A Contrarian Analysis of the Software Selloff
The SaaSpocalypse unfolded in three distinct waves, each amplifying the fear that AI would render the entire SaaS business model obsolete.
SaaS underperformed S&P 500 by 24 percentage points. CIOs redirected budgets toward AI infrastructure. IT budget growth decelerated to 3.4%. Hyperscalers committed to $470B+ in AI infra spending for 2026. The money was moving from software humans use to infrastructure AI runs on.
Anthropic released Claude Cowork. Narrative shifted from "AI helps SaaS" to "AI replaces SaaS." Claude Code: zero to $2.5B ARR in nine months. A Google engineer said one year of work was done in an hour. CNBC built a Monday.com clone live for ~$15 in compute. Monday.com has a $4B market cap.
NOW -11% despite beating earnings (9th straight). TEAM -35% in one week. MSFT shed $360B in a single day. Feb 20: "Claude Code Security" announced — CRWD lost $20B in 2 days. PANW RSI 24. ZS RSI 19. Feb 22: CitriniResearch named the "Intelligence Premium Unwind." IGV posted its worst January since Oct 2008.
Bank of America analyst Vivek Arya identified the intellectual incoherence at the heart of the selloff. The market is simultaneously pricing in two mutually exclusive scenarios.
Hyperscalers are spending $650 billion on infrastructure that won't generate adequate returns. AI adoption will disappoint.
AI adoption is so pervasive and successful that it makes the entire software business model obsolete. Every SaaS company dies.
If A is true, the agentic revolution that kills software doesn't happen.
If B is true, the infrastructure spending is justified and accelerates.
The market is pricing both simultaneously — and that is, in BofA's words, "internally inconsistent."
"Agents Use Tools, Not Replace Them"
NVIDIA CEO Jensen Huang, February 26, 2026
Tool companies (NOW, SAP, CRM) will use agentic systems to develop their software. But enterprises will use agents to use those tools. Agents won't replace the tools — agents will use them.
"I have 42,000 biological employees, and I'm going to have hundreds of thousands of digital employees, and together we're going to use a lot more tools." C compilers, Python instances, software usage — all growing.
CRM data, IT tickets, security telemetry, HR records — agents need somewhere to read from and write back to. These systems don't get replaced; they get used more.
"Nobody's going to understand customer service better than ServiceNow, and they're going to come up with agents really fine-tuned and optimized for the type of work that uses the tools they have."
Jensen Huang is talking his own book. More agents using more tools = more GPU demand = more NVIDIA revenue. His "hundreds of thousands of digital employees" framing is literally a GPU demand story. This doesn't invalidate the logic — his argument stands on its own structural merits — but it should be weighted accordingly.
Scope limitation: Jensen's argument was specifically scoped to mission-critical enterprise platforms with deep systems of record. He never defended single-function tools. Monday.com and Asana were not protected by this thesis.
The market is treating software as a monolith. It is not.
The market is pricing all three tiers identically. That is the category error. That is the opportunity.
Beneficiary — more AI = more demand
Market treatment: Sold as "SaaS"
Our view: Contrarian Bull
Adapting — switching costs + AI monetization
Market treatment: Sold as "SaaS"
Our view: Contrarian Bull
Existentially Threatened — low moats
Market treatment: Sold as "SaaS"
Our view: Bear case is correct
The bears have a compelling narrative. But narratives must survive contact with how enterprises actually operate.
Replacing Workday's HCM: 12-24 months of data migration, retraining, compliance re-cert, and parallel running. Open-source has been free for 20+ years — and commercial software still flourished.
The CNBC demo showed 2% of the product. The other 98%: security hardening, SOC 2, GDPR, 200+ integrations, RBAC, 99.99% uptime SLAs, 24/7 support, multi-tenant scale, edge cases, years of patches.
CrowdStrike processes 500B events/day. Decades of threat intelligence, global telemetry, breach response. Every new AI agent = new attack surface needing security. More AI = more cybersecurity demand.
SOC 2, HIPAA, FedRAMP, CMMC, PCI-DSS — years to obtain, ongoing maintenance required. CMMC wave: 220,000+ defense contractors must certify 2025-2028. A vibe-coded tool has none of this.
Procurement cycles. Vendor risk assessments. Legal review. Board approvals. Change management. Training. The inertia that slows adoption also protects incumbents.
Intellectual honesty requires acknowledging what the bear case gets right.
Is SaaS per-seat pricing more like Adobe, or like newspapers?
$600-750B in loans to software companies underwritten against growth assumptions AI is systematically dismantling.
We are currently analogous to late 2015 (post-Third Avenue freeze, pre-energy bankruptcy wave) or August 2007 (post-BNP Paribas canary, pre-Lehman). The critical variable: whether the PE-insurer nexus transforms a manageable ~$100-200B sector loss into a systemic event. This is a risk to the financial system, not to the operating companies themselves.
Against the bearish narrative — the data that argues for asymmetric opportunity in Tier A and Tier B companies.
These are not companies being disrupted by AI.
They are companies monetizing AI.
The market is pricing them as victims while their revenue says otherwise.
The resolution pathways — from the wreckage, the signal emerges.
The "best of breed" era is over. Surviving platforms absorb adjacent markets. Palo Alto's $25B CyberArk acquisition is the template. Weaker companies become targets at distressed valuations. The M&A cycle at current multiples could be generational.
Per-seat to usage-based to outcome-based. Near-term revenue variability and multiple compression (the current pain). But ultimately more durable pricing that grows with AI adoption rather than shrinking with headcount.
Over the next 2-3 quarters, platform companies will demonstrate AI is a revenue driver, not destroyer. Agentforce trajectory, NOW agentic deals, CRWD Charlotte AI attach rates — each print chips away at the apocalypse narrative.
Jensen's thesis: ServiceNow doesn't get replaced by an AI agent — the AI agent runs on ServiceNow. CrowdStrike doesn't get replaced by Claude Code Security — Claude Code Security needs to be secured by CrowdStrike.
In January 2025, NVIDIA lost $600 billion in market cap on DeepSeek fears — the argument that Chinese open-source AI would destroy the commercial AI infrastructure model. Within a year, NVIDIA briefly touched a $5 trillion valuation.
The SaaSpocalypse has the same structural characteristics: a narrative-driven panic in companies that are beneficiaries, not victims, of the trend the market fears.
Composite scores across five dimensions: AI monetization traction, moat depth, valuation dislocation, technical oversold severity, management response quality.
Derived from BigPic Capital's research system — not opinion, but a composite of the data.
Every name in the top 5 trades at a 29-44% discount to its own 3-year average P/S multiple.
We are not calling a bottom. We are not recommending trades. We are presenting data that shows the market is pricing an extinction event into companies whose operating metrics argue otherwise. Private credit contagion could drive prices lower before the re-rating occurs. Position sizing and risk management remain critical — the data says these companies are mispriced, not that they can't get more mispriced in the short term.