Sapphire Ventures partner says AI startups face a harder proof test

AI is pulling software valuations in two directions, and that gap matters for private investors trying to judge which startups can become durable companies. In a Crunchbase News interview, Sapphire Ventures partner Anders Ranum said public software stocks are being valued near decade-low multiples while private AI startups keep drawing record prices.

Ranum, who has spent nearly 15 years at Sapphire Ventures after 12 years in product and strategy roles at SAP, focuses on B2B software, security and industrial infrastructure. His recent investments include LangChain, WorkOS and Tractian.

Ranum told Crunchbase News that the current split between public and private market signals is unusual. He said public investors are pricing in the risk that AI could disrupt existing software businesses, while company-level fundamentals such as gross margins, free cash flow and net dollar retention have improved.

Net revenue retention, sometimes called NRR or net dollar retention, measures whether existing customers spend more or less over time after churn and upgrades. Ranum said the metric still matters because it shows whether customers see value, but he described it as a backward-looking signal. He said he now spends more time asking whether removing a product would interrupt a customer’s day-to-day operations.

Exit paths are changing

Ranum pushed back on the idea that large tech mergers and acquisitions have stopped. He told Crunchbase News that software M&A deal value rose 40% year over year in 2025 to $334 billion across 678 transactions, and said Sapphire saw more than half a dozen portfolio-company acquisitions over the past six months.

He said pricing has changed, with valuations resetting even as deals continue. On initial public offerings, or IPOs, Ranum said 2026 could become a major year if the largest private technology companies complete listings. He cited SpaceX as having gone public, Anthropic as having filed, and OpenAI as reportedly preparing to file soon.

For companies below that top tier, Ranum said the threshold for going public is higher. He said some companies may wait until 2027 or later for better conditions, which puts more pressure on startups to show both revenue growth and margins. He also pointed to the secondary market, where existing private-company shares change hands, as a source of flexibility for companies and investors.

AI claims need evidence

Ranum said traditional software-as-a-service companies are not automatically poor investments because they were not built as AI companies from day one. He described the current market as a “show me” period in which investors want proof of free cash flow, a credible path to profitability and evidence that AI is helping a company win customers.

He said companies can no longer expect investor enthusiasm from saying they are adding AI features. According to Ranum, the market is looking for signs that AI can be sold, used and paid for in a way that changes how work is done.

For infrastructure startups, Ranum said the opportunity comes from becoming deeply embedded in enterprise workflows, even as large model providers and data companies expand into adjacent tools. He said standalone startups can defend themselves if real business processes run through their products.

Factory-floor AI has clearer payback

Ranum told Crunchbase News that near-term industrial AI demand is strongest in defined settings such as packing, picking, inspection and maintenance. He said those areas have clearer labor economics, lower deployment risk and real purchasing cycles.

He pointed to Tractian as an example. According to Ranum, unplanned downtime costs the world’s 500 largest companies about 11% of revenue each year. Tractian combines sensors with AI software that looks for early signs of equipment failure, giving customers a measurable reason to pay before a contract is signed, he said.

Ranum said he favors software and sensors added to existing industrial equipment over replacing old machinery outright. In his view, factories are unlikely to remove decades-old machines just because a startup offers a newer option, while software that learns from sensor data can create longer-term value.

This story draws on original reporting from Crunchbase News.