The street is already rewriting the script. Microsoft had been off roughly 20-25% at its worst point in early 2026, the IGV ETF cratered, and suddenly a violent snapback has everyone chanting the classic buy-the-dip mantra. Oracle and Microsoft led the charge. CNBC desks are buzzing about high-quality enterprise names finally getting their due after the semis rotation reversed. Fundamentals are still growing, they say. Short covering plus macro relief equals the bottom is in. You’ve heard this movie before—weak hands shaken out, smart money steps in, and we all move higher.
You should ignore the barking. This is textbook relief rally noise masking a structural repricing that AI agents are accelerating, not pausing. The market wants to treat any bounce as vindication for the old software playbook. Reality is the punchline: enterprises are already testing agents that consolidate workflows and reduce the total number of seats and licenses they pay for. One Claude-powered agent handling tasks that used to require multiple human seats doesn’t expand the TAM—it shrinks the dollars flowing into traditional per-seat contracts. That dynamic didn’t disappear because the ETF had a green week.
Look past the hopium at the hard numbers. IGV surged about 14% in that historic week ending mid-April 2026, yet it remained down roughly 19-20% year-to-date. Microsoft clawed back some ground but had suffered one of its worst drawdowns since the financial crisis amid explicit AI disruption fears. In Q1 2026, software stocks delivered their worst relative performance versus the S&P 500 in recorded history—21% underperformance, eclipsing the dot-com bust, GFC, and 2022 bear market. Goldman Sachs analysts directly called out that valuations had priced in 15-20% medium-term growth now sitting at real risk from slower license expansion.
The multiples compression tells the sharper story. Application software EV/NTM revenue multiples contracted 41% over the prior twelve months to around 3.4x as of early 2026 data points—well below the pre-pandemic average of 7.8x. That’s not random rotation noise; it’s the market embedding the fact that AI agents replace bundles of licenses rather than simply layering on top. CEOs can tout AI add-ons and defend margins on earnings calls, but the cohort’s Rule of 40 metrics and revenue guidance have already slowed to low-teens median growth. Filings keep surfacing the ignored datapoint: persistent seat compression as buyers consolidate tools.
Screenshottable stat line: IGV +14% (best week since October 2001 per Bloomberg reporting) yet still -19% YTD as of mid-April 2026; application software EV/NTM revenue at 3.4x (41% contraction from prior levels, vs 7.8x pre-pandemic average); software vs S&P 500: -21% relative underperformance in Q1 2026, worst in history. Deadpan fact bomb: Software stocks just delivered their best week in 25 years… while sitting 19% lower for the year. Exactly the kind of violent snapback that often precedes the next leg of repricing when the AI license-reduction math keeps showing up in 10-Qs and customer renewals.
The contrast with the infrastructure layer sharpens the point. Semis and capex-heavy plays had run ahead by double digits in stretches before this flip because the real near-term money in AI flows to picks-and-shovels: data centers, chips, power. Application-layer software faces the disruption risk. You saw over $1 trillion in value evaporate earlier when Anthropic’s Claude agent plugins and similar tools triggered the initial rout—fears were explicit and tied to workflow consolidation, not temporary macro jitters. Lower rates hopes or short covering can spark a bounce, but they don’t rewrite the embedded slower growth now reflected in those compressed multiples.
This isn’t a blanket “software is dead” call. High-quality incumbents retain real moats—switching costs, data advantages, and bundled ecosystems that buy them time to adapt. But the lazy consensus that any relief rally equals the bottom is simply wrong. Enterprise buyers are actively piloting agents that reduce total license needs, and that pressure builds regardless of one strong ETF week.
Stay underweight the application software rebound. Position accordingly for continued multiple compression or selective pain in names most exposed to seat replacement. The infrastructure capex layer still captures the bulk of AI spending dollars long-term; the application dogs barking changes none of that math.