Ev Randle of Benchmark, the early stage firm behind eBay and Uber, argues a single engineer using Anthropic's Claude Code can cost as much as an entire annual SaaS contract, inverting the metrics investors and CFOs have trusted for a decade.
A single engineer using Anthropic's Claude Code can run a company about $3,000 a month, or roughly $36,000 a year, according to Benchmark general partner Everett "Ev" Randle in his recent Podcast Alpha breakdown (The $3,000-a-Month Developer). That is roughly the price of an entire annual enterprise SaaS contract for one user, the kind of line item that defined enterprise software budgets for the last decade. Randle argues the collision is the most important pricing fact in software right now, because it inverts the metrics that investors and procurement teams have relied on since the cloud era.
On the 20VC podcast, Randle frames this as the inversion of what he calls the golden rules of SaaS. For ten years, software was sold per-seat: cheap to deliver, easy to scale, and rewarded with gross margins that climbed as customers added users. In his framing, AI flips both the pricing and the signal. Products now bill for inference, the compute used each time a model answers a prompt, which moves the cost line from a near-fixed cloud bill to a variable, consumption-driven expense. The products customers run most heavily, where AI is doing real work, now show lower gross margins because every action routes through expensive inference capacity. A high gross margin in 2026, on Randle's read, signals a product customers barely use, not a healthy business.
That shift is visible in the public numbers around the AI labs themselves. Anthropic closed a $30 billion Series G on February 12, 2026, at a $380 billion post-money valuation, a figure the company confirmed in its own announcement and that was independently reported by the New York Times, Reuters, and CNBC. Behind the headline is a business that has to keep buying the advanced chips and data centers needed to run frontier models, the largest, most capable AI systems from labs such as Anthropic, OpenAI, and Google DeepMind, at a pace the SaaS era never demanded. The capital intensity is the other side of the inference coin.
For software buyers, the practical consequence is a line-item reset. A team of twenty engineers running Claude Code at the run rate Randle cites could spend $720,000 a year on a single vendor, more than many mid-market software budgets for entire departments. Anthropic's own Claude Code pricing documentation and third-party breakdowns from Finout and Business Insider put per-developer monthly spend in that neighborhood for sustained use, not casual experimentation. Procurement teams that built budgets around seat counts now have to plan around tokens and inference volume, with the unit of value shifting from "how many users" to "how much intelligence the company is consuming."
For public-software investors, the inversion cuts in a different direction. Gross margin, the metric that separated great SaaS companies from mediocre ones for a decade, now functions as a tell for AI exposure. A horizontal tool, broad cross-industry software such as a word processor or project tracker, will see margins compress as customers route more work through an AI assistant and the inference bill grows. A vertical product, industry-specific software such as electronic health records or compliance tooling with embedded proprietary data and regulatory moats, can keep higher margins because the value is the data and the workflow, not the chat. Randle's framing, summarized in a Sourcery write-up of the same thesis, is that the old screening templates need to be re-read from the bottom up.
Benchmark itself is not making a market call. Randle describes the firm's AI portfolio as founder-first and theme-agnostic, in his 20VC interview and the Podcast Alpha breakdown, a curated map that only looks thematic in retrospect. The argument is not that everyone should sell software stocks. It is that the unit of value has changed: a seat is no longer a unit of value when one engineer can burn through what used to be a department's annual budget in a quarter.
The forward question is whether the public markets catch up. In a scenario Randle floats, a $1.5 trillion Anthropic IPO would deliver a multiple on Benchmark's position roughly thirty-five times larger than the multiple Snowflake's pre-IPO investors saw when the data warehouse company went public in 2020 and turned roughly $500 million of venture stakes into about $2.5 billion. That math is hypothetical, not a forecast, and it depends on the AI labs delivering on both revenue and the inference cost curves that determine whether the new economics actually work at scale. What is not hypothetical is the buyer-side math: every CFO planning a 2026 software budget now has to decide how much of the line item is seats, and how much is intelligence, before the year is out.