Anish Acharya: Is SaaS Dead in a World of AI?

Episode Summary

Executive Summary: Anish Charya argues AI will not “kill SaaS” so much as reshape it: models are best used to extend core products, reduce switching costs, and create new app-layer opportunities where multiple models, richer workflows, and human-in-the-loop design matter. He sees incumbents as durable in many systems of record, startups as advantaged in native AI categories, and margins, pricing, and defensibility as changing—but not disappearing.

Main Topics: SaaS is not dead; AI will reallocate value (Priority: 5/5): The core claim is that software is oversold in the sense that not every workflow should be rebuilt with AI, but SaaS revenue remains durable in many enterprise categories. AI is more likely to augment the remaining 90% of spend outside software than replace payroll, ERP, or CRM wholesale. Switching costs and systems integration are falling (Priority: 5/5): Coding agents reduce the pain, time, and risk of moving between SaaS providers, which should increase competition and force incumbents to improve products rather than simply rely on customer lock-in. Incumbents vs startups: where each wins (Priority: 5/5): Capable incumbents will improve existing product categories, while startups are better positioned to create and own new AI-native categories that did not exist before the product cycle. The app layer is underhyped (Priority: 5/5): In a multi-model world, downstream app companies can aggregate best-in-class models, provide richer UI, and specialize by use case; this creates meaningful value even if foundation models remain powerful. Margins, pricing, and AI business models (Priority: 4/5): AI-native companies often have worse blended margins because they subsidize usage and free trials, but power users can now justify much higher ARPU and consumption revenue. The relevant unit economics need to separate acquisition spend from durable paying-user margin. Defensibility still matters, but moats are changing (Priority: 4/5): Traditional moats like networks and proprietary live data remain valuable, while weak moats based on generic data-network effects are less convincing. Systems with high human workflow density remain hard to displace. Founder quality, market size, and investing process (Priority: 4/5): Charya emphasizes founder authenticity, domain obsession, and market-size underestimation. He argues Series A is the key stage because it offers the best balance of signal, price, and ownership, and that a strong firm must see and win the deals in its domain.

Key Arguments: AI should not be pointed at rebuilding low-ROI enterprise systems like payroll or ERP when IT spend is only a small slice of enterprise budgets; it should extend core workflows or attack the larger non-software spend pool. Switching costs are dropping because coding agents make systems integration and migration easier, which benefits the ecosystem by increasing competition and improving product quality. Incumbents like ServiceNow are not helpless; they are capable businesses that can raise prices and improve products, so broad claims that all incumbents are toast are overstated. The app layer can create more value than foundation models because apps can aggregate multiple models, handle specialized workflows, and offer richer feature surfaces that model vendors are unlikely to prioritize. Model companies will often recreate primitives, but they are less well-suited to build the opinionated, domain-specific UX and surrounding product surface that makes a category durable. AI products are shifting monetization upward: power users can pay much more than the historical consumer ceiling, and consumption-based pricing plus high-end subscriptions are becoming viable. Defensibility still exists; networks remain powerful, and live proprietary data can improve outcomes even with commodity models. Margins should be analyzed by separating CAC-like subsidized usage from the durable margin profile of converted users; free trials and negative-margin onboarding can be healthy if they convert well. The best founders are deeply, sometimes irrationally, connected to the problem and can use investors as force multipliers, but they do not need investors to create value. A strong venture firm must see all relevant deals in its sector and should expect to win the deals it pursues; luck is not a strategy. Many markets labeled as one market are actually industries with many specializations, so there may be room for dozens of winners rather than one monopoly winner. In AI, human-in-the-loop systems will remain important because ambiguity, exception handling, and broader judgment cannot be fully automated yet.

Data Points: IT spend as share of enterprise budgets: 8% to 12% - Used to argue that even dramatic AI-assisted rewrite of software only affects a limited slice of enterprise spend. Price increases among public SaaS companies since ChatGPT: 75% have raised prices - Cited as evidence that SaaS pricing power has not collapsed post-ChatGPT. Typical SaaS price increase: 8% to 12% mean increase - Average price increase across public SaaS names since ChatGPT. Large price increase cohort: 25%+ raises - A meaningful subset of SaaS companies raised prices by 25% or more. OpenAI top line: $20 billion - Referenced as evidence that demand is absorbing supply rather than suggesting overbuild. OpenAI capacity scaling: 3x capacity and 3x top line - Used to argue inference supply is fully spoken for. Power-user subscription examples: $200 to $300 per month - Examples included ChatGPT at $200, Grok Heavy at $300, Gemini Ultra at $250. Historical consumer software ceiling: $20 to $25 per month - Spotify was used as the old benchmark that AI is now exceeding by 10x or more. Potential enterprise outcome size: $3 to $5 billion - Discussed as a substantial but not necessarily venture-maximal outcome for a UK or similar market. Legal software market size discussed: $50 billion to $500 billion - Charya argued AI may expand the effective market size far beyond traditional software estimates. Autonomous model token cost decline: 100x lower for GPT-4.0 tokens - Used to show model cost has fallen sharply even as capability rises. Voice/product activity metric: M1 and M2 retention framing - He suggested treating month one as traffic and month two as the first real retention test for AI products. Retention benchmark: 50% solid; 60% to 70% very good - Suggested as strong M2 retention for AI products with lots of top-of-funnel traffic.

Pivotal Quotes: "The general story that we're going to vibe code everything is flat wrong and the whole market is oversold software." — Anish Charya: Core rebuttal to the idea that AI will replace all SaaS and enterprise software. "Why would you point it at rebuilding payroll or ERP or CRM?" — Anish Charya: Explaining why AI models are more valuable when applied to higher-impact business opportunities. "I don't think we're allowed to believe in luck at Andreessen. We have to see 100% of the deals in our domain and that we win 100% of the deals that we go after." — Anish Charya: Describing the firm’s disciplined investing philosophy and deal ownership expectations.

Implications: AI is likely to expand software markets, not simply compress them. Winners will be incumbents that adapt, startups that own new native categories, and app-layer companies that orchestrate models, data, and workflows better than foundation vendors.