AI Is Repricing the Entire Category

$2 trillion in market cap, gone

The numbers are not subtle. $2 trillion in SaaS market capitalization has evaporated. The iShares Expanded Tech-Software ETF is down over 21% year-to-date and roughly 30% from its September 2025 peak. SaaStr has called it "The SaaS Rout of 2026" - and the label is deserved. For the first time in the history of enterprise software, the sector trades at a discount to the S&P 500.

Revenue multiples have compressed from 6.7x in 2024 to 5.9x in 2025, and the data suggests further compression in 2026. This is not a cyclical downturn. It is a structural repricing driven by a single question: what happens to SaaS when AI can do the work that the software was designed to organize?

The market is not punishing all software equally. Companies with deep data moats, platform architectures, and AI-native products are holding or gaining value. The companies losing the most are point solutions - single-feature products that automate one workflow and charge per seat for the privilege. AI is making their core value proposition obsolete faster than they can pivot.

The token cost trap: when AI costs more than the employee it replaced

Here is the irony nobody saw coming. Companies spent 2024 and 2025 laying off workers and replacing them with AI tools. Now they are discovering that the token bills can exceed what they were paying those employees in salary. Nvidia VP of Applied Deep Learning Bryan Catanzaro said it plainly in April 2026: "For my team, the cost of compute is far beyond the costs of the employees."

He is not alone. Uber burned through its entire 2026 AI coding tools budget in four months after Claude Code adoption surged across roughly 5,000 engineers. Usage jumped from 32% of engineers in February to 84% classified as agentic users by March. COO Andrew Macdonald publicly questioned whether the spend is producing results: "That link [between token spend and consumer innovation] is not there yet." Microsoft reportedly began canceling most of its direct Claude Code licenses and pushing engineers toward GitHub Copilot CLI to control costs.

The phenomenon now has a name: tokenmaxxing. Engineers and knowledge workers maximize their AI token consumption, sometimes racking up $150,000 per month individually, either because they believe it signals productivity or because the tools have become embedded in daily workflows to a degree that drives up costs regardless of intent. Uber even ran internal leaderboards ranking teams by how much they used the tool, which gave engineers every reason to use Claude aggressively and no reason to hold back.

The math behind this is counterintuitive. Token prices have fallen 280x over the past two years. But total enterprise AI spend has risen 320% in the same period. Why? Because agentic AI workflows consume 5 to 30 times more tokens per task than a standard chatbot interaction. Every agentic task triggers 10 to 20 LLM calls, RAG architectures inflate context windows 3 to 5x, and always-on monitoring agents consume compute around the clock. Gartner predicts that inference on a one-trillion-parameter model will cost 90% less by 2030, but cheaper tokens will not translate to cheaper enterprise AI because consumption is scaling faster than unit costs are falling.

AI-native retention: a different kind of churn problem

AI-native SaaS products have a retention profile that looks nothing like traditional B2B software. ChartMogul's data shows AI-native SaaS at 40% gross revenue retention - comparable to B2C, not B2B. The traditional B2B benchmark sits at 75% GRR. That gap is enormous and it reflects a fundamental difference in how customers evaluate AI tools versus traditional software.

There is a silver lining. AI-native GRR is improving rapidly - it jumped from 27% to 40% in just nine months. The products are getting better, the use cases are maturing, and buyers are developing more realistic expectations about what AI can and cannot do. But 40% GRR still means you lose 60% of your revenue every year. At that rate, you need extraordinary new logo acquisition just to stay flat.

Net revenue retention for AI-native SaaS sits at 48%, which is well below the 108% median for traditional B2B. This means AI-native companies are not just losing customers - they are failing to expand the customers they keep. The expansion motions that work for traditional SaaS (seat-based growth, feature upsells, usage expansion) have not yet been proven at scale for AI-native products.

The seat-replacement thesis: AI agents versus human users

The most disruptive thesis in enterprise software right now is simple: AI agents will replace human users, and per-seat pricing will collapse. If an AI agent can do the work of a customer service rep, a data analyst, or a junior developer, why would a company pay per-seat for the software those humans used?

The incumbents are fighting back by embracing the thesis rather than denying it. Salesforce launched Agentforce - AI agents that live inside the Salesforce platform and perform tasks that previously required human users. Microsoft Copilot is embedded across the entire Office and Azure stack. ServiceNow is building AI agents for IT service management. The message from the incumbents is clear: the agents will run on their platforms, and they will charge for agent consumption rather than human seats.

This creates an existential question for mid-market SaaS companies built on per-seat economics. If Salesforce and Microsoft are giving away AI agents as part of their platform, what happens to the standalone tools that charge $50/seat/month for workflow automation? The answer is already visible in the valuation data: point solutions are being repriced to reflect a future where their primary user might be an AI, not a human.

Vertical SaaS: where AI creates value instead of destroying it

While horizontal SaaS faces existential pressure from AI commodification, vertical SaaS is experiencing the opposite effect. Specialized AI for healthcare, legal, construction, and logistics is creating new value because these industries have domain-specific data, regulatory requirements, and workflow complexity that general-purpose AI cannot address.

A healthcare SaaS company with HIPAA-compliant training data and clinical workflow integration has a moat that ChatGPT cannot cross. A legal AI tool trained on case law and integrated with court filing systems has domain depth that no horizontal tool will replicate. The specialization is the defense.

The research shows vertical SaaS companies with AI capabilities are seeing faster growth and better retention than their horizontal counterparts. They benefit from higher switching costs (industry-specific integrations), deeper data moats (proprietary training data), and clearer ROI narratives (measurable outcomes in specific workflows).

AI in customer success: the multiplier effect

91% of companies say AI will have moderate to significant impact on their customer success operations (Gainsight 2025 Pulse Survey). More than 50% have already integrated AI into their core CS workflows. This is not a future state - it is happening now, and the results are measurable.

CS teams using AI-augmented tooling are seeing 3x expansion revenue growth compared to teams without AI assistance. Onboarding times have dropped by 40%. The mechanism is straightforward: AI handles the pattern recognition (which accounts are at risk, which are ready to expand, what signals matter), and humans handle the relationship work (executive conversations, strategic reviews, complex negotiations).

Digital CS is growing at 15% annually as a category. Online communities grew from 42% to 73% adoption among SaaS companies. The combination of AI-powered health scoring, automated playbooks, and digital community engagement is creating a CS model that scales without proportionally scaling headcount.

What survives the reshuffling

Not all SaaS dies. But the categories that survive look very different from the categories that thrived in the 2020-2023 era. Three archetypes are emerging as durable in the AI-native future.

The first is platforms, not point solutions. Companies that serve as the system of record - the place where data lives, workflows connect, and integrations are managed - are gaining value as AI makes individual features less defensible. Salesforce is a platform. A standalone email tracking tool is a point solution. The platform survives because it is the substrate on which AI agents run.

The common thread: defensibility comes from being where the work happens, not from doing any single piece of the work. AI can do the work. AI cannot be the system of record, the integration layer, or the compliance framework. The companies that own those layers will be the ones that survive the reshuffling.