AI is making some impressively tantalising leaps forward, but the cost of developing such advanced technology isn’t cheap. That has been clearly outlined by Google sinking an additional £1.6 billion in capital into Deepmind, its UK-based AI business. The company racked up losses of £477 million over the course of 2020, with an additional £1.1 billion expense booked by Google through the write-off of a £1.1 billion loan extended by the parent company.
Google won’t be too worried, given the depth of its pockets, and is clearly committed to ongoing investment in Deepmind with a long-term view. Last month a Deepmind algorithm was responsible for one of the most significant scientific breakthroughs in years when it learned how to accurately visualise the structure of proteins. It’s been one of biology’s biggest challenges.
The London-based Deepmind lab has developed some of the most AI tools in the world in recent years. That’s acting as a draw to some of the world’s leading scientists and mathematicians, in a virtuous loop further accelerating the pace of new breakthroughs. Google uses Deepmind’s AI tools in its core business, with improving its voice assistant, search engine and digital advertising platforms all benefiting.
Deepmind’s revenue currently mainly comes from its sister companies within Alphabet, the holding company that also owns Google. They pay it fees to use its intellectual property, which rose to £266 million in 2019, from £103 million in 2018. However, turnover is still much smaller than expenses, with a high wage bill another reason why the company has managed to attract so many leading specialists.
Deepmind is only ten years old, having been founded by former chess prodigy turned games developer Demis Hassabis, now the company’s chief executive, alongside Shane Legg and Mustafa Suleyman in 2010. Google acquired Deepmind in 2014 for a price reported at around £400 million, and has since ploughed money into funding and extending its research.
From its UK base, Deepmind employs around 1000 people worldwide, some of whom earn seven-figure salaries. But with Apple, Microsoft, Amazon and Facebook also competing for the world’s brightest research scientists, the best-of-the-best don’t come cheap. Deepmind’s total bill for salaries, plus bonuses and expenses, came in at £468 million in 2019, up from £398 million a year earlier.
As well as the “once-in-a-generation advance” of being able to compute which shapes proteins fold into, Deepmind has also produced software able to beat the world’s top human players of the board game Go.
In its most recently published accounts, Deepmind warned that the machine learning sector is one constantly changing and is highly competitive. Those factors, it cautioned, could have a “significant impact”, on its future financial success. But with a backer as wealthy as Alphabet committed to continuing to fund its research, few would bet against it eventually monetising its research and becoming a hugely valuable company in its own right.