Today, at the India AI Impact Summit in New Delhi, Mistral AI CEO Arthur Mensch told CNBC something that should be on every enterprise leader's radar: more than 50% of the software currently running inside enterprises could be replaced by AI.
Not augmented. Not enhanced. Replaced.
His exact framing matters. Mensch described a "replatforming" — enterprises coming to Mistral with the explicit intention of getting rid of software they bought 20 years ago. Not because the software doesn't work. Because AI can now do the same job faster, cheaper, and without the recurring licence fees.
This wasn't a keynote abstraction. Mensch said Mistral now has more than 100 enterprise customers actively pursuing this transition — building custom AI applications in days that previously required vertical SaaS platforms to accomplish the same workflows.
He's not the only one saying this. But the timing — at the largest AI summit in history, with 110 countries, every major AI CEO, and $100 billion in prospective investment discussions happening simultaneously — signals that this is no longer an emerging trend. It's the central narrative of enterprise technology in 2026.
The $285 Billion Wake-Up Call
To understand why Mensch's comments matter today, you have to understand what happened two weeks ago.
On February 3, Anthropic released industry-specific plug-ins for Claude Cowork — its AI-powered workplace productivity suite. The plug-ins targeted legal research, financial analysis, marketing, sales, and customer support — functions that millions of enterprises currently pay for through traditional SaaS subscriptions.
Wall Street's reaction was immediate and severe. In a single trading session, approximately $285 billion in market capitalisation was wiped out across software, financial services, and asset management stocks. Thomson Reuters fell nearly 18% — its worst single-day drop on record. RELX (LexisNexis) dropped 14.4%. Intuit, PayPal, Equifax, Salesforce, and Atlassian all saw double-digit declines. The Goldman Sachs basket of US software stocks fell 6% — its steepest daily decline since the tariff-driven selloff in April 2025. The S&P 500 Software & Services Index extended its losses to eight consecutive sessions and is now down approximately 20% year-to-date.
The sell-off spread globally. Indian IT services firms Tata Consultancy Services and Infosys saw some of their steepest single-day declines in months. Advertising holding companies Publicis, WPP, and Omnicom each fell between 9% and 12%. Private equity firms Ares Management and KKR — which had invested heavily in software portfolios — each dropped 10%.
Analysts called it the "SaaSpocalypse." But the more accurate term is structural repricing. Wall Street wasn't panicking about one product launch. It was repricing the future of the entire SaaS business model.
What's Actually Happening — and What's Overblown
Let's separate signal from noise.
The signal is real. AI is now capable of automating workflows that previously required dedicated software platforms staffed by dedicated teams. Contract review, compliance triage, financial analysis, CRM management, marketing asset generation, data analytics — tasks that enterprises have been paying per-seat SaaS licences for are increasingly executable by AI agents at a fraction of the cost.
Gartner's analysis of the Cowork launch was measured but significant. They wrote that Claude Cowork and its plug-ins are "potential disrupters for task-level knowledge work" and that the release "exposes how much day-to-day knowledge work remains manual, making it ripe for automation."
The noise is the Armageddon narrative. Wedbush Securities pushed back, noting that enterprises "won't completely overhaul tens of billions of dollars of prior software infrastructure investments" to migrate to AI platforms. They have a point. Large enterprises took decades to build their current software stacks. Trillions of data points are ingrained in their existing infrastructure. Ripping everything out and replacing it with AI agents isn't realistic — and it's not what smart enterprises are doing.
The reality sits between the extremes. Enterprise software isn't dying tomorrow. But it is being restructured from the bottom up. The question for enterprise leaders isn't whether this transition happens — it's how to navigate it without creating new dependencies that are worse than the ones you're leaving behind.
The Replatforming Playbook
Here's what the evidence actually points to — and the strategic framework we recommend to our clients.
1. The Workflow Layer Is Being Absorbed by AI
The most vulnerable software categories are those that primarily organise, route, and present information. CRM dashboards. Legal research databases. Financial data terminals. Marketing asset management platforms. Document review tools.
These products are valuable not because of proprietary technology, but because they structure workflows around data that enterprises already own. When AI agents can access, process, and act on that same data — directly connecting to enterprise systems through protocols like MCP — the workflow layer becomes redundant.
Alphabet's numbers illustrate the scale. Gemini Enterprise has sold more than 8 million paid seats in roughly four months. Gemini processes more than 10 billion tokens per minute via direct APIs. Alphabet plans $175 billion to $185 billion in capital expenditure for 2026 — and Sundar Pichai, who met PM Modi in New Delhi today, told analysts the company is still "supply constrained."
The demand for AI-native workflow tools is not speculative. It's infrastructure-constrained.
2. Licensing Models Are Already Changing
Salesforce's response to the Cowork threat is instructive. Rather than defending the traditional per-seat SaaS model, they've introduced what they call an Agentic Enterprise License Agreement (AELA) — an all-you-can-eat flat fee that bundles their Agentforce platform with their broader cloud suite.
Salesforce's President and Chief Revenue Officer Miquel Milano explained the rationale directly: "AELA is for customers that have already experimented. They're ready to scale. They want to go all in so we agree on a flat fee, and then it's a shared risk."
Translation: the per-seat model is under enough pressure that Salesforce — the company that defined SaaS — is willing to take a loss on current licensing to lock in long-term platform dependency. They're playing for the renewal, when the customer is fully committed.
For enterprise buyers, the strategic implication is clear: every SaaS vendor will soon offer similar bundled AI licensing. Some of these deals will be genuinely valuable. Others will be vendor lock-in disguised as innovation. The enterprises that can distinguish between the two will save millions. The ones that can't will simply trade one form of dependency for another.
