Episode Summary
Executive Summary: Ramp economist Eric Harazian argues that enterprise AI adoption is still early but accelerating, with spend rising sharply even as firms try to control costs. He says Anthropic may benefit from government scrutiny, open-source models are growing from a small base, Google is underrated, and the supposed SaaS apocalypse is overstated because seat-based software remains dominant and AI-native vendors are taking share.
Main Topics: Anthropic, federal scrutiny, and the 'forbidden fruit' effect (Priority: 5/5): The discussion centers on whether White House export controls and prior DoD risk labeling will hurt Anthropic. Harazian argues the opposite may happen: government attention can strengthen Anthropic’s brand, increase perceived model power, and even boost adoption. How much companies actually spend on AI per employee (Priority: 5/5): The conversation examines Ramp data on AI spend intensity across firms. Harazian emphasizes that AI budgets are rising quickly, but for most companies AI is still a small share of spend, with heavy spending concentrated in a small cohort of tech-forward firms. Why companies stay sticky across AI vendors (Priority: 4/5): The hosts explore whether model outages or price wars would cause switching. Harazian says adoption is sticky because users care about workflow integration and product experience more than raw model quality, and advanced users often keep multiple vendors. Cost control, token management, and open-source alternatives (Priority: 5/5): A major theme is firms’ growing concern about unpredictable token spend. Harazian says businesses want routing and spend controls, which is driving experimentation with open-source or orchestration layers, though usage remains small overall. DeepSeek’s resurgence and what it really means (Priority: 4/5): DeepSeek’s renewed growth is framed as a sign that firms are testing cheaper alternatives, but Harazian argues the move is overstated and likely temporary because American model vendors can respond with lower prices and better controls. SaaS apocalypse claims vs. actual buying behavior (Priority: 4/5): The episode challenges the idea that AI will wipe out SaaS. Harazian says legacy SaaS is not collapsing; instead, AI-native competitors are growing, and most software still sells on seats rather than tokens or usage. Methodology and representativeness of Ramp data (Priority: 3/5): A brief methodological defense explains that Ramp’s data skews tech-forward but is useful for forecasting because AI adoption is led by these firms first. Harazian argues traditional data sources undercount AI adoption and miss the leading edge.
Key Arguments: Anthropic’s government controversy may help rather than hurt it by increasing brand legitimacy, curiosity, and perceived capability. Model outages do not automatically cause users to defect; workflow integration and user habits make AI tools sticky. AI spend is rising extremely quickly, but for most firms it is still a modest share of total business spend. The real demand from enterprises is not just lower prices, but controls and routing tools that make AI budgets predictable. OpenAI and Anthropic are incentivized to maximize token spend, so firms increasingly look for middleware, open-source models, or other cost controls. DeepSeek’s growth is real but small and likely not durable because incumbent model providers can cut prices and match performance. Google is underappreciated because it can compete on AI without relying solely on token revenue and already has broad enterprise distribution through Workspace and Cloud. The so-called SaaS apocalypse is overstated: seat-based software remains the dominant pricing model, and AI-native vendors are taking share faster than general-purpose model companies are replacing SaaS. Advanced AI users typically do not choose one vendor; they use multiple models and vendors to optimize for task, cost, and workflow. Ramp’s data is skewed toward tech-forward firms, but that makes it valuable for identifying where adoption is headed rather than merely describing the current average firm.
Data Points: Top 1% AI spend per employee: $7,449 per month - Harazian cites Ramp data on the most AI-intensive firms Top 10% AI spend per employee: $611 per month - Spending among the top decile of firms using AI Median firm AI spend per employee: $11 per month - Roughly the price of a basic enterprise chatbot subscription Per-business AI token spend growth: 15x - Ramp data from January 2025 through May 2026 among firms already spending on AI Monthly AI spend growth: 14% month over month - AI spend still rising even as firms attempt cost discipline Top quartile share of business spend on AI: ~2% excluding payroll; ~1% including payroll - AI remains a small share of total business spending for most adopters Firms using AI at all: 54% - Share of firms paying for or using AI in some way Firms using Anthropic: 41% - Anthropic overtook OpenAI in U.S. business adoption in Ramp data Firms using OpenAI: 39.1% - OpenAI remains large but flatter in growth than Anthropic Open-source platform usage by firms: 5% - Up from 1% last year, but still a small share of firms DeepSeek usage at peak in early 2025: ~0.5% - Temporary spike during its buzzy launch DeepSeek usage last month: 0.4% - Fast-growing from a tiny base DeepSeek earlier trough: 0.1% - Usage fell sharply after the initial spike Vendors used by top 1% AI spenders: 8 vendors on average - Advanced users spread spend across multiple AI vendors Vendors used by median AI spender: 2 vendors on average - Less advanced firms tend to use far fewer vendors SaaS spend mix: seat-based contracts: 60%–75% - Still the dominant software pricing model Metered usage as share of spend: <5% - Usage-based billing remains a small part of overall software spend Adobe metered revenue share: ~0.5% - Example showing that usage-based pricing is still limited even at a major software company AI companies’ business revenue mix: ~80% token-based - Harazian argues OpenAI and Anthropic are structurally incentivized to drive token usage
Pivotal Quotes: "If anything, it’s probably the case that the Department of Defense’s supply chain risk labeling accelerated Anthropic’s adoption with businesses." — Eric Harazian: On whether government scrutiny will hurt or help Anthropic "The single most important factor they list for why they have not adopted AI comprehensively throughout their organization is a concern around the cost." — Eric Harazian: On why enterprise AI adoption is still incomplete "There is no indication, at least in our data, that there are even early signs of a slowdown amongst these kinds of SaaS vendors." — Eric Harazian: On claims that AI is killing SaaS
Implications: Enterprise AI is still in an experimentation-and-optimization phase: spend is rising, but control and routing matter more than raw model hype. Vendors that solve cost governance and workflow integration may win share, while fears of a near-term SaaS collapse look premature.
About Big Technology Podcast
The Big Technology Podcast takes you behind the scenes in the tech world featuring interviews with plugged-in insiders and outside agitators. Alex Kantrowitz, a Silicon Valley journalist who's interviewed the world's top tech CEOs — from Mark Zuckerberg to Larry Ellison — is the host.