Why AI Isn’t Killing SaaS Yet

Episode Summary

Executive Summary: The discussion argues that the feared “SaaS apocalypse” is not supported by Ramp’s business spend data. Businesses are not broadly abandoning SaaS or switching to token-based pricing; instead, adoption is shifting toward multiple models, cost-conscious routing, and new AI-adjacent software categories like AEO. AI is reshaping software, but more through competition, workflow layers, and selective adoption than mass vendor collapse.

Main Topics: SaaS apocalypse vs. actual spend data (Priority: 5/5): Ara Karazian argues the narrative that AI will wipe out SaaS and force a new pricing model is premature. Ramp’s data shows traditional SaaS buying remains largely intact, with no meaningful collapse in adoption or spend. How businesses buy software today (Priority: 5/5): The transcript distinguishes between two claims: users shifting from incumbent SaaS to model-company competitors, and users shifting from seat/platform pricing to tokens or agentic pricing. Ramp sees neither at scale yet. Model competition and multi-model adoption (Priority: 5/5): Businesses increasingly use more than one AI model, and early adopters are broadening rather than consolidating around a single provider. OpenAI, Anthropic, Google, and others compete on cost, performance, and routing. Cost pressure and routing to cheaper models (Priority: 4/5): Ramp observes rising AI spend intensity and increasing use of platforms like OpenRouter to route tasks to cheaper or open-source models. The trend is strongest among the highest-spending firms. New AI-native software layers and categories (Priority: 4/5): Growth is occurring in adjacent software categories such as answer engine optimization (AEO), model-routing tools, and AI-native alternatives like newer CRM vendors. These are separate from frontier model providers and may be underappreciated. Legacy firms, adoption, and defensive positioning (Priority: 3/5): Legacy players are responding unevenly. Some, like accounting firms, are hedged or cautious, while others are adopting AI more aggressively or using it in marketing and client positioning. Not all change is directly attributable to AI. Productivity, labor, and measurement challenges (Priority: 3/5): The conversation notes that spend does not equal productivity. Future research should track adoption intensity, headcount, PR quality, and firm outcomes to determine whether AI improves performance or just changes workflows.

Key Arguments: Ramp’s data from 50,000 businesses and $100B in annual spend does not support a broad shift away from SaaS or a collapse in SaaS buying behavior. Token-based pricing is still a tiny share of actual software spend, so the new pricing model narrative is not yet visible in enterprise purchasing. Businesses are using multiple AI models rather than standardizing on one winner, which points to a competitive, multi-provider market. High-usage firms are increasingly cost-conscious and routing tasks to cheaper models or open-source options through tools like OpenRouter. Many feared “at-risk” vendors, such as Figma and Perplexity, remain healthy or fast-growing in Ramp’s data because they offer products the frontier labs do not directly replace. The most meaningful AI-driven growth may happen in adjacent software categories—like AEO or model-routing—rather than inside frontier model labs alone. AI adoption should be evaluated through productivity and workflow outcomes, not just spend; layoffs and headcount changes are not yet clearly attributable to AI. Legacy companies are not uniformly anti-AI or pro-AI; their responses vary by industry, incentives, and customer needs.

Data Points: Businesses tracked by Ramp: 50,000 - Ara Karazian says Ramp sees spend data from this many businesses. Annual spend observed by Ramp: $100 billion - Scale of spend data used to assess software buying behavior. Receipt-based contracts share of spend: 65%–75% - Still the dominant software purchasing model. Flat platform fees share of spend: 20%–30% - Second major software pricing format, still significant. Token-based pricing uptake: ~0.5% of spend - Usage on platforms that have launched token-based offerings. Token-based products on earnest platforms: <1% of actual spend - Reinforces that token pricing remains marginal. Token costs for typical high-intensity AI spenders: 13x increase over the last year - Shows rising AI usage costs among token-heavy firms. AI/token spend as share of business spend excluding payroll: ~2% - For high-intensity spenders, AI remains a small but growing expense. Ramp AI Index share for open router: ~3% of AI spend - Direct spend through OpenRouter is rising but still small. DeepSeek adoption: <1% of firms on the platform - Even after an initial spike, DeepSeek did not gain broad enterprise adoption. DeepSeek early adoption spike: Occurred about a year ago - Brief interest did not translate into broad usage. Anthropic adoption: Now the #1 most popular model used by businesses on Ramp - Anthropic recently overtook OpenAI in Ramp’s business adoption data. XAI adoption: ~2%–3% adoption within a few months of launch - Presented as a strong early feat, though still below leading models.

Pivotal Quotes: "“Neither aspect of SaaS apocalypse is supported by actual business spend.”" — Ara Karazian: Core thesis rejecting the claim that AI is already collapsing SaaS buying or pricing models. "“Businesses are increasingly using multiple models.”" — Ara Karazian: Explains why the market appears competitive rather than winner-take-all. "“SaaSpocalypse as a pronouncement has come way too soon.”" — Ara Karazian: Summarizes the argument that the dominant narrative is ahead of the data.

Implications: Listeners should expect AI to reshape software through competition, routing, and new product layers rather than an immediate SaaS collapse. The winners may be infra, workflow, and AI-adjacent tools, while adoption and productivity effects remain uneven and data-dependent.