Deepmind has been unable to keep up with its growing expenses since it was acquired by Google, making it something of a cash-burning machine. Now DeepMind’s profit growth, helped by Alphabet, Google’s parent, has brought new hope to itself and the artificial intelligence industry. But DeepMind’s murky revenue stream, high talent, and technology costs pose challenges for its further commercialization.
Recently, Google artificial intelligence laboratory DeepMind released the latest results, showing that it has made a profit in the fiscal year 2020, ending seven consecutive years of losses.
DeepMind’s revenues more than tripled to 826 million pounds ($1.13 billion) in 2020 and 265 million pounds ($361 million) in 2019, according to its financial reports. Its spending rose from # 717M ($976M) to # 780m ($1.06 bn) over the same period. The company ended the financial year with a profit of 44m pounds ($60m) , up from a loss of 477m pounds ($650m) in 2019.
At the same time, the report did not give much detail about DeepMind’s revenue stream, saying only that it had signed a service agreement with another group that provides research and development services, including the DeepMind Research and Development Initiative.
DeepMind’s customers are Alphabet and its subsidiaries and do not sell products or services directly to consumers or companies. The extent of its revenue surge is unclear.
The source of revenue puzzle, AI solution business model stuck
Deepmind’s main area of research is deep reinforcement learning, a branch of machine learning that has been used extensively in scientific research. Deepmind and other AI labs use deep reinforcement learning to perform complex games, train robots, predict protein structures, and simulate autopilot. DeepMind scientists believe that advances in reinforcement learning will also promote the development of general artificial intelligence.
It is worth noting that deep reinforcement learning is expensive to research and has limited commercial applications, compared with other deep learning systems such as image classifiers and speech recognition systems, deep reinforcement learning models often require training in the environment to be used and can not be directly ported and integrated into new applications, resulting in further technical and financial costs for the company.
In addition, DeepMind’s deep-learning focus does not translate directly into a profitable business model. Take, for example, the real-time strategy game Starcraft II mastered by the augmented learning system AlphaStar, a $one million project funded by Google with a lot of cloud computing resources, but this significant scientific value only exists during the duration of the project and is of little value in promoting the use of artificial intelligence.
At the same time, Alphabet is using DeepMind’s enhanced learning technology in some of its operating systems, such as power reduction technology in Google’s data center and related technology from Waymo, Alphabet’s autonomous driving company, but the technology may be used to outsource some AI tasks to DeepMind, rather than apply AI lab technology directly to the product. Meanwhile, one of DeepMind’s divisions is working on Google’s and Alphabet’s applied artificial intelligence projects, but this work is not directly related to the general artificial intelligence research being done by DeepMind Labs.
DeepMind’s payroll costs reached # 467M in 2020, about two-thirds of its total. The company has more than 1,000 employees, only a small fraction of whom are highly paid scientists, researchers and engineers.
In addition, DeepMind’s development relies heavily on Alphabet’s help. Alphabet wrote off Deepmind’s # 1.1 bn ($1.5 bn) debt in 2019 and helped it turn a profit in 2020. Whether Alphabet will be able to continue to help DeepMind in the long term remains to be seen, but if Deepmind loses Alphabet’s help, DeepMind will lose customers, money and fierce competition from the tech giants, its technology talent will be poached by other big companies to realise its enterprise value.
DeepMind, a cutting-edge artificial intelligence lab, has not been as profitable as the industry had hoped, but its 2020 report card shows it has turned a profit, it still offers new hope for an artificial intelligence industry that has struggled to commercialize.