Mistral pivots to enterprise data to train AI after exhausting internet sources

French artificial intelligence (AI) company Mistral is embedding its engineers directly into client offices to solve a costly problem: most businesses see no return from their AI investments. The Paris-based company will integrate its AI scientists and engineers into partner organizations to train models on proprietary data.

CEO Arthur Mensch said the move responds to widespread failures in implementing AI across corporate sectors. “The very high-tech companies [and] a couple of banks are able to do it on their own,” Mensch told the Wall Street Journal. “But when it comes to getting some [return on investment] from use cases, in general, they fail.”

A July study from the Massachusetts Institute of Technology found 95 percent of organizations report zero return from their AI spending. Companies worldwide have invested billions in AI tools that have failed to deliver the promised results.

Mistral’s solution embeds its technical staff within client companies to work directly with company data that has never been used for AI training.

“The curse of AI is that it looks like magic. So you can very quickly make something that looks amazing to your boss,” Mensch explained. But those flashy prototypes rarely scale to solve real business problems.

Public data sources that powered the first wave of AI models have reached their limits. Major tech companies have already scraped most available internet content for training purposes.

“For the last three years, we’ve been able to compress human knowledge and make models increase across the board,” Mensch stated. “But now we’re reaching a saturation point there and that means the next frontier is in getting access to a new kind of environment.”

Private company data represents what Mensch calls “the last untapped data reserves” for improving AI performance. Manufacturing firms, logistics companies, and financial institutions hold decades of operational data that have never been exposed to AI systems.

The embedded team model generates revenue through service fees while improving Mistral’s core technology. Knowledge gained from working with enterprise data enhances models available to all customers.

Shipping giant CMA CGM provides an early success story. Mistral’s embedded team automated the complex process of tracking containers as they leave cargo ships.

The AI solution connected multiple software systems and reduced costs by 80 percent, according to Mensch. Such results require deep integration that no surface-level chatbot deployments can achieve.

The company secured €1.7 billion in funding last month, led by Dutch semiconductor equipment maker ASML. Both firms will work together under the new model, with Mistral teams helping ASML develop AI solutions for its chip manufacturing processes.

“This investment brings together two technology leaders operating in the same value chain,” Mensch said in a press release. “We aim to help ASML and its partners meet current and future technical challenges through AI.”

Mistral plans to expand this model globally. The company recently announced offices in Montreal and partnerships across Europe, Asia, and Africa.

Organizations may need to restructure to benefit fully from AI integration. Mensch suggests companies could operate with fewer middle managers as AI improves information flow between departments.