Salesforce shed 4% of its value on Tuesday. Cursor, a four-year-old coding startup nobody outside engineering circles had heard of two years ago, is now growing faster than any software company in history. Both facts trace back to the same nine-day-old model.
OpenAI's GPT-6 Astra didn't just raise the ceiling on what AI can do. It split the software industry into winners and casualties along a line that has nothing to do with company size or balance sheet and everything to do with what kind of code a business sells. The same week the model reignited fears that AI would swallow entire categories of enterprise software, one European private equity firm moved to merge two of its portfolio companies for scale, while an AI-native coding editor kept adding customers as if the disruption applied to everyone but its own market.
The casualties showed up first. The S&P 500 ended lower on Tuesday, with Salesforce and other software makers losing ground, with Salesforce and Intuit falling about 4% while ServiceNow lost 5%. The S&P 500 software and services index fell 1.4%, down for a second straight day. The trigger, according to Reuters, was OpenAI's own product: OpenAI's launch last week of its newest model, GPT-6 Astra, reignited speculation that AI could compete with services now offered by specialized software companies. Argent Capital Management's Jed Ellerbroek put it plainly to Reuters: "Astra has kind of reignited the software disruption fears and it's resumed that old trend that we got used to for a while, where semiconductor stocks and data center capex beneficiaries do well, while software stocks do poorly."
That fear has a name now. Traders and analysts have been calling it the "SaaSpocalypse" — the theory that as AI models get good enough to do the job a subscription software product used to do, the subscription stops making sense. The software sector has had a choppy 2026 amid fears that the rapid development of AI could impact companies. Some of that damage has already reshaped ownership structures. Silver Lake, the private equity firm behind a string of enterprise software bets, announced Tuesday that it will merge two of its French portfolio companies, Cegid and Silae, in a deal with an enterprise value of more than 10 billion euros ($11.6 billion), per a statement announcing it. French firms Cegid, which sells business management software, and Silae, a payroll and HR platform, would combine in a deal that would bring together payroll, accounting, e-invoicing, digital finance and payments data, and create a 1,400-person developer team to invest in AI.
The logic is defensive consolidation dressed up as strategy. The companies are expected to have an annual revenue of 1.6 billion euros ($1.9 billion), the FT reported. Silver Lake's Christian Lucas, who will chair the combined business, framed it as an offensive move rather than a retreat: "Together, Cegid and Silae can offer a unique end-to-end suite of software and AI-powered products serving the SMBs and their professional advisors who are the dynamic heart and soul of the European economy." Translation: two mid-sized vendors betting that scale, and a dedicated AI engineering bench, buy enough time to figure out how to compete with the thing that's scaring their investors.
Now flip the lens to the tool that's doing the scaring. OpenAI didn't pitch Astra as a chatbot upgrade. It called the model "the best model for software engineering to date", and staked the claim on benchmarks: on ExploitBench, Astra achieved a perfect score of 100%, compared with 78.5% for GPT‑5.6 Sol, our previous frontier cyber-capable model. The company's Preparedness Framework now classifies Astra's cyber capability as "Critical" — the first model to cross that line — meaning it can already assist with secure code review and patching, with more advanced exploit-generation capabilities gated behind a restricted access program. It isn't cheap: Astra's API runs $10 per million input tokens and $50 per million output tokens... above GPT-5.6 Terra's $2/$12 and Claude Opus 5's $5/$25 rates. OpenAI's bet is that better reasoning burns fewer tokens overall, which is a different argument than "our model is cheap" — it's "our model wastes less of your money failing."
None of that pricing has slowed the AI-native coding tools riding the same wave that's sinking Salesforce. Cursor, the code editor built by Anysphere, hit roughly $4 billion in annualized revenue by June, a run-rate climb that outpaces the fastest previous SaaS growth curves — Slack needed about seven years to reach its first billion in revenue; Cursor did it in about eighteen months, according to data compiled by Gradually AI. Cursor now holds an estimated 18% share of workplace AI-coding usage, tied with Anthropic's Claude Code and trailing GitHub Copilot's 29%, per JetBrains' January 2026 developer survey cited by AI Business Weekly. The category itself has outgrown its own hype cycle: AI coding tools pulled in $12.8 billion in revenue in 2026, more than double the $5.1 billion recorded in 2024, and more than half the code now committed to GitHub carries some AI assistance, according to figures reported by TechCrunch.
That's the split running through the industry right now. The companies selling horizontal business software — payroll, CRM, service tickets — are the ones whose stock charts turned over this week, because a general-purpose model can increasingly do a passable version of what they sell. The companies selling picks and shovels to the people writing that model's competitors are printing revenue growth curves nobody in enterprise software history has matched. Astra doesn't threaten Cursor. It's the reason Cursor exists.
The uncomfortable truth for boardrooms watching their software multiples compress is that AI didn't create one disruption. It created two, running in opposite directions, and which side of the line a company sits on was mostly decided years before anyone had heard of GPT-6.
