Software Stocks Rally 21% in May: AI Agents Are Sorting Winners From Losers

Three months ago, Wall Street held something close to a funeral for enterprise software. In a 48-hour span in early February 2026, traders coined the term "SaaSpocalypse" after roughly $285 billion was wiped from software stock valuations — a shock widely attributed to fears that Anthropic's Claude Cowork, an agent platform capable of handling contract review, compliance checks, and sales preparation, could compress demand for seat-based software licenses. Analysts queued up to write obituaries for the per-seat model.

Three months later, those same analysts are quietly buying it back. The iShares Expanded Tech-Software ETF (IGV) rose 21% in May, its best monthly performance since October 2001, and trades roughly 35% above its April low. Whether smart money is bottom-fishing a genuine recovery or chasing momentum after missing the cycle, the honest answer is that neither description cleanly fits. What looks like one rebound is, on closer reading, a market doing something more specific: re-rating the sector name by name on a single question — is this company on the path autonomous agents have to travel, or is it priced for what agents replace?

Relief Rally With Force Behind It

Begin with what is undeniable. The February selloff overshot. No recession arrived, no central-bank tightening materialized, no broad earnings disappointment followed. The trigger was a product announcement, and the day's biggest losers were companies whose underlying results were unchanged before and after. Forced de-risking and stop-outs did the rest. Once the panic faded, the names sold hardest bounced hardest — a familiar market mechanic that requires no AI thesis to explain.

Options-flow data also showed heavy call demand in IGV late in May, trading more than five times the 30-day average volume, with calls outpacing puts four-to-one. Sentiment rallies have short half-lives; the question is what carries the tape from here.

Snowflake Earnings Supplied the New Story

The fuel under May's move was Snowflake. After the close on May 27, the company reported first-quarter fiscal-2027 revenue of $1.39 billion, up 34% year over year, with product revenue of $1.33 billion, both beating consensus. It also raised its full-year product-revenue guidance to $5.84 billion from a prior $5.66 billion. The move that ignited the chart came from two additional announcements: a five-year, $6 billion commitment to AWS for Graviton compute and AI infrastructure, and the acquisition of Natoma, a startup building enterprise governance for the Model Context Protocol — the emerging standard for how AI agents access enterprise tools and data. Snowflake surged roughly 37% after the report, with Constellation Research analysts applying the framing that did the most work: "selling shovels."

The metaphor matters more than the bounce. Snowflake's results offered a concrete answer to the question that had hung over the sector since February: in an agent-driven enterprise, is software an obstacle that AI routes around, or infrastructure that AI consumes? Snowflake's pitch is that if a company sits on the path agents must travel — storing the data, governing the tool calls — autonomous agents are not a threat. They are the most demanding new customer.

The story is potent and, at this point, more thesis than proof. Natoma's MCP gateway had been deployed at only a handful of enterprises as of the acquisition announcement, and the bulk of agentic workloads inside Fortune 500 IT departments today remain pilots and proofs-of-concept rather than production systems running mission-critical processes. Natoma's total acquisition price was approximately $110 million in Snowflake stock and cash — meaningful for a startup with a handful of customers, but a small anchor for a $5.84 billion revenue business. Other names, including Datadog, MongoDB, and Okta, also contributed to May's rally, but Snowflake's move was unusually large precisely because investors attached it to the AI-infrastructure narrative.

Bank of America Split ServiceNow and Salesforce in Opposite Directions

The most useful analytical framework for reading the rebound arrived from Bank of America. On May 18, analyst Tal Liani reinstated coverage of two software giants in opposite directions: ServiceNow at Buy with a $130 price target, arguing the workflow platform could become a central layer for autonomous AI operations across IT, employee, and customer processes; and Salesforce at Underperform with a $160 price target, describing what Liani called an "AI-driven structural reset." Same sector, same analyst, same day. The contrast was the message.

The logic is straightforward. ServiceNow sits inside the approval, routing, and compliance infrastructure of large enterprises. The more autonomous agents proliferate, the more an organization needs a system of record for who is allowed to do what, with which data, and who is accountable when something goes wrong. Liani argued that agentic AI deployments "would elevate the need for orchestration, permissions, approvals, policy enforcement and auditability, aligning directly with ServiceNow's core capabilities." ServiceNow's Q1 2026 results — $3.77 billion in revenue, 22% year-over-year growth, 630 customers above $5 million in annual contract value — supplied numbers consistent with that thesis. By the final week of May, the stock had recovered roughly 33% from its 2026 lows.

Salesforce's problem runs closer to inverse. Its core revenue is denominated in CRM seats — dollars per salesperson per month. When an agent handles part of a sales rep's job, the natural buyer reflex is not to add an agent seat but to reduce a human one. Liani's three pressure points were muted net-new customer additions, limited upsell potential, and an underwhelming monetization path for Agentforce. Then came the earnings report.

Salesforce Q1 FY2027 Results Complicated the Bear Case

On the same evening as Snowflake's report, Salesforce disclosed its own first-quarter fiscal-2027 results. Revenue came in at $11.1 billion, beating the $11.05 billion consensus, with adjusted earnings per share of $3.88 against a $3.12 expectation — a beat that crushed the bar by nearly 24%. Agentforce annual recurring revenue reached $1.2 billion, up 205% year over year, crossing the billion-dollar threshold for the first time. The company delivered 3.8 billion Agentic Work Units — the consumption-based metric it uses to track agent task completions rather than software access — and counted more than 29,000 Agentforce deals closed since launch.