3. The Infrastructure Gap Is Real — And It's the Opportunity
Here's the counterpoint that most coverage of the SaaSpocalypse misses.
MIT's Project NANDA found that 95% of organisations are seeing zero measurable return from their generative AI investments. A separate study found that two-thirds of enterprise IT leaders say their AI environments are too complex to manage. Infrastructure, power, and operational foundations required to run AI at enterprise scale simply aren't there for most organisations.
This is the gap between what AI can theoretically do and what most enterprises can actually execute. It's also where the real opportunity lies.
The enterprises that will capture the replatforming advantage aren't the ones that rush to replace their SaaS stack with AI agents. They're the ones that build the infrastructure to evaluate, orchestrate, and govern AI-driven workflows alongside their existing systems — transitioning incrementally, measuring outcomes at every step, and maintaining optionality throughout.
4. The Platform Layer — Not the Model Layer — Determines Who Wins
Mistral's Mensch, Anthropic's Cowork, Salesforce's AELA, Google's Gemini Enterprise, OpenAI's Frontier — they're all competing for the same position: becoming the platform through which enterprises operate.
But here's what enterprise leaders need to understand: adopting any single AI platform as your operating layer creates the same dependency risk as adopting any single SaaS vendor. The vendors change. The lock-in doesn't.
The enterprises that win the replatforming are the ones that control their own orchestration layer — the infrastructure that connects AI models to enterprise data, governs agent behaviour, manages permissions, and allows swapping providers without rebuilding workflows.
Model Context Protocol (MCP) — now under the Linux Foundation's Agentic AI Foundation with support from Anthropic, OpenAI, Microsoft, and Google — provides the standard connectivity layer. But connectivity alone isn't orchestration. Orchestration requires intentional architecture: agent management, governance frameworks, monitoring, and the ability to run multiple AI providers simultaneously.
What This Means for Gulf Enterprises
The replatforming wave is global, but it has specific implications for enterprises operating in the Gulf region.
SaaS contracts are leverage points. Every enterprise in the region running legacy SaaS platforms is sitting on a renegotiation opportunity. As vendors scramble to retain customers with bundled AI licensing, enterprises with clear AI strategies and optionality can negotiate dramatically better terms — or transition to AI-native alternatives for specific workflows.
Sovereign data requirements accelerate the shift. Gulf enterprises with data residency requirements have often been constrained to specific SaaS vendors with regional data centres. AI-native architectures — particularly those built on open-source models deployable on-premises or on sovereign cloud infrastructure — offer more flexibility, not less, for meeting compliance requirements while modernising workflows.
The services layer needs restructuring. Many Gulf enterprises rely on systems integrators and managed service providers to operate their software stacks. As AI absorbs workflow-layer functions, the value of these relationships shifts from operating software to managing AI infrastructure, governance, and orchestration. Enterprises should be evaluating their service partners' AI capabilities now — not when contracts come up for renewal.
Multilingual capability matters more in a replatformed stack. When software workflows move to AI agents, the agents need to operate effectively in Arabic alongside English and other languages. Most Western SaaS platforms have limited Arabic support. AI-native architectures built with multilingual capability from the ground up deliver better outcomes for Gulf enterprises processing documents, serving customers, and operating across markets.
How to Think About This Strategically
The replatforming is real. But it's not a cliff — it's a transition. And enterprises that approach it strategically will extract significantly more value than those that react to vendor pressure or market narratives.
Audit your software stack against AI capability. For each SaaS platform you're paying for, ask: what specific workflow does this enable, and can an AI agent perform that workflow at equivalent or better quality? Start with the highest-cost, lowest-complexity tools. That's where the first wins are.
Build your orchestration layer before you build your agents. The enterprises getting burned on AI aren't the ones deploying agents. They're the ones deploying agents without the infrastructure to manage, monitor, and govern them. Orchestration — model management, agent identity, permissions, monitoring, audit logging — comes first. Agents come second.
Negotiate from strength, not urgency. Every SaaS vendor will approach you with bundled AI licensing in 2026. Before you sign, understand what you're getting and what you're giving up. Flat-fee AI licensing sounds attractive until you realise it locks you into a single provider's ecosystem for the next three to five years.
Measure everything. The 95% of enterprises seeing zero ROI from AI aren't failing because the technology doesn't work. They're failing because they can't measure whether it's working. Define outcome metrics before deployment. Track them continuously. If something isn't delivering value, stop doing it. If it is, scale it.
The Bigger Picture
The India AI Impact Summit — now extended to February 21 due to overwhelming demand, with over 300,000 registrations and 110 countries participating — is the backdrop for a fundamental restructuring of enterprise technology.
Two weeks ago, Claude Cowork triggered a $285 billion repricing of the software industry. Today, the CEO of Mistral told the world's largest AI gathering that half of enterprise software can be replaced. Tomorrow, PM Modi will deliver his inaugural address to 20 heads of state and 40 CEOs, framing India's vision for AI as the driver of a $400 billion IT industry by 2030. On Thursday, Sundar Pichai — who met Modi today and is planning $185 billion in AI infrastructure investment — delivers the keynote.
The message from New Delhi this week is unmistakable: the enterprise software stack that powered the last two decades of business is being restructured. The models are ready. The platforms are competing. The investment is flowing. The governance frameworks are being built.
The enterprises that move deliberately — building infrastructure, maintaining optionality, measuring outcomes, and governing the transition — will define the next era. The ones that wait will spend the next five years catching up to the ones that didn't.
The replatforming has begun. The only question is whether you're navigating it or being navigated by it.