GV's Tom Hulme: 35% of the world's AI researchers are in Europe, and 80-90% of GV's deals are AI

Summary

Google Ventures managing partner Tom Hulme sees AI as the dominant thesis across his firm's portfolio, with 80-90% of GV's current activity involving AI investing. Europe holds 35% of the world's AI researchers and hosts four of the top technical universities globally—Oxford, Cambridge, Imperial, and ETH Zurich—giving the continent a foundational advantage that has historically been its main source of founding talent.

Europe's best founders tend to be repeat entrepreneurs. Guy Pajani founded Sneeq and later Tessel with GV backing. Successful companies spawn networks of follow-on founders in a dynamic Hulme compares to the PayPal mafia. GV was an early investor in GoCardless, which then seeded companies like Monzo. These networks are now producing broader effects across fintech and beyond.

DeepMind deserves more credit for its role in UK tech than it typically receives. The company was founded in 2012, five years before the 2017 transformer paper, making its trajectory far less obvious at the time. Demis Hassabis's decision to maintain a research base in the UK proved consequential. It anchored technical talent and catalyzed a wave of spinout companies, including NeoLabs, reinforcement learning companies like Recursive Superintelligence, and world model companies including GV portfolio company Odyssey.

Research-stage companies like NeoLabs fill a strategic gap in the ecosystem. Current frontier companies optimize for large language models and diffusion models, chasing benchmarks and executing well on existing S-curves. The unexplored territory lies in world models and reinforcement learning, areas where novel breakthroughs could create complementary intelligence rather than direct competition. NeoLabs typically services breakthroughs via API rather than building full-stack products, though that occasionally happens.

Q, an Israeli voice-interface company, broke out of Europe and became Apple's second-largest acquisition. GV co-led its seed and Series A rounds with Kleiner Perkins and Spark Capital. Hulme's thesis was that voice communication at 150 words per minute with intonation cues offers higher information throughput than typing at 90 words per minute, making it a rich input channel. Q built private voice communication as the non-invasive counterpart to the throughput gains from neural implants. Hulme expects Q's integration into Apple to deliver a major moment in 2028.

In AI hardware, Hulme argues that if AI models overshoot user requirements, the real value accrues to physical devices as distribution points. He points to GV's investment in Nothing, a smartphone maker, as a bet on whoever controls the hardware distribution layer capturing disproportionate value.