Those are not numbers that support a straightforward bear case. Salesforce is simultaneously running three pricing models — per-conversation, flex-credit, and per-user — which is itself a signal that the company has not yet settled on the answer. Running three models in parallel suggests a company testing rather than executing. The structural question Liani identified — whether Agentforce's consumption revenue will scale fast enough to offset any compression in the legacy seat base — remains unanswered by a single quarter of acceleration. Full-year revenue guidance of $45.9 billion to $46.2 billion, up 11% year over year, slightly undershot Wall Street's expectations, and the stock was little changed in after-hours trading despite the earnings beat. The market was not celebrating; it was still waiting.

AI Agents Reshape the Seat-Based Pricing Model

Pulled back, the rebound looks less like a binary call on AI than a repricing of who gets paid and who gets disintermediated when agents do enterprise work. The traditional sales conversation pitted one software product against another. The new one increasingly pits a software product against a human salary: an agent at roughly $30,000 a year in consumption cost against a junior analyst at $90,000. The yardstick shifted from feature checklist to labor substitution, which both raises the ceiling — software can now claim a slice of the payroll budget, not just the IT line — and tightens the floor, because finance teams cut what they cannot tie to return on investment first.

Gartner has projected that 35% of point-product SaaS tools will be replaced by AI agents by 2030. Bain & Company and Deloitte published reports in late 2025 predicting that AI agents would cannibalize per-seat revenue in task-level productivity tools within 24 months. Seat-based pricing adoption has dropped from 21% of vendors to 15% in the span of roughly 12 months.

That reframing produces a usable map, though not a deterministic one. The market appears to be rewarding companies closer to agent infrastructure — data platforms like Snowflake, observability players like Datadog, and identity, workflow, and security platforms like ServiceNow and CrowdStrike — whose consumption- or workflow-anchored pricing scales with the volume of agent activity, not the number of human seats it replaces. It is applying more skepticism to seat-denominated systems where AI can reduce human-user counts, including Salesforce's most exposed surfaces, Workday, and parts of the collaboration and content-creation stack. The uncomfortable middle belongs to companies not facing extinction but watching legacy lines erode faster than new ones scale — precisely where Salesforce sits today.

What the SaaS Rebound Is, and What It Is Not

Underneath the IGV print there is a quieter signal worth weighing. Even with index-level volatility calmer, single-stock moves through the rebound remained violent — Snowflake's 37% day and Salesforce's muted response to a large earnings beat both happened the same evening. That pattern is consistent with a market sorting winners and losers rather than lifting the whole sector evenly. Such regimes do not last indefinitely. When a shock arrives that pushes everything in the same direction at once, correlations snap back, and what felt like a stable rally can give back gains quickly. For an investor picking software names, the cost of being wrong on a single ticker is far higher than the index suggests.

As of Monday, the software basket is back green for 2026 at the index level. That headline conceals a wide dispersion: Snowflake is up roughly 50% on the week of earnings; HubSpot is up 10% in a single session but still down nearly 46% on the year; Salesforce closed May down roughly 30% for 2026 despite its earnings beat; and the median software company is still climbing out of a deep hole. The infra names carry the index; the cap-weighted basket reads green; most individual names do not.

The cleanest read of May is that the rebound is partly mechanical mean reversion, partly a genuine fundamental re-rating, and partly a market doing the harder work of segmenting an industry that used to be priced as a single basket. The names rallying hardest are those whose business models align with what agents do — processing data, governing tool calls, routing approvals. The laggards are those whose models price what agents replace — human seats completing tasks. The risk for bulls is not that the infrastructure thesis is wrong but that it is already priced in. The risk for bears is that this is not 2001 — there is a real customer underneath the new spend, and the picks-and-shovels analogy, for a few names, looks like it might actually fit.

The funeral in February was premature. The triumphalism in May is also premature. What is happening is a market finally pricing software companies one by one on the question that ought to have driven their valuations all along: when autonomous agents become a real part of how enterprises work, are you on the path they have to travel — or are you what they are routing around?

Frequently Asked Questions

Why did software stocks recover 21% in May 2026?

The iShares Expanded Tech-Software ETF's 21% May gain — its best month since October 2001 — was driven primarily by Snowflake's May 27 earnings report, which showed 34% year-over-year revenue growth, a raised full-year guidance figure, a $6 billion AWS infrastructure commitment, and the acquisition of Natoma, a governance platform for AI agent tool access. The results persuaded investors that data-infrastructure companies sitting in the path of agent workloads would benefit from AI adoption rather than be displaced by it.

What is the SaaSpocalypse and is it over?

The SaaSpocalypse refers to the roughly $285 billion wiped from software stock valuations in a 48-hour span in early February 2026, triggered by fears that AI agents would render the per-seat software licensing model obsolete. At the index level, those losses have been recovered as of June 2026, but the recovery is deeply uneven: companies whose revenue scales with agent activity have rebounded sharply, while seat-dependent names like Salesforce remain down roughly 30% for the year despite a strong earnings beat in May.

What software stocks benefit most from AI agents?

Companies whose revenue grows with agent volume rather than human head count are the ones the market rewarded in May. Snowflake, which stores and governs the data agents consume, surged 37% after earnings. ServiceNow, which routes, approves, and audits agent actions across enterprise workflows, recovered roughly 33% from its 2026 lows. Bank of America analyst Tal Liani specifically reinstated ServiceNow at Buy and Salesforce at Underperform on the same day, drawing the sector's clearest analyst line between the two pricing structures.

How does Snowflake's Natoma acquisition relate to AI agents?

Natoma is a startup that built an enterprise governance platform for the Model Context Protocol, the standard that governs how AI agents connect to external tools and data sources. Snowflake acquired it for approximately $110 million to extend its data governance capabilities to AI-driven actions — meaning enterprises can manage not just what data agents access but what actions they are permitted to take inside business applications. The deal signals Snowflake's intent to become the control layer for enterprise agent deployments, not just the data warehouse underneath them.