Software stocks extended their brutal week Thursday, with the iShares Expanded Tech-Software Sector ETF down more than 9% for the week and nearly 30% from its September 2025 high. This marks the sector's worst performance since the post-Liberation Day crash.
The catalyst: Anthropic's latest Claude updates demonstrated AI agents handling legal clerical work, raising existential questions about software that bills per user seat. If one AI agent can do the work of 50 administrative users, the enterprise software licensing model has a problem.
The bull case remains straightforward. John Campbell at Allspring Global Investments argues incumbents won't be easily disrupted because they're building their own agents. Citi's Tyler Radke sees selective opportunities in companies exposed to hyperscale data volumes, naming Microsoft, MongoDB, and Snowflake. According to Jefferies, 73% of software stocks are now oversold—an eight-year high.
The bear case is more troubling. Greg Swenson at Leuthold Select Industries Fund notes the software ETF still trades at 40x trailing earnings. History suggests emotional sell-offs don't bottom at median valuations—they overshoot. "Things don't usually tend to bottom at historical median or average levels," he told CNBC.
What makes this different from last year's DeepSeek scare: This isn't about Chinese competition on price. It's about whether AI agents fundamentally break the per-seat licensing model that's powered software margins for decades. Fundstrat's Hardika Singh framed it clearly: If software companies can pivot, this is a healthy correction. If they can't, "this is a rupture of the AI trade" and software will be "the first victims of the AI industrialization."
The spillover is already visible. Private credit firms Blue Owl and Ares Management are down 9% and 16% respectively this week, suggesting concerns about tech debt exposure.
Context matters here: Enterprise AI spending is under scrutiny across the board in 2025. This software rout follows Microsoft's earnings highlighting AI capex concerns, and comes as CIOs increasingly question ROI on generative AI pilots. The sector's P/E ratio has fallen from 85x at last summer's peak to below 60x now—still not cheap by historical standards.
The rotation is clear. Money is moving to "real economy" sectors: energy, industrials, materials. The AI infrastructure trade continues. Pure-play software? We'll see if the incumbents can prove they're building the agents, not being displaced by them.