Gartner Projects $234 Billion in SaaS Spending Under Threat from AI Agents by 2030

The Number That Reshapes the CIO's Perspective

Gartner released a report on July 1 that quantifies, for the first time with a specific figure, the risk that autonomous AI agents pose to the enterprise software market. The consultancy estimates that by 2030, $234 billion in annual spending on SaaS applications will be exposed to what it terms "agentic arbitrage," approximately 20% of the total. This projection accompanies comments from George Brocklehurst, Managing Vice President at Gartner, who authored the document.

The definition is technical and precise. According to Brocklehurst, agentic arbitrage occurs when an AI agent performs tasks across multiple SaaS platforms simultaneously, delivering results to the user without the need to open each interface. "Agentic systems deliver outcomes directly and make software invisible," Brocklehurst said during the presentation. "This breaks the link between user growth and revenue growth for many vendors."

Salesforce, ServiceNow, and Workday in the Crosshairs

The report targets the per-seat pricing model that underpins the three largest American corporate SaaS vendors. Salesforce closed 2025 with $37.9 billion in revenue supported by 150,000 customers paying for user licenses. ServiceNow surpassed $10 billion following the same model. Workday exceeds $8 billion. If Gartner’s standard materializes, each company loses the metric that investors have relied on for 15 years to justify their high multiples.

So far, the vendors’ reaction has been to anticipate the agent itself. Salesforce launched Agentforce in October 2024 and charges per conversation, $2 per interaction. ServiceNow has embedded Now Assist Agents within its platform. Workday has partnered with Anthropic for deploying agents in HR processes. However, the commercial model still relies on paid seats. Gartner argues that this transition will only be successful if the vendor can capture the incremental value of ROI that automation releases, something Brocklehurst describes as "less apocalypse and more metamorphosis."

The Perspective Outside Silicon Valley

In Germany, SAP faces the same pressure from the opposite side. As the dominant ERP provider in Europe, SAP argues that its proximity to data and workflows makes Joule more defensible against replacement by third-party agents than Salesforce. Christian Klein, CEO of SAP, told analysts in May that the company is not following the per-seat model in selling agentic functionalities and will charge by outcome starting in 2027. In Bangalore, the effect is the opposite: Infosys and TCS gain integration pipeline because each agent needs to be connected to the customer's record system. Margins drop while volumes rise.

In Latin America, mid-sized companies are already facing the dilemma of whether to continue paying full licenses for HRIS and CRM systems or outsource the flow to an agent that queries both platforms via API. TOTVS, Brazil's leading ERP provider with 40,000 clients, addressed the topic in its May earnings call, stating that TOTVS Agent, launched in November 2025, is the piece that ensures retention against external agent attacks. It’s the same thesis as SAP's, adapted for regional scale.

What Weakens Gartner's Projection

The contrary view is well-known. Analysts from Morgan Stanley and Bernstein argue that Gartner's projections regarding software replacement have historically overestimated the timing. The prediction that on-premises ERP would become extinct by 2015 was off by a decade. The adoption curve for agentic functionalities depends on structured data that many companies still do not possess, and the cost per agent operation today, between $0.50 and $4.00 per complex task, still makes the math favorable only in high-frequency workflows.

The point that the projection does not lose: even if Gartner is off by 40% in the figure, the per-seat commercial model will lose traction in the 2027 and 2028 renewal cycle. Those selling software to CIOs need to start renegotiating contracts based on outcomes now, or they will arrive in 2029 trying to salvage revenue that has already leaked to the agent that no one approved formally. For the buyer, the reading is symmetrical: agreeing to renew a three-year contract in a per-seat model between 2026 and 2028 is signing off on a metric that Gartner itself has just indicated is becoming obsolete.