DeepMind, the A.I.-centric subsidiary of Alphabet, is bleeding money: last year, it lost $571 million, according to the Financial Times. If that wasn’t enough, it owes its corporate parent roughly $1.4 billion (although there’s apparently no risk of actual default).
What’s behind that intense cash burn? In 2018, DeepMind paid out salaries of $485 million, up from $242 million in 2017 and $127 million in 2016. LinkedIn says that the firm employs 838 people, which breaks down to $581,000 per head. And that’s before you consider the budget spent on hardware, services, and everything else that goes into cutting-edge artificial intelligence and machine-learning research.
In return, DeepMind has generated roughly $125 million in revenue, all of it from sales to Alphabet. DeepMind’s A.I. products end up integrated into YouTube, Google’s advertisement and cloud businesses, Waymo’s autonomous-driving platform, and so on. It’s also spent a good deal of time publicizing its work with deep learning, particularly in the context of gaming; its A.I. agents have gotten very good at beating humans in games such as “Go,” DOTA 2,” and “Quake III,” but that’s not the sort of thing that translates into massive profit.
“Mostly, DeepMind has been pursuing simulations, and games, where you can get quick successes and a lot of attention. There is definitely pressure on them to deliver at Google,” Pedro Domingos, an A.I. scientist who authored “The Master Algorithm,” told FT.
If that budgetary crunch wasn’t enough, there are new rumblings of internal issues. DeepMind co-founder Mustafa Suleyman was recently placed on leave for undisclosed reasons, according to Bloomberg. Not only does Suleyman oversee DeepMind’s applied division, but he’s also regarded as an ambassador for A.I., explaining its potential (and ethics) to the public and government officials.
Under normal circumstances, a substantial cash burn and the departure of a high-level executive would be enough to spark rumors of an imminent shutdown. That probably won’t happen, but DeepMind is nonetheless in a weird spot; as a subsidiary, it’s supposed to provide A.I.-related assets to Alphabet’s various companies and products—but Google Brain, Google’s in-house A.I. shop, already occupies a similar role within Alphabet’s ecosystem. Nor does Google seem inclined to allow DeepMind to market its products to other firms, which would potentially solve some of the revenue concerns; for the moment, the subsidiary is very much locked in its silo.
If more prominent executives and researchers leave DeepMind, then we can assume that the subsidiary’s problems are only deepening—especially if they migrate to a rival A.I. shop such as OpenAI. As it stands, it seems like DeepMind is at a very odd juncture: burning tons of cash, making some inroads into deep-learning research, but not living up to its massive early hype.