That is a statement likely to send a shiver through the major software publishers, already strained by the lukewarm reception from financial analysts. In an interview with CNBC, Arthur Mensch, CEO of Mistral AI, asserts that more than 50% of today’s enterprise software could eventually be replaced by applications built on generative AI.
In the crosshairs? Productivity tools, workflow systems, and lightweight CRMs—essentially the core business of Salesforce, Workday, or ServiceNow.
The Promise of Replatforming
To explain it, Arthur Mensch invokes the concept of “replatforming”—a deep overhaul of enterprise IT architectures that would abandon standardized SaaS subscriptions in favor of bespoke applications built directly on AI model APIs. The commercial argument is compelling: lower license costs, closer alignment with internal processes, development times compressed from months to days.
The movement already has its precursors. Klarna, the Swedish fintech, has turned away from certain Salesforce and Workday components to build its own AI-powered stack. A textbook case that Arthur Mensch willingly cites to illustrate the tangible feasibility of this bold technological leap.
Not All Will Be Carried by the Wave
The head of Mistral takes care to nuance the claim. The data infrastructure—storage, backup, security, data platforms—does not face a threat. It will emerge even stronger from the transition, since it is what feeds the models.
Bipul Sinha, CEO of Rubrik, shares the assessment: the “front office software” will be reshaped, while the “back office data” will consolidate as a critical and indispensable layer.
But this shift requires serious prerequisites: clean and unified data, a modern cloud infrastructure, and teams capable of governing these new agents. We are far from the comfort of turnkey SaaS.
An Opportunity Tailored for Mistral
Of course, the statements from the head of Mistral AI serve his business model. He already claims more than a hundred large enterprise clients seeking modernization. His platform—open and proprietary models, personalized assistants, enterprise search—positions itself precisely as the toolkit for this replatforming.
Yet the forecast remains contested. Skeptics point to the considerable inertia of existing information systems, regulatory constraints, and the many dashed AI production projects. For many observers, AI will first integrate as an augmentation layer above current tools rather than as a bulldozer.
But the message to traditional software publishers is unambiguous: transform into AI platforms, or risk becoming the software that others replace.
