Enterprise AI Displacement — Digest for May 2026
TL;DR
The enterprise software sector is repricing itself around a single fear: AI agents will collapse the per-seat subscription model that built $100B+ companies. The market has already priced in $1 trillion in disruption risk, but the real action is now in who survives the transition. Zendesk is betting on outcome-based pricing to align with customer ROI; Anthropic has overtaken OpenAI in enterprise adoption by shipping agentic code tools; and SaaS gross margins are compressing structurally as inference costs rewrite unit economics. The winners will be companies that can either govern AI agents (ServiceNow's thesis) or monetize them without destroying margin (Zendesk's bet). The losers are those still selling seats.
The Repricing: $1 Trillion in Value Destruction
The software sector has entered a historic repricing cycle, with market fears about AI-driven disruption now baked into valuations across the board. The iShares Expanded Tech-Software ETF fell more than 24% in Q1 2026 — its steepest quarterly drop since Q4 2008 — as software forward P/E multiples fell below the S&P 500 for the first time in recent memory.
"AI agents like Anthropic's Claude Cowork and OpenAI's Codex can execute tasks currently performed by human workers at computers. This means legacy SaaS companies could end up selling fewer seats — compressing the per-seat subscription revenue model that built companies like Salesforce into $170B+ enterprises." — The "SaaS Rout of 2026"
SAP and Salesforce are each down roughly 33% from their 52-week highs, while ServiceNow has lost roughly half its value over 12 months despite consistent 20% revenue growth. HubSpot is down roughly 50% in 2026 year-to-date. The repricing reflects a structural anxiety: if AI can do the work, why buy more seats?
The market has drawn a sharp distinction between companies that own the ontology — the living map of how a customer's business actually works — and those that don't. Palantir trades at roughly 35x forward sales because it sits at the center of that ontology layer. SAP and Salesforce trade at just 4x forward sales, suggesting the market expects little AI-ready innovation from incumbents.
What to watch: Whether the May 2026 rebound in tech stocks (triggered by softened views on the "existential AI threat") holds or marks a dead-cat bounce ahead of Q2 earnings season.
The Margin Squeeze: AI COGS Rewriting Unit Economics
AI inference costs are introducing a structural variable cost that legacy SaaS never had to contemplate. For every $1 million in AI product revenue booked in 2026, roughly $230,000 exits as inference cost before any personnel expense, meaning the 80% gross margin era is over.
"ICONIQ January 2026 data: average AI product gross margin at 52%, up from 41% (2024) and 45% (2025) — improving but far below traditional SaaS. Bessemer Venture Partners: LLM-native company gross margins around 65% vs 80-90% ceiling of prior cloud era." — AI COGS Problem: SaaS Gross Margin Compression 2026
HubSpot's gross margin slid from 85% to 84% as AI rollout costs accumulated. Snowflake's product gross margin sits at 67.2% and is targeting 75% for fiscal 2027 — implicitly conceding that AI workloads drag unit economics. The exception is Datadog, whose gross margin holds at 80% because its LLM Observability product is software about AI workloads, not an AI inference product itself.
The best operators are recovering margin through model routing (small models handle 80% of queries, frontier models only the complex 20%), prompt caching (which offers ~90% discounts on cached tokens), and shifting to consumption pricing to pass variable cost to customers. But for most legacy SaaS companies, the Rule of 40 calculation — growth plus margin — now penalizes the AI transition even when revenue is accelerating.
What to watch: Whether public SaaS companies can stabilize gross margins above 70% by Q3 2026, or whether the compression deepens as AI feature rollout accelerates.
The Pricing Model Shift: Outcome-Based Pricing Challenges the Status Quo
Zendesk is making a strategic bet that outcome-based pricing — charging only for verifiably resolved interactions rather than per-seat access — can disrupt the SaaS status quo and force competitors to justify their traditional models.
"By charging only for resolved interactions, it aligns incentives with enterprise buyers demanding ROI from AI investments. If successful at scale, this could force competitors like Salesforce and ServiceNow to justify their traditional seat-based or consumption models." — Zendesk Bets on Outcome-Based Pricing
Zendesk's internal "Zen on Zen" program achieved 60%+ autonomous resolution, 30% reduction in manual ticket volume, and a 20% CSAT increase. A BritBox deployment hit 47% autonomous resolution with 27% faster resolution times. A major DMV customer achieved 70% automated resolution in just three days. These aren't incremental wins — they're proof that outcome-based pricing has a real foundation.
The strategic significance is profound: if Zendesk can deliver auditable, verifiable outcomes at scale, it forces every competitor selling seat-based or consumption-based models to answer a harder question: Why should we pay for seats if we only care about outcomes? Salesforce and ServiceNow will counter by highlighting their governance frameworks and orchestration layers, but the category tension is now visible.
What to watch: Whether Zendesk can scale outcome-based pricing to enterprise customers without collapsing into disputes over what counts as a "resolved interaction," and whether competitors adopt similar models or double down on governance narratives.
