More Than Half of Enterprise Software Could Switch to AI

Arthur Mensch, CEO of Mistral AI, told CNBC this week that more than 50% of enterprises’ current software could be replaced by AI. The statement landed in the middle of a brutal sell-off in software stocks.

Even the iShares Expanded Tech-Software Sector ETF went down more than 20% this year, with names like Salesforce, ServiceNow, SAP, and Oracle losing between 5% and 20% of their value in a single week.

The panic also has a name now, the 'SAASpocalypse'. Wall Street is pricing in the possibility that AI agents and autonomous systems will make large portions of traditional SaaS obsolete. But Mensch’s framing is more precise than the market panic suggests.

He did not say AI will eliminate software. He said enterprises are replatforming, getting rid of things they bought 20 years ago that have become expensive and inefficient, and rebuilding those workflows on AI infrastructure.

The distinction matters. This is not about AI making software irrelevant. It is about AI changing what software is, how it is built, and who controls it. And many of the people selling traditional SaaS are not positioned to survive that shift.

Mensch’s argument is backed by what Mistral is seeing with its own enterprise customers. He described companies creating fully custom applications in a couple of days to run procurement workflows or supply chain operations, tasks that five years ago would have required buying vertical SaaS and spending months on integration.

Generative AI, combined with agentic systems that can autonomously execute multi-step workflows, is genuinely compressing development cycles that used to take quarters into timelines measured in days.

The strategic implication is great. If enterprises can build custom software at the speed Mensch describes, the value proposition of traditional SaaS, which is fundamentally about offering pre-built, standardized solutions faster than an internal team could develop them, collapses.

And if the software can be built faster internally, it can also be tailored more precisely to the business’s actual workflows rather than forcing the business to adapt to the software’s assumptions.

Mensch was careful to note that systems of record are not going to change. Bipul Sinha, CEO of Rubrik, agreed with this view in a separate CNBC interview, predicting that workflow software will be significantly disrupted by AI, but data infrastructure software that enables AI will see positive momentum. The distinction cuts cleanly through the SaaSpocalypse panic. Not all software is equally at risk.

Legacy platforms that sit on top of proprietary, high-quality data like Oracle’s enterprise databases or SAP’s ERP systems have structural moats. The data itself has value, and switching costs remain high. But workflow SaaS built on top of open or semi-open data sources has no such protection.

If the primary value is orchestrating tasks rather than owning data, and if AI can now handle that orchestration autonomously, the software layer becomes redundant. The value moves to the AI system running the work, not the interface that used to mediate it.

Who Will Control The Replatforming

Forrester Research predicts that 30% of enterprise app vendors will launch their own Model Context Protocol (MCP) servers in 2026, creating an open ecosystem where businesses are not locked into a single AI provider.

This is the real battleground. Vendors that embed MCP servers and allow external AI agents to collaborate with their platforms will enable cross-functional, agentic workflows that span multiple systems.

Vendors that do not will be bypassed. Enterprises will simply route around them using custom AI agents that pull data from APIs and execute workflows independently.

“The replatforming is a big opportunity for us, because we now have more than 100 enterprise customers coming to us also with that will of maybe changing and replatforming their IT system, so maybe getting rid of certain things that they bought 20 years ago, and that is starting to be a bit expensive,” Mensch said. “They see AI as a way to replatform the thing so that it becomes more efficient and less costly.”

Forrester also predicts that half of enterprise ERP vendors will launch autonomous governance modules in 2026, combining explainable AI, automated audit trails, and real-time compliance monitoring. The vendors who move first on governance will capture competitive advantage through compliance-ready platforms.

The rest will face customer defection. Governance is not a feature, it is the foundation that makes autonomous AI deployable at enterprise scale. Vendors who treat it as an afterthought will lose to those who architected it from the beginning.

While software stocks have been hammered, industrial companies are thriving. Gilead and John Deere hit new highs this week even as technology stocks were sold off again. Futuriom’s AI Enterprise Index, which tracks more than 100 enterprise AI use cases, shows that the two leading categories are process optimization (17%) and data-driven insights (9%).

The value is being created, it is just being captured by enterprises using AI to improve their own operations rather than by the software vendors selling them tools.

The market is correct to be skeptical of traditional SaaS. But the skepticism is being applied too broadly. Nvidia CEO Jensen Huang called the software sell-off “illogical,” and he is partially right. Companies with strong data moats, governance infrastructure, and AI-native architectures are not at risk.

Companies selling workflow software built on generic data sources with no differentiation beyond the interface are in serious trouble. The market has not yet cleanly separated the two.

What This Means for Leaders

For CIOs evaluating their software portfolios, the replatforming question is not abstract. It is operational. The first decision is whether to continue renewing workflow SaaS that could be replaced by custom AI agents built internally.

Andreessen Horowitz found that over 90% of surveyed enterprises are testing third-party AI apps for customer support, and one public fintech stopped an internal build to buy a third-party tool instead. That suggests the build-versus-buy calculus is shifting unevenly.

The second decision is which vendors to bet on for the next decade. CIOs should interrogate their business app vendors on their MCP strategy, their autonomous governance roadmap, and whether they are building AI-native architectures or retrofitting AI onto legacy systems.

Vendors who cannot answer those questions coherently are high-risk holds. Nearly 30% of pilot AI projects get dropped, and skill gaps still slow rollouts. The reasons are predictable. Enterprises underestimate the data quality and governance work required before AI can run reliably.

They assume existing staff can handle AI operations without reskilling, which creates execution gaps. And they deploy AI agents without clear ROI metrics, making it impossible to justify continued investment when the pilot phase ends. The vendors who survive will be the ones who make AI deployment easier, not harder.

Mensch is right. More than 50% of enterprise software could be replaced by AI. But replacement is the wrong framing. The better framing is replatforming. Enterprises are not abandoning software, they are rebuilding it on AI infrastructure. The vendors who control that infrastructure, provide the governance frameworks, and enable cross-platform agentic workflows will capture the value.