a16z, Anish Acharya: Is SaaS Dead in a World of AI? | Who Wins the Dev Market: Cursor or Claude Code
TL;DR
Anish Acharya argues that rumors of SaaS death are greatly exaggerated, suggesting AI will lower enterprise switching costs and expand software demand rather than replace existing systems, while value accrues to application layers that aggregate increasingly commoditized foundation models.
🏢 The SaaS 'Massacre' is Overstated 3 insights
Vibe-coding everything is flat wrong
Enterprises spend only 8-12% of budgets on IT, making the cost savings of rewriting payroll or ERP systems negligible compared to optimizing the remaining 90% with AI.
SaaS pricing power remains strong
75% of public SaaS companies have raised prices since ChatGPT's release by a mean of 8-12%, with some increasing 25% or more, contradicting narratives of collapsing durability.
AI eliminates the 'hostage' dynamic
Coding agents dramatically lower the cost and risk of switching between enterprise systems like SAP and Oracle, ending eras where customers were trapped by integration complexity.
⚔️ Incumbents vs. Startups in the AI Era 3 insights
Incumbents will dominate existing categories
Capable incumbents like ServiceNow and Adobe will leverage AI to improve existing products (Word, Photoshop, CRM) rather than be disrupted, making them better than ever.
Startups capture native AI categories
New categories that didn't exist before AI—such as AI-assisted movie production—will be won by startups, as incumbents focus on core product enhancement over pioneering new primitives.
Distribution typically beats innovation
History suggests incumbents typically acquire or match innovation before startups achieve distribution, though reduced switching costs may accelerate competitive pressure.
💡 Where Value Accrues in the AI Stack 3 insights
Foundation models are becoming commodities
With multiple providers innovating in lockstep and strong open-source alternatives, 80% of model capabilities are now substitutable while 20% represent specialized functions.
Value accrues to the aggregation layer
Applications like Cursor capture value by orchestrating multiple specialist models (Gemini for frontend, Codex for backend) into unified interfaces, sparing users from switching between CLIs.
Developer tools will fragment by use case
The market will resemble cloud providers (AWS/Azure/GCP) rather than winner-take-all markets like Uber/Lyft, with different tools serving distinct developer archetypes simultaneously.
Bottom Line
Invest in application-layer companies that aggregate commoditized AI models into specialized workflows and in startups creating entirely new AI-native categories, while recognizing that capable SaaS incumbents will likely defend their core markets through distribution advantages.
More from 20VC with Harry Stebbings
View all
Nikesh Arora on The Future of Token Costs | Memory Becoming the Moat & Why Enterprise AI Isn't Ready
Nikesh Arora argues that while frontier AI models chase consumer breadth with high false-positive tolerance, enterprise AI adoption requires expensive depth and context to handle edge cases, predicting token costs will fall 90% and traditional G&A functions will halve within three years as AI transitions from passive software to opinionated agents.
Flexport CEO: Why Revenge and Patriotism are the Best Founder Traits
Flexport CEO Ryan Peterson reveals the company is on track for $450 million in net revenue and approaching profitability, while sharing contrarian views on why fear of losing drives founders, remote work is "white collar fraud," and AI dependency poses existential operational risks.
Anthropic's Fable Banned by US Government | Wix & Adobe Hit All-Time Lows | Mistral Raising at $20BN
SpaceX completed the largest IPO in history at
How Export Controls Helped Not Hurt China & Power is the Bottleneck to AI | Perplexity CEO
Perplexity CEO Aravind Srinivas argues his company forced Google's strategic pivot to AI mode while asserting that sustainable AI value lies not in frontier models but in orchestration interfaces that maximize token value per watt for high-spending power users running autonomous agents.