DeepMind's Growth, Innovations, and Acquisition by Google

Key Insight

DeepMind's initial fundraising was a significant challenge; after securing seed money, they calculated needing another 2 million pounds ($3 million) for salaries and expensive hardware clusters. 'AI' was a taboo term in early 2010s venture capital, with VCs operating on 5-10 year horizons, whereas AGI required decades. Peter Thiel, a cofounder of the Singularity Summit, became their crucial early investor. After an unconventional meeting where DeepMind's founders 'barged' into Thiel's party and Hassabis engaged him with a chess theory, Thiel's Founders Fund provided a large share of the requested funds despite his quip about investing in 'Somalia.' With funding, DeepMind attracted top global talent, pioneering reinforcement learning where models learn through trial and error, like playing a game and adjusting strategies based on points scored. They also secured Geoff Hinton and Yann LeCun as technical advisors. A subsequent 10 million pound ($15 million) B-round, led by Solina Chau, was successfully closed, with Skype cofounder Jaan Tallinn also investing.

A turning point occurred in 2012 with the 'cat classification paper,' featuring AlexNet, developed by Geoff Hinton and his University of Toronto PhD students. This deep learning model, trained on massive image volumes, demonstrated unprecedented accuracy in identifying over 1000 categories in the ImageNet challenge, proving that 'deep learning had arrived.' DeepMind's continuous need for compute power and talent led to a 40 million pound ($60 million) C-round in 2013, led by Elon Musk, despite Suleyman finding Musk's understanding 'high variance' and distrusting his competitive nature. Crucially, Musk showed Google cofounder Larry Page a demo of DeepMind's deep Q-network (DQN) learning to play the videogame Breakout, which, without strategic tips, surpassed human-level performance within hours. This breakthrough immediately captured Page's interest, leading him to seek a meeting with DeepMind's founders.

The tech industry was already in an 'AI arms race' for talent, with Google actively acquiring expertise, including the 'acquihire' of Hinton's DNNresearch for $44 million. Faced with constant fundraising, the high cost of talent (DeepMind engineers earned $100000 annually compared to two or three times that at larger companies), and the realization that building a Google-sized company while solving AI was untenable, DeepMind decided to sell. Despite a higher counteroffer from Facebook, DeepMind's founders chose Google for $650 million in early 2014. This decision was contingent on remaining in London, prohibiting the use of their technology for state surveillance or military purposes, and Google establishing an independent ethics and safety board—a non-negotiable priority for Suleyman that Facebook refused, reinforcing DeepMind's commitment to 'irrevocably attach' ethics to AI. Following the acquisition, Suleyman became Head of Applied AI, tasked with finding practical and profitable applications for their technology.