Tech • AI • Robotics
SaaS Apocalypse: Why Wall Street Got It Wrong!
8/10TL;DR
Salesforce’s sharp rebound and a broader reassessment of enterprise software suggest the predicted SaaS apocalypse is fading, even as weaker software categories remain exposed to AI disruption.
KEY POINTS
Wall Street’s reversal on SaaS
In the first half of 2026, investors heavily sold software stocks on the thesis that increasingly capable AI agents would let companies build internal tools instead of paying recurring subscriptions. Salesforce fell to around $150 in June, roughly 45% below earlier levels, and some funds built explicit bearish positions against the sector. That narrative has now been challenged by recent earnings and product moves, with Salesforce surging 22% in one session and returning to a record high.
Why Salesforce rebounded
The rebound was driven less by traditional seat growth than by evidence that Salesforce can remain central in an AI-driven enterprise stack. A key catalyst was its agreement with Anthropic, allowing Claude to act as an interface over Salesforce data through Cloudforce. Investors also focused on the mark-up value of Salesforce’s stake in Anthropic, estimated at a $2.6 billion gain, reinforcing the view that the company is positioned on both the application and AI infrastructure sides of the market.
Enterprise software is not one category
The latest market reaction has strengthened a distinction inside software between highly customized enterprise platforms and more replaceable tools. Systems such as CRM and ERP products, including Salesforce and SAP, are deeply embedded in billing, compliance, security, identity management and company-specific workflows. By contrast, lighter products with more standardized deployments face greater risk if AI can reproduce their interface and basic functionality at lower cost.
The power of the system of record
Large enterprise platforms benefit from being the system of record, storing years of operational data and the logic of how a company actually runs. That creates high switching costs and extremely strong retention, with customers often expanding spending over time rather than leaving. In that model, the value is not just a user interface but the underlying data structure, workflow history and custom implementation that AI tools still struggle to replicate safely.
Why rebuilding software internally remains hard
The idea that companies could simply recreate core business software with AI-generated code has run into practical constraints. Custom-built replacements still require testing, debugging, compliance checks and long-term maintenance, especially in finance, tax, invoicing and multi-entity operations. For many businesses, using engineering teams to rebuild a CRM or back-office stack is increasingly seen as a distraction from core product development rather than a meaningful cost saving.
AI may expand software revenue, not destroy it
A major shift is emerging in pricing. Instead of charging only per human user, software groups are starting to bill for agents, actions and usage. Salesforce’s Agentforce reportedly increased annual recurring revenue from about $1.2 billion to $1.5 billion in a quarter, illustrating how AI can create new monetization layers. In this model, agents become heavy users of enterprise data, turning underused software environments into higher-consumption platforms.
Marc Benioff’s strategic bet
Marc Benioff appears to be embracing a risky but potentially decisive idea: the enterprise interface no longer has to belong solely to the software vendor. Letting Claude become a front end to Salesforce data acknowledges that conversational AI may become the main operating layer for workers. Rather than defending a legacy interface, Salesforce is trying to own the records, permissions, billing and agent activity underneath, where the stickiest enterprise value still sits.
Not all software is recovering
The reassessment does not mean every software company is safe. Some categories remain vulnerable, especially products built around aggregation, scraping or generic workflows. Examples cited by market watchers include companies such as ZoomInfo, whose core model may be more exposed than infrastructure-oriented or deeply embedded platforms like Snowflake. The new divide is no longer software versus AI, but software with durable data gravity versus software with limited defensibility.
The broader AI boom remains intact
The software rebound is unfolding alongside continued strength in AI infrastructure. Nvidia reported $96 billion in quarterly revenue, up 106% year on year, a pace that suggests spending on AI compute remains intense. At the same time, acquisitions and licensing deals across the market indicate that major players are racing to combine models, infrastructure, open-source distribution and enterprise data into bundled platforms rather than isolated tools.
CONCLUSION
The selloff that treated all software as obsolete now looks increasingly simplistic. Enterprise platforms that control critical data, customization and agent usage may emerge from the AI transition stronger, while more generic SaaS products still face a harsher reckoning.
Explain this