NEA’s Tiffany Luck on AI IPOs, personal agents, and the ROI reckoning

NEA partner Tiffany Luck says the artificial intelligence industry is confronting a return-on-investment reckoning as enterprises pull back from indiscriminate spending on AI tools, a shift she discussed on TechCrunch's Equity podcast [1]. The pullback follows a period of aggressive AI adoption that some in Silicon Valley called "tokenmaxxing," where companies pushed employees to maximize use of large language models [1]. Uber reportedly exhausted its annual AI budget within a few months, certain organizations cut Claude licenses for segments of their workforce, and Meta Platforms killed an internal leaderboard tracking AI usage [1]. Meta, which rebranded from Facebook in 2021 to signal a strategic pivot toward the metaverse, derived 97.8 percent of its total revenue from advertising in 2023 and ranked as the world's third-largest R&D spender as of 2022, with expenditures reaching US$35.3 billion [4]. The company sits among the Big Tech firms that collectively represent about a quarter of the S&P 500 [5]. Luck, who previously helped companies adopt e-commerce, now focuses on AI investment at NEA [1]. She told the Equity podcast that personal agents represent a key area for AI adoption in consumer businesses, and that startups are emerging to help enterprises track returns on AI spending [1]. She also noted that enterprises are mixing and matching models from different providers rather than committing to a single vendor, and that forward-deployed engineers are functioning as a conduit for AI adoption inside organizations [1]. Luck argued that value is being created across the entire AI stack, not solely at the model layer [1]. The broader technology industry continues to invest heavily in infrastructure that supports AI workloads. Content delivery networks, which carry more than half of today's Internet traffic, are evolving toward self-operated Meta-CDN architectures that integrate third-party CDNs to offload traffic during capacity shortages — a model Apple has used to handle events such as major iOS updates reaching an estimated 1 billion devices [3]. On the privacy front, researchers are developing systems like Cookie Monster, a differential privacy budgeting component designed to improve advertising measurement APIs from Google, Apple, Meta, and Mozilla [2]. The approach enforces well-defined differential privacy guarantees while enabling advertisers to conduct more private measurement queries accurately, a development that intersects with the advertising-dependent business models of companies like Meta [2].