用210亿的估值告诉市场：开源也能长出大公司，也能获得高估值。这会吸引更多资本进入开源AI领域，形成正循环。资本进来了，人才就来了；人才来了，模型就进步了；模型进步了，商业价值就兑现了。
不是终局，是开局
当然，210亿欧元估值的Mistral也面临巨大的挑战。
最大的挑战是算力。前沿大模型的训练越来越烧钱，30亿欧元听起来很多，但在大模型训练面前，可能也就够训练一两个旗舰模型的。OpenAI一年的算力投入是百亿美元级别的，Mistral要追这个差距，30亿只是起步。
第二个挑战是商业化。开源模型的商业化一直是个难题。Mistral目前的收入主要来自API服务和企业定制，这两块的竞争都非常激烈——上面有OpenAI、Anthropic的闭源模型压着，下面有无数开源小模型在卷价格。如何在保持开源初心的同时实现大规模商业化，是Mistral必须回答的问题。
第三个挑战是人才。欧洲的AI人才储备虽然不错，但和美国比还是有差距。Mistral要扩张团队，必须从全球招募顶尖人才，这意味着更高的薪酬、更激烈的人才争夺。
但无论如何，Mistral的这一轮融资都是一个标志性事件。它标志着几件事：
第一，开源AI终于有了一个能站在舞台中央的玩家。之前开源模型虽然热闹，但更多是社区驱动、小团队玩票。现在有了三星背书、210亿估值的Mistral，开源阵营在和闭源巨头对话时，腰板可以更硬了。
第二，AI的地缘格局正在从「美国独大」走向「多极化」。美国有OpenAI/Google/Anthropic，中国有智谱/DeepSeek/通义，欧洲有Mistral——三足鼎立的雏形已经显现。这对整个行业的健康发展是好事，没有任何一个地区应该垄断AI的未来。
第三，硬件和软件的结盟正在成为新常态。三星+Mistral、英伟达+一众云厂商、微软+OpenAI、Google+Anthropic——每一个前沿AI玩家背后，都站着一个甚至多个硬件巨头。AI竞争已经不是模型之间的竞争，而是生态系统之间的竞争。
30亿欧元不是终局，是Mistral下一轮冲刺的开局。AI产业的故事，远比我们想象的更漫长、更精彩。
明天见。
Three billion euros.
This isn't another overhyped funding round from a San Francisco AI unicorn. It's a Series D announced on Tuesday by a French company founded less than three years ago. On September 8, Mistral AI officially confirmed closing a €3 billion (~$3.5 billion) Series D at a post-money valuation of over €21 billion (~$24.5 billion). Samsung Electronics led the round, with EQT's Scaleup Europe Fund and existing investor PSG Equity co-leading.
This round sets three records: Europe's largest single equity round in tech history, one of the highest valuations in the open-source LLM camp, and the largest AI company funding round globally this year. Most importantly — it didn't happen in San Francisco, or Seattle, or Beijing. It happened in Paris.
Why Samsung, Why Mistral
Samsung leading the round is more telling than the €3 billion figure itself.
Over the past few years, global AI compute and model power have been heavily concentrated in Nvidia and a handful of American companies. OpenAI is tied to Microsoft, Anthropic to Google, xAI to Musk and its own data centers — behind every frontier LLM company stands a compute giant. This deep "software + hardware" coupling has made it extremely hard for newcomers to break in.
But Samsung is the exception. The Korean giant has the world's most advanced memory chips and HBM capacity, its own Exynos processors, and massive end-device scenarios across phones, TVs, and home appliances. What it's missing is a competitive AI foundation model. It has invested in AI companies before, but mostly financially — this time is different.
Mistral's value to Samsung is completing the last piece of Samsung's AI strategy puzzle: an end-to-end chain from chip to device to model. Think about it — Samsung ships over a billion devices globally: phones, TVs, appliances, wearables. All of them need on-device AI capabilities. Mistral's small-model approach (7B, 8x7B MoE) is naturally suited for edge deployment. The fit is almost perfect.
And Samsung's value to Mistral is tangible hardware support and scenario deployment. With Samsung as a strategic investor, Mistral can get the lowest-level hardware information directly for edge model optimization, and can move faster than any other model in deploying to phones, TVs, and more. This isn't purely financial — it's a deep strategic alliance.
"As closed-source models grow more expensive and more locked down by the day, the open-source camp needs a player that can stand on the front row. Mistral is becoming that player."— An AI industry investor's assessment
Open-Source AI's Second Pole Finally Steps Into the Spotlight
Mistral's rise fits into a bigger picture: the open-source LLM landscape is reshuffling.
For the past two years, the open-source narrative has been dominated by Meta's Llama series. Llama 2, Llama 3 — each release has driven a wave of open-source innovation. But Llama has a fundamental problem: it's not truly open source. Commercial use requires licensing, there are usage thresholds, all kinds of restrictions. It's less "open source" and more "open weights."
Mistral is taking a different path. From Mistral 7B to Mixtral 8x7B, to Mistral Large and Mistral Small, the French company's reputation in the open-source community has skyrocketed. Its models don't just perform well — they're extremely developer-friendly: open weights, cheap APIs, clear documentation, an active community.
More critically, Mistral has proven something: open-source models don't just score well in benchmarks — they can make money. Industry estimates put Mistral's annualized revenue at several hundred million euros, with customers across Europe and North America. That's rare among open-source model companies — most either burn through funding or rely on cloud vendor purchases, with weak self-sustaining revenue.
What does a €21 billion valuation mean? For context, Anthropic was valued at roughly $60 billion around this time last year, and OpenAI at over $150 billion. Mistral's valuation is only a fraction of theirs — but considering Mistral's open-source route, much smaller team, and much shorter history, the number is already staggering.
What matters more is position. On the global LLM map, the U.S. has OpenAI, Anthropic, Google, Meta. China has Zhipu, Moonshot, DeepSeek, Qwen. Europe has been a blank spot — research, talent, and market exist, but there's been no competitive homegrown LLM company. Mistral is filling that gap.
Three Layers of Ripple Effects
The impact of this round will unfold on three levels.
Level one: market competition. With €3 billion in ammunition, Mistral will accelerate on three fronts simultaneously — compute investment, model R&D, and commercial expansion. Many people used to think open-source models could never catch up to closed-source first-tier players — the performance gap is real, but Mistral is narrowing it with actual execution. With sufficient compute and talent, the next generation of Mistral models could approach or even surpass closed-source models on more benchmarks. That's good for the entire industry — more competition means faster progress and lower prices.
Level two: geopolitics. As U.S. AI export controls tighten further and further, companies in Europe and Asia are looking for alternatives that "don't depend on the U.S." As a European company, Mistral is naturally not directly subject to U.S. export controls — a huge draw for many enterprises. Samsung's investment reinforces this positioning: a European model company plus an Asian hardware giant equals an AI supply chain that bypasses the U.S.
Level three: the open-source ecosystem. Historically, the biggest pain point for open-source models has been "nobody wants to pour serious money into R&D" — the consensus has been that open source doesn't make money. But Mistral, with its €21 billion valuation, is telling the market: open source can also grow big companies and command high valuations. This will attract more capital into open-source AI, creating a positive flywheel. Capital comes in, talent follows; talent arrives, models improve; models get better, commercial value materializes.
Not the Endgame — the Opening
Of course, Mistral at €21 billion also faces enormous challenges.
The biggest challenge is compute. Training frontier LLMs is getting more and more expensive. €3 billion sounds like a lot, but in the face of large model training costs, it might only be enough to train one or two flagship models. OpenAI's annual compute spend is in the tens of billions of dollars. For Mistral to close that gap, €3 billion is just the starting line.
The second challenge is commercialization. Monetizing open-source models has always been hard. Mistral's current revenue comes mainly from API services and enterprise customization — both fiercely competitive markets. Above them are the closed-source incumbents like OpenAI and Anthropic; below them, countless open-source small models are racing prices to the bottom. How to achieve large-scale commercialization while staying true to its open-source roots is the question Mistral must answer.
The third challenge is talent. Europe has decent AI talent reserves, but still lags behind the U.S. To expand its team, Mistral will have to recruit top talent globally — meaning higher compensation and fiercer talent competition.
Regardless, this round is a landmark event. It signals three things:
First, open-source AI finally has a player that can stand center stage. Previously, although open-source models had buzz, it was mostly community-driven, small-team side projects. Now, with Samsung's backing and a €21 billion valuation, the open-source camp can stand a little taller when negotiating with closed-source giants.
Second, AI's geopolitical landscape is shifting from "U.S. dominance" to "multipolarity". The U.S. has OpenAI/Google/Anthropic, China has Zhipu/DeepSeek/Qwen, Europe has Mistral — the outline of a three-pole world is already visible. That's healthy for the industry's long-term development. No single region should monopolize AI's future.
Third, hardware-software alliances are becoming the new normal. Samsung+Mistral, Nvidia+cloud vendors, Microsoft+OpenAI, Google+Anthropic — behind every frontier AI player stands one or more hardware giants. AI competition is no longer between models — it's between ecosystems.
€3 billion isn't the endgame. It's the starting line for Mistral's next sprint. The story of the AI industry is far longer and more fascinating than any of us imagined.
See you tomorrow.
Mistral · €3B Series D · Samsung lead · €21B valuation · open-source LLM · European AI · second pole of open source · hardware-software alliance · multipolar AI
