Feb 9, 2026, 10:04 AM ETFireside Chat, Interview

a16z, Anish Acharya: Is SaaS Dead? Do Margins Still Matter? Why We Are Not in an AI Bubble?

Andreessen Horowitz (a16z) operates on a "100% visibility" mandate, requiring the firm to review 100% of deals in their domain and win 100% of the deals they pursue, avoiding reliance on luck.
The firm maintains high elasticity on pricing (check size) at early stages (Series A) to secure deals, but remains inelastic on valuation multiples and ownership percentages to maintain partnership integrity.
a16z focuses heavily on Series A because founders with shipping products and revenue provide the optimal signal for product-market fit and entry valuation compared to Seed or later rounds.
The firm explicitly rejects "king-making" as a strategy, viewing investor selection as a catalyst for already winning companies rather than a tool to anoint losers.
a16z measures partner performance via bi-annual 360-degree feedback from founders, prioritizing responsiveness and truthfulness over short-term financial returns.

The narrative that AI will "vibe code" core enterprise systems like ERP, payroll, or CRM is described as "flat wrong," as software currently represents only 8-12% of enterprise spend, yielding insufficient savings to justify the risk of rebuilding.
AI capabilities are better positioned to extend core business advantages or optimize the remaining 90% of non-software spend rather than replacing legacy software.
Traditional SaaS incumbents are not "seers" but highly capable competitors; 75% have raised prices significantly (mean 8-12%, some >25%) post-ChatGPT, indicating strong pricing power and product-market fit.
Switching costs in enterprise software are decreasing dramatically due to coding agents, transforming "hostages" (locked-in SAP/Oracle customers) into customers with viable exit options.
Venture economics are shifting: the $3-$5 billion exit is no longer considered "sufficient" for a top-tier fund, as the goal is now trillion-dollar enterprises, requiring initial assumptions to support massive scale.
The market is not in a bubble; unlike previous cycles, current AI supply build-outs (e.g., 3x capacity) are 100% absorbed by demand, and customers are paying higher prices rather than experiencing price compression.

Incumbents vs. Startups: Incumbents typically win by improving existing categories (e.g., better Word processors), while startups win by creating "native categories" that did not exist before (e.g., AI movie making).
Defensibility Shifts: "Data network effects" are evolving from static proprietary datasets to "live" data (e.g., real-time health or product usage data), which creates powerful moats even with commodity foundation models.
Network Effects: Traditional network effects remain the gold standard for defensibility; synthetic networks (like some AI products) are viewed as less durable than physical networks (e.g., Airbnb).
System of Record Risk: Systems of record with no engagement layer or human workflow are vulnerable to disruption, whereas core financial/transactional systems (e.g., banking ledgers) remain highly defensible due to accuracy demands.
Foundation Model Fragmentation: The market is shifting from a single monopoly model to a fragmented ecosystem where models act as substitutes (80%) or specialists (20%), increasing the value of an "aggregation layer" (apps).

The "Boring" vs. "Weird" Thesis: Startups thrive in "weird" categories involving human emotion, persuasion, or sexuality (e.g., companionship) where large corporations are constrained by committees and safety filters.
UI Evolution: Chat interfaces are overstated for mass consumer use; "browse-based" interfaces will remain dominant for users with intent to "spend time" rather than "save time," while chat serves high-agency, intent-based workflows.
Agent Reality: The "autonomous agent" maximalist view is overhyped; effective AI requires humans in the loop for exception handling and navigating ambiguity, which models cannot yet resolve.
Market Underestimation: Venture investors consistently underestimate market size and overestimate the difficulty of the "zero to one" transition, leading to missed opportunities in markets like payroll or credit scores.
Area Under the Curve: "Slow-growth, high-defensibility" companies (e.g., Figma) are as valuable as "high-slope" companies; long-term value often comes from the depth of the customer base and retention rather than initial velocity.

Inference as Sales & Marketing: For top AI companies, inference costs function as the new sales and marketing spend, replacing traditional CAC with consumption-based pricing.
Margin Nuance: Blended margins for AI-native companies are often lower due to subsidized free tiers (credits), but the "durable margin" on converted power users is high and justifies the initial loss.
Price Elasticity: Consumer price ceilings for AI are rising drastically (e.g., ChatGPT Ultra at $200/month vs. historical $25 ceiling), driven by power users willing to pay for high-value consumption.
LTV Uncertainty: Calculating LTV is difficult in the AI era due to retention volatility; successful metrics focus on Month 2 (M2) retention (targeting 50-70%) to filter out "tourist" traffic in Month 1.
Budget Shifts: Enterprise spending is transitioning from the SaaS budget to the human labor budget, with AI acting as a wedge to replace or augment human tasks in support, sales, and operations.

2026 Prediction: While early leaders in existing markets (coding, support) are likely to maintain dominance, entirely new "AI-native" categories are expected to emerge in 2026.
Verticalization: Model companies (e.g., OpenAI) are less equipped to build broad, opinionated feature surfaces (e.g., full legal suites) compared to specialized app companies that can orchestrate multiple models.
Companion Economy: AI companions for education, elderly care, and social interaction are growing, offering "contextual indirection" that allows users to explore relationships without the stigma of direct human therapy.
Open vs. Closed Models: While open models are used for specific cost or quality benefits (e.g., lack of safety restrictions), closed models currently offer an advantage in ambition and capability, with costs dropping 100x since release.
Future Workforce: AI will not remove jobs entirely but will shift the "area under the curve" of productivity, enabling humans to pursue more ambitious goals rather than reducing headcount by 20%.