European AI model rising star Mistral's valuation jumped 80%! The era of closed source and open source AI is flying hand in hand, and computing power demand is moving towards a “sea of stars”

The Zhitong Finance App learned that Mistral AI, a European AI startup supported by lithography giant Asmack, said on Tuesday local time that the company raised about 3 billion euros (about 3.5 billion US dollars) in a new round of financing led by global memory chip giant Samsung Electronics. The French startup is seeking to position itself as the European strategic AI leader, leading European sovereign AI systems, and seeking to accelerate the replacement of non-European competitors such as OpenAI and Anthropic in the European market.

According to information, participants in this round of investment are also EQT Investment Management Company, Scaleup Europe Fund, a large-scale investment fund supported by the European Union, and PSG Equity, an existing institutional investor. Mistral said that the current round of financing made its latest valuation after investment exceed 21 billion euros. By contrast, a year ago, after completing a round of financing led by Dutch semiconductor equipment manufacturing giant Asmack, Mistral's valuation was around 11.7 billion euros, which meant that the AI startup's valuation jumped 80% in just one year.

According to information, recent global technology industry leaders, including NVDA.US (NVDA.US), Microsoft (MSFT.US), Facebook parent company Meta (META.US), established US tech giant IBM (IBM.US), and Silicon Valley venture capital giant Andreessen Horowitz, are urging the US government to support the development trajectory of AI big model technology with open-weight AI (open-weight AI) as the core.

Mistral AI is one of them, and the AI startup's technology development path also follows the open source AI big model path around open weights. Open weighting means that enterprises can fully deploy, fine-tune, and access the business in their own servers or private clouds according to specific license download model weights, and enhance data autonomy and supplier selection rights. However, Mistral also provides a proprietary business model, so Mistral AI is more accurately positioned as a European AI developer that uses an open model to expand the ecosystem and commercialize AI applications with high-end enterprise customization, complete deployment services, and large-scale computing power platforms.

In the AI developer ecosystem, open-weight AI usually refers to an “open weight AI big model”, that is, developers can download parameters after model training and deploy, fine tune, and reason in their own server or cloud environment; however, model developers do not necessarily disclose training data, data processing methods, training code, and complete architecture details, and may also restrict commercial use or specific applications through licenses.

Open source AI in the strict sense of the word is broader: in addition to weight, it should also provide information sufficient to allow users to research, modify, and reproduce the system's code, architecture, and training data, and guarantee basic freedom of use, research, modification, and redistribution. Therefore, open weight is “the finished model can be used and modified,” while true open source AI further opens up “how the model is manufactured”; the former can be part of the latter, but open weight is not necessarily equal to open source all technical aspects and code.

As for the AI computing power industry chain and AI super bull market investment trend that global investors are focusing on, opening up weight, model routing, and improving architectural efficiency will reduce the cost of intelligence per unit, but may amplify total computing power demand through the “Jevens paradox.” As simple call costs decrease, enterprises will deploy more continuously running agents, parallel sub-agents, long context analysis, code automation, and real-time multi-modal services; a single task may consume less computing power, but the number of tasks, inference steps, and deployment nodes may grow faster. Even if only a small number of experts are activated, very large MoE models such as KIMI K3 still need to distribute huge weights in high-capacity memory and high-speed interconnect clusters. Therefore, the popularity of open models will spread computing power requirements from a few closed source laboratories to global cloud computing service providers, sovereign clouds, giant AI data centers and local inference clusters for enterprises.

Recent research by Google's parent company Alphabet clearly indicates that a reduction in the unit token price, an extension of task execution time, and an increase in the number of applications can occur simultaneously, driving up the total demand for AI inference and the total value of inference, which can be described as an accurate response to the so-called “Jevens paradox.” According to the forecast report of TrendForce, a well-known market research agency, in 2027, shipments of NVL72 racks increased by more than 50% year on year. The total output value of GB200/GB300 and next-generation NVL rack systems around Nvidia AI GPUs based on the Vera Rubin architecture is expected to exceed US$710 billion, an increase of 214% year over year, including the impact of product upgrades and price increases.

Behind the $24 billion valuation: Mistral accelerates construction of European AI alternatives

Mistral CEO Atil Mensch said in an interview with US media on Tuesday that the huge amount of money from this financing will be used to build more infrastructure, including the company's own data center, and develop computing power leasing business.

“In the long run, our plan is to fully rely on the European AI computing power cluster we have built ourselves, so this means that the scale of computing power we have will grow exponentially by about 100% over the next five years,” Mensch said.

The CEO added that the company will train “larger and faster models.”

This Paris-based AI startup is actively developing original AI models and is expanding the layout and construction process of its own AI data center. The company hopes to adopt a closed-source AI model path different from OpenAI and Anthropic, and cooperate with enterprises one by one while promoting an open weighted AI model to create a customized AI application superplatform that meets their needs, and actively integrate these tools into the enterprise business.

Mistral has incorporated cutting-edge AI technology into manufacturing processes in Asmack, the Netherlands. Menshi said that the latest in-depth cooperation between Mistral and South Korea's Samsung Electronics will focus on similar customization fields.

Mensch said earlier this year that he expects Mistral's annual recurring revenue data this year to surpass the $1 billion mark. As new financing and partnerships are reached, Mensch said he expects to “significantly exceed” this figure “if everything evolves according to current trends.” He declined to provide a revised revenue forecast.

“We are very confident that today's financing is also accelerating growth and setting the stage for further strong growth in 2027.” Menshi said in an interview with the media.

Mistral has always sought to position itself as a European alternative to the AI industry that is neither the US nor China, in line with the growing strong demand for “sovereign AI systems” in the European market and other regions. Investors in the EU-supported large-scale investment fund Scaleup Europe Fund, which is one of the main players in this round of financing, include European Commission funds, as well as large European companies such as Novo Holdings and Santander, which are also from Europe.

Unlike US AI application giants OpenAI and Anthropic, which have always followed a closed source path, Mistral focuses on open weighted AI models, in stark contrast to OpenAI and Anthropic's proprietary closed systems. However, the company is facing fierce competition from strong Chinese competitors.

Mensch said that the models that Mistral will release “soon” will be “very competitive.” He pointed out that China's cutting-edge AI laboratories are not actually doing business or cooperating with corporate clients on a large scale outside of China. Mistral said that in some cases, the big open source AI model led by Chinese companies can be deployed on Mistral's computing power infrastructure, but this “will not form a strong dependency on Chinese AI laboratories” because the data remains here at Mistral.

Mensch said that it is difficult for European companies to rely on China's open source AI models for long-term support, because it is unclear whether these models will continue to be upgraded over a long period of time, and it is also unclear whether they will be subject to any export restrictions.

“Currently, we see that the fluctuations in this area are actually quite drastic,” Monsch said. “Now we also need to be a trusted partner for our customers. “They want us to clearly guarantee that in a year they will be able to use a better model than today, and the only way we can provide that guarantee is to continue to train the model exclusively on our own,” Monsch said.

Samsung bets on new European AI forces: from model financing to AI computing power demand and accelerated expansion of industrial applications

The core significance of Mistral's current round of financing is to transform Europe's “sovereign AI” demand into its own computing power, model development, and continuous investment in industrial-side applications for enterprises. For Mistral, competitive advantage needs to be built on model capabilities, data control, deployment services, and continuous upgrade guarantees; for European and North American AI computing power providers, more deployable models mean a wider pool of potential customers. Therefore, Samsung's investment in Mistral not only relates to Europe's independent AI needs, but also a major business opportunity for industrial companies to embed AI into their daily production processes.

Samsung's takeover of Asmack's lead investment also gave Mistral an opportunity to further deepen cooperation into the semiconductor manufacturing scene. Its business model focuses on combining enterprise data, customized models, business system integration, and long-term services; management expects this year's recurring revenue to exceed 1 billion US dollars, which largely indicates that the company is using corporate payment requirements to support an infrastructure expansion path around increasingly strong AI computing power requirements.

In addition to the huge demand for computing power brought about by the big open source AI model and the Astra model that OpenAI has just released, sparked a buzz on AGI — Nvidia CEO Hwang In-hoon also made a big statement on social media on Sunday that the launch of GPT-6 Astra means “AGI has arrived”. The demand for computing power has already been reflected in chip revenue and multi-year procurement commitments. Nvidia announced quarterly revenue of US$96.2 billion in August, up 106% year on year; data center revenue was US$89 billion, up 117% year on year.

On the AI computing power resource procurement side, according to media reports, Anthropic signed a cloud computing agreement worth 35 billion US dollars with Nvidia's supported Lambda and reached a six-year, 45 billion US dollar computing power lease arrangement with Nscale. The latter involved about 460 megawatts of infrastructure. The two deals totaled 80 billion US dollars, reflecting the strength of Frontier Laboratories to lock in computing power supplies for many years in advance, and also provided a foundation for training and the expansion of intelligent reasoning.

The upstream AI computing power industry also sent a strong signal. South Korea's total exports in August reached 98.25 billion US dollars, up 68.7% year on year; of these, semiconductor exports around SK Hynix and Samsung, the world's two largest memory chip giants, reached 46.65 billion US dollars, an increase of 209 percent over the previous year, a record high. The South Korean government directly linked the strong performance of semiconductor exports to hyperscale cloud vendors such as Google and Amazon to expand investment in AI computing power infrastructure.

At the same time, AI server memory chip components are still the clearest supply bottleneck at the AI computing power industry chain level. TrendForce predicts that in 2026, server DRAM contract prices will increase by about 270%, and enterprise-grade SSD prices will increase by about 235%; HBM contract prices may still rise 70% to 140% in 2027. These data reflect the combined effects of AI computing power expansion and storage price increases. According to TrendForce's latest estimates, the combined share of DRAM and NAND in capital expenditure of major cloud service providers will rise from 47% in 2026 to 68% in 2027, behind which there is a simultaneous increase in procurement volume and price increases.

The dual growth drivers of AI computing power demand: open models broaden applications, and closed source models such as Astra are expected to continue to raise the upper limit of capacity

China's open weight model competition is driving the spread of AI deployment from a few cutting-edge laboratories to private enterprise environments, regional clouds, and sovereign clouds. Major US technology companies such as Nvidia and Microsoft support open weights. Their industrial logic is to lower the threshold for model acquisition and customization, so that more companies around the world can participate in AI application development, thereby boosting global demand for AI computing power resources.

China's open weighting models such as Kimi are lowering the threshold for enterprise migration models, customization models, and autonomous deployment. Kimi K3 has a total parameter of 2.8 trillion yuan and supports about 1 million token contexts; as of September 8, the Moonshot AI vendor channel price shown by OpenRouter is $3 for each million input tokens and $15 for output tokens. Its competitiveness comes from a combination of capability, price, freedom of deployment, and interface compatibility. Open weighting allows companies to run and transform models on their own infrastructure, but this does not mean that training data and complete R&D processes are all open source.

In addition, “model routing platforms” (such as OpenRouter), which combine open source and closed source AI models, have begun to enable enterprises to allocate AI budgets according to tasks, and economic evaluation has also shifted to “the total cost of each successful task.” Simple tasks are handed over to low-cost models, and complex tasks are called on cutting-edge models, which can reduce unnecessary expensive reasoning. According to the enhanced search generation test published by Amazon, intelligent routing allocates 87% of requests to Claude 3.5 Haiku, saving an average of 63.6% of costs while maintaining the larger Sonnet 3.5 v2 benchmark accuracy of the larger model.

Astra represents another demand expansion mechanism: the ability of large models is increased, making tasks that were previously difficult to complete reliably into a commercialable range. OpenAI revealed that Astra achieved 98% in the FrontierMath Level 4 test and 99.9% in ARC-AGI-3. Based on this, Hwang In-hoon expressed the judgment that “AGI has arrived” and stated that the model used more than 100,000 Nvidia GPUs for training, and 400,000 GPUs will soon be launched. It is worth noting that “AGI has arrived” is still a controversial judgment, and the announcement of the deployment of the larger Nvidia AI GPU cluster directly reinforces the strong AI computing power demand expectations that the large front-end AI model will continue to expand investment in training resources.

According to a research report published by Rich Privorodsky, the head of the Wall Street financial giant Goldman Sachs Delta One trading desk, what the Goldman Sachs trading desk values most is that Astra may shift the overall demand curve — that is, when the AI model is smarter, companies can try work that could not be reliably done before, and competitors also need to continue to invest in R&D and training, which provides new support for the AI spending cycle; the SoftBank ADR as described in its research report has risen sharply by more than 10%, and Oracle's 5.5% increase, reflecting the market's potential for Astra-led computational power expansion Repricing. Implemented to Nvidia, Astra's computing power requirements and future updates and iterations, and working with the Recursive Self-Improvement (RSI) technology route to strengthen cutting-edge computing power requirements, Hugging Face expands developer coverage, Blackwell and Vera Rubin undertake product delivery, and strong profit growth supports valuation expansion.

The Delta One trading desk said that the cutting-edge AI capabilities brought by Astra have broken through to the AGI era, can continuously create new applications and force competitors to continue to invest, thereby extending the AI capital expenditure cycle. This is a major complement to the open weighting model to reduce costs: the cutting-edge model increases the final complexity of deliverable tasks, and this breakthrough in cutting-edge capabilities has effectively expanded the scope of business that AI can handle. Open weighting and model routing expand the range of global customers that can bear the AI operating model. Open weights and model routes can be used through flexible deployment and optimization models. Improve the economic viability of more application scenarios and jointly promote the expansion of the scope of AI use and deployment scale.