Nathan Benaich on the State of AI: 90% of research uses Nvidia, compute index up 100x in 3 years, and EU's $1.1B plan is a house for ants

Summary

Nathan Benaich, founder of Air Street Capital, released his annual State of AI report this week, touring San Francisco and New York to present findings from the roughly 300-page document. Two headline metrics define the current moment: a proprietary compute index tracking documented cluster sizes across major companies shows approximately a 100x expansion over the past three years, and a programmatic analysis of open-source research papers finds that 90% of AI research relies on Nvidia chipsets. AMD is beginning to register in that dataset, consistent with broader market signals around its growing competitiveness.

Hardware Alliances and the Nvidia Moat

The Nvidia dominance figure carries strategic weight given the capital now flowing into alternatives. OpenAI holds warrants in AMD at a strike price of $600, creating a financial incentive to make AMD infrastructure viable at scale. Benaich draws a parallel to Anthropic's relationship with AWS Trainium, characterizing Anthropic as effectively providing IT support to help AWS close the hardware gap with Nvidia. Both arrangements reflect the same underlying dynamic: hyperscalers and frontier labs are motivated to erode Nvidia's pricing power, but the engineering lift remains substantial.

Early-Stage Dynamics

Round sizes at the seed stage have inflated, but the underlying logic of early AI investing has arguably improved. Founders no longer need to consume a $3-5M seed budget building proprietary models and assembling training data before testing customer demand. Off-the-shelf model capabilities that would have seemed implausible a decade ago now let teams run product-market-fit experiments from day one. Benaich frames this as a return to fundamentals, where team execution velocity matters more than the initial idea.

Consumer Monetization and Agentic Commerce

A survey of approximately 200 AI practitioners conducted for the report found that 95% use AI in both personal and professional contexts, 75% pay out of pocket, and 10% spend $200 per month or more. The survey skewed toward educated professionals in North America and Europe, so it likely overstates broad consumer willingness to pay, but the figures are directionally notable.

Benaich sees agentic commerce as the more significant near-term revenue vector. Data in the report indicates that e-commerce conversions originating from agent-driven conversations convert at a higher rate than direct traffic. Brands that invest early in content optimized for agent retrieval may compound that advantage as labs use reinforcement learning to improve personalization, creating a self-reinforcing cycle.

The EU's $1.1B AI Plan

The EU's recently announced $1.1 billion AI initiative draws little enthusiasm. Benaich's framing is blunt: the amount is structurally inadequate relative to the capital being deployed by US and Chinese competitors. He points to two prior European efforts as cautionary precedents. The Quaero project, a government-backed search engine positioned as a European alternative to Google, failed without making a meaningful dent. The French government spent years refusing to procure Palantir on sovereignty grounds, ultimately adopted it anyway after burning comparable sums on domestic alternatives.

The broader concept of AI sovereignty is, in Benaich's view, largely illusory for non-American nations. Purchasing Nvidia chips and US-built software and installing them domestically does not constitute sovereign AI capability, because the underlying systems require updates and can effectively be switched off by their vendors. True AI sovereignty would require full-stack independence that no government outside the US and China is close to achieving. The primary beneficiaries of the sovereign AI narrative, he argues, are the technology suppliers who gain additional addressable markets.

Defense and the Broader Outlook

Beyond software, Benaich flags national security and defense as a substantial and underpenetrated market for AI, specifically citing electronic warfare and autonomous systems. Air Street has positions in Delian and other European defense-adjacent companies. His overall read on the sector remains constructive, anchored to whether AI continues to unlock use cases that were previously not economically feasible, which he characterizes as the core investment thesis and the metric worth watching as macro pressures build.