Microsoft building new AI model to take on Google

Microsoft is reportedly building a new AI model designed to compete against powerful offerings from Google and Anthropic, but the exact purpose of the model has yet to be determined.

The Information reported the new model, dubbed MAI-1, could also potentially compete against OpenAI, which Microsoft has made a multibillion-dollar investment in.

A representative from Microsoft told Mobile World Live the company declined to comment on the reporting but pointed to a blog yesterday (6 May) by CTO Kevin Scott that stated it plans to continue to build large supercomputers for OpenAI well into the future.

“Our partner Open AI uses these supercomputers to train frontier-defining models; and then we both make these models available in products and services so that lots of people can benefit from them.  We rather like this arrangement,” Scott stated on LinkedIn. “There’s no end in sight to the increasing impact that our work together will have”.

The news site reported MAI-1 is being overseen by Mustafa Suleyman, a Google DeepMind co-founder and former CEO of AI start-up Inflection prior to Microsoft hiring him and his team in March.

The Information reported that while Microsoft paid $65 million for the rights to Inflection’s intellectual property, MAI-1 is separate from the models that Inflection released.

MAI-1 will be “far larger” than the previous smaller, open-source models Microsoft has trained, according to the news site, meaning it will require more data and computing power, which will make it more expensive.

It will have roughly 500 billion parameters while OpenAI’s GPT-4 has more than 1 trillion parameters, the report stated. Smaller open-source models released by companies such as Meta Platforms and Mistral have 70 billion parameters. Microsoft’s Phi-3 mini model released last month offers 3.8 billion parameters.