Microsoft has announced it will invest $1 billion in San Francisco–based research group OpenAI. OpenAI said the investment and collaboration with Microsoft would help the two organizations pursue Artificial General Intelligence (AGI), development of an Azure AI computing platform, and an agreement to port OpenAI’s services to run on Azure.
The partnership is intriguing given that Microsoft already has one of the largest (in terms of employees and spending) and most sophisticated AI-related research units. Recent reports suggest that Microsoft now employs some 8,000 AI researchers and computer scientists — so it is perhaps surprising that Microsoft would want to enlist the help of OpenAI. It is also surprising that this project will focus on developing a computer platform given OpenAI’s research focus. OpenAI has published papers about scaling AI training, but very little about computer platforms themselves.
The agreement suggests that OpenAI software solutions will also be delivered across the Azure platform. OpenAI has an impressive portfolio of open source technologies such as AI-Gym (used for training reinforcement learning algorithms). AI-Gym could complement the reinforcement learning systems Microsoft gained through its acquisition of Bonsai. The agreement opens the door to the possible launch of a commercial platform that leverages both organizations’ reinforcement learning technologies.
The partnership will include an agreement that Azure will be the preferred host of OpenAI’s software services. This is significant given that OpenAI currently doesn’t have any commercial services, and suggests that OpenAI will be bringing a new technology to the market shortly. OpenAI has not given any indication as to what services it could be launching.
Many media outlets in their reporting of the partnership paid significant attention to the mention of AGI in the announcement. These discussions on AGI are very premature, given the current state of the technology, and provided a useful distraction from the real motivation of the partnership, which will be to develop a computing platform and software services. If Microsoft and OpenAI can develop a computing platform that can process data on orders of magnitude ahead of what is possible today, this would be a useful step in the development of AGI.
Microsoft’s obvious rival here is Google, both as a competitor in terms of emerging AI technology and in the cloud infrastructure market. Google already has its own AI computer platform in the form of the TPU, of which it is already on its third iteration. The TPU is used widely by Google’s research teams, Google Brain, and DeepMind. The TPU is also deeply integrated and optimized for the AI framework TensorFlow, giving Google strong hardware and software firepower, which Microsoft isn’t currently matching. Developers can also now access the TPU through Google Cloud, creating an additional future revenue opportunity from the companies’ investment in the platform.
Microsoft appears to be looking to replicate the same strategy as Google, leveraging the skills of OpenAI to help develop an equivalent computer platform to the TPU. The news that Microsoft is looking to develop its own AI supercomputing platform will be particularly concerning for Nvidia and Intel; as more cloud computing giants become vendors of hardware, their dominance as the market’s leading providers of AI-related cloud computing platforms could be seriously challenged.
Elements of the political media have been pushing for greater strategic investment from U.S. organizations to counter the rise of China’s growing AI sector and capabilities. This partnership, however, is much more related to the commercial interests of Microsoft than anything geopolitical. OpenAI has made commitments in the past not to involve itself in any military projects, and says it is focused on developing “advanced digital intelligence in the way that is most likely to benefit humanity as a whole.” Both OpenAI CEO Sam Altman and Microsoft CEO Satya Nadella have come out to stress that this joint venture will remain within the constraints of an ethical framework.