The Vendor Crossover: Anthropic Overtakes OpenAI in Enterprise Adoption
Anthropic has overtaken OpenAI in enterprise adoption for the first time, driven by Claude Code — a terminal-native agentic coding tool that has become the fastest-growing product in Anthropic's history.
"In April 2026, more U.S. businesses paid for Anthropic's Claude than for OpenAI's ChatGPT for the first time, per the Ramp AI Index (50,000+ companies): Anthropic at 34.4%, OpenAI at 32.3% — a 2.1-point gap." — Anthropic Overtakes OpenAI in Enterprise AI Adoption
Claude Code was generating $2.5B+ in annualized revenue by February 2026, with business subscriptions quadrupling since January 1. An estimated 4% of all GitHub public commits worldwide are authored by Claude Code. Anthropic's annualized revenue run rate crossed $30B in April 2026, above OpenAI's ~$24-25B. More than 1,000 enterprise customers now spend $1M+ annually on Anthropic.
But the crossover came with friction. Anthropic shifted enterprise billing from bundled seat-based pricing to per-token billing in April 2026, introducing a structural pricing conflict: the company earns more when customers consume more tokens, creating incentives to push users toward expensive flagship models. Uber's CTO reported burning through the entire 2026 AI budget in four months, with individual monthly costs of $500-$2,000 per engineer. At a 1,000-engineer organization, annual AI tooling costs could reach $6M–$24M from a single vendor.
OpenAI responded with a $4B+ counter-offensive: launching the OpenAI Deployment Company (backed by TPG, Bain Capital, Goldman Sachs, and others) and acquiring Tomoro, a London-based AI consultancy with ~150 engineers and enterprise clients. CEO of applications Fidji Simo called Anthropic's gains a "wake-up call."
What to watch: Whether Anthropic's per-token pricing model drives customer churn as organizations realize the true cost of scaling, or whether the company's product velocity keeps competitors at bay despite pricing friction.
The Lock-In Mechanics: AI Dependency Hardens Faster Than Cloud Ever Did
Enterprise AI dependency is calcifying faster than cloud lock-in ever did, driven by API integration, agent frameworks, fine-tuning investment, and infrastructure entanglement. Yet enterprises remain dangerously underestimating the switching cost.
"A Zapier survey of 542 U.S. executives with active AI vendor contracts found that 81% worry about vendor dependency, yet only 6% said they could lose their primary AI vendor without disruption." — AI Vendor Lock-In Builds Faster Than Cloud Lock-In Ever Did
The lock-in is invisible until a swap attempt reveals it. When enterprises integrate against a provider's API, they tune prompts to that model's instruction-following style, design error handling around specific failure modes, and calibrate quality thresholds against its outputs. All of that work is lost when switching vendors. Real-world events compound the problem: OpenAI shut down DALL-E models with only six months notice; Anthropic retired multiple Claude models on short notice and revoked OAuth access for OpenClaw users in early 2026.
The Zapier survey revealed a critical disconnect: nearly 9 in 10 executives believe they could switch AI vendors within a month. Among the two-thirds who've actually tried, 58% say it failed or took far more effort than expected. 47% of leaders said at least one key business function would stop working if their primary AI vendor had significant downtime or a major policy change.
What to watch: Whether enterprises begin negotiating data portability provisions and exit clauses into AI vendor contracts, or whether lock-in deepens unchecked through 2026.
What Surprised Us
-
Intuit's 3,000-person layoff (17% of workforce) signals that cost-cutting is now the dominant narrative among legacy software companies, even those with strong fundamentals. The company is betting that AI-driven efficiency will let it do more with less — a bet that may or may not pay off, but signals how quickly the industry has internalized disruption anxiety.
-
Salesforce is trading at a forward P/E of ~14x despite $41.5B in revenue and $72.4B in total remaining performance obligations. Bank of America's Underperform rating crystallizes the bear case: Agentforce is only ~2% of quarterly revenue annualized, and the market is pricing in that the AI transition will destroy seat-based revenue faster than new AI revenue can replace it. This isn't a valuation miss — it's a structural bet that the business model is broken.
-
Figma is delivering 46% YoY revenue growth, 139% net dollar retention, and 85% gross margins, yet analysts are cutting price targets. This is the clearest signal that category disruption, not company execution, is driving repricing. If AI-native design tools can commoditize design output, even exceptional results won't drive multiple expansion.
-
The Pentagon's supply chain risk declaration against Anthropic (and the conflicting legal guidance between DC and SF courts) has created a hidden friction point for defense contractors and federal agencies. 100+ enterprise customers raised concerns. This could fragment the AI vendor ecosystem in ways that aren't yet visible in public markets but will matter for government and regulated verticals.
Open Threads Worth a Vote
-
[question] SAP-Anthropic partnership: How deep is the integration and what does it mean for SAP's AI positioning? — SAP trades at 4x forward sales partly because the market doubts its ability to make customer data intelligible to AI agents. If Anthropic has cracked that ontology layer, it changes SAP's repricing calculus.
-
**[question] Which AI-native ERP/