As generative AI sweeps across the globe, the "golden rules of investment" from the traditional software era have become completely obsolete. AI is fundamentally upending conventional valuation wisdom in the venture capital world by shattering the inverse relationship between scale and risk and reshaping financial metrics.
In a recent episode of the Sourcery with Molly O'Shea podcast, Ev Randle, a partner at the established, top-tier Silicon Valley venture capital firm Benchmark, engaged in a deep conversation with host Molly O'Shea. Their discussion covered a new framework for valuing AI companies, inference economics, the potential impact of an Anthropic IPO, and the logic behind Benchmark's AI investment portfolio.
Ev Randle immediately presented a core judgment: the entire venture and growth-stage investment ecosystem is experiencing a sense of "disorientation"—a world turned upside down. This confusion, he argues, stems from the failure of a classical principle.
He pointed out that under the new AI paradigm, the traditional inverse relationship between risk and scale has collapsed. In the past, as startups grew, they would sequentially eliminate risks related to product-market fit (PMF), unit economics, and total addressable market (TAM). But in today's AI era, "you can have businesses with well over a billion dollars in revenue that have not proven their unit economics and have not proven durable product differentiation."
This phenomenon has plunged the entire venture ecosystem into a "disorienting phase." Not only has the "spreadsheet investing era" ended, but many golden rules from the SaaS domain are now operating in reverse within the AI context.
The End of the Spreadsheet Era and the Reversal of SaaS Dogma
In the traditional SaaS investing era, high gross margins, pure software, asset-light models, and high customer retention were the "north star metrics" for evaluating great companies. These metrics fit perfectly into financial models and gave rise to quantitative standards like the "Rule of 40."
However, Ev Randle stated bluntly: "All of the golden rules that defined the spreadsheet investing era are gone. The most in-vogue companies and categories are almost the exact opposite of those golden rules."
He illustrated with three core changes. First, a shift in distribution strategy. "There's a tweet that says 'FTE is the new PLG,' which means everything is 'Palantir-izing'." Previously, implementation services heavily reliant on human labor were seen as a drag on gross margins; now they are becoming standard.
Second, a complete inversion of gross margin logic. In AI applications, "if you have high gross margins, that's actually a bad thing because it means no one is using your AI features." Due to the high cost of AI inference, frequent usage inevitably lowers gross margins.
Finally, a surge in capital intensity. To build moats against large model providers, many AI software companies are starting internal research labs to train or fine-tune their own models. This brings unprecedented capital expenditure (Capex), completely breaking the old rule that SaaS businesses didn't need to invest in heavy assets like GPUs.
A New AI Valuation Formula: Sky-High Contracts Under a P×Q×M Model
With the old rules broken, how does one evaluate an AI company's business model? Ev Randle proposed a new taxonomic perspective: P (Price) × Q (Quantity) × M (Margin).
Compared to traditional SaaS, the Q (number of target customers) for AI applications is typically similar, and the M (gross margin) is almost certainly lower than the 70%+ seen in SaaS. However, the P (price per customer) can reach staggering heights.
"Right now, you see inference platforms signing contracts with startups for nine figures (hundreds of millions of dollars), which is incredibly rare in SaaS," Ev Randle noted. He highlighted the astonishing spending power of AI coding assistants as an example: "Developers within our portfolio are spending $3,000 a month per person on Claude Code. Wow, that's $36,000 per developer per year."
This combination of business model and product innovation is seen by Ev Randle as the most important shift since the cloud era. The buyer's mindset has shifted from "purchasing a software license" to "buying intelligence or white-collar labor on-demand."
This has blown open the ceiling for corporate software budgets. Where a customer might have contributed $200,000 in revenue before, it could now be "a $20 million revenue customer for a normal company, or for some of the giants, $500 million a month."
Inference as a "Money-Printing Waterfall" and the "Trillion-Dollar Question" for Frontier Models
Discussing the explosion in AI infrastructure and the agent economy, Ev Randle used a vivid metaphor to describe the immense current demand for inference. "It's like you're sitting on the bank of a river with a bucket trying to fill it with water, but there's a giant waterfall over there. What you should do is not sit on the bank thinking, you should go stand under the waterfall... That waterfall is inference." He noted that inference-based business models are turning revenue growth curves from the old "1-3-9-20" into extremely steep "1-20-100" or "1-30-300" trajectories.
Yet beneath this massive demand, frontier models face a severe test of pricing power. Ev Randle proposed an "AI Mom Test": for a non-technical average user, 100% of their daily needs could likely be met by highly cost-effective open-source models.
He pointed out that if AI capabilities eventually hit an absolute ceiling and open-source models reach 95% of frontier model capability, "that's a really scary situation for the frontier labs," as it would severely undermine their ability to charge a premium. Conversely, if frontier labs can achieve Recursive Self-Improvement (RSI), they would possess strong pricing power. This is termed the "multi-trillion dollar question" that will determine the direction of the AI field.
Venture Ecosystem Upheaval: Massive Liquidity Impact and "Atypical" Venture Capital
On the capital side, AI's capital-intensive nature and the trend of companies staying private longer are altering the shape of the entire venture industry.
Ev Randle warned the market to prepare for the massive liquidity shock that future AI giants could bring. Using Anthropic as an example for a striking calculation, he stated: "If Anthropic goes public and gets liquidity at a $1.5 trillion valuation... it generates 35 times the return of Snowflake's Pre-IPO round. That's 35 Snowflake Pre-IPO rounds in a single financing."
He revealed that some funds have invested as much as $3 to $4 billion in a single AI company and could see a 5x return in under five years. The speed and scale of this wealth creation will send massive shockwaves through the entire Silicon Valley ecosystem, impacting real estate, new company formation, and secondary investments.
Faced with this frenzied market, Ev Randle observed that many former venture capital firms have effectively become "Alternative Asset Managers." Benchmark, however, sticks to its core philosophy: backing founders, not themes. "Great founders never go out of style, whereas business models go in and out of style."
Concluding the interview, Randle offered a direct assessment of the venture industry's structural evolution. He believes the industry's biggest problem is using the single term "venture capital" to describe two fundamentally different types of firms. "General Catalyst and Andreessen Horowitz are alternative asset managers, not venture capital firms. They have a venture product, but they themselves are not venture capital firms—they have growth products, debt products, health insurance products, wealth management products. Venture capital in many ways is still the same venture capital, but for these large firms, it's just one product among many, not the firm itself."
Comments