We are receiving more evidence that Microsoft is looking to split with OpenAI. A new report from The Information says Microsoft is developing its own in-house reasoning models to compete with OpenAI and has been testing models from Elon Musk’s xAI, Meta, and DeepSeek to replace ChatGPT in Copilot, its AI bot for the workplace.
The development would have been unheard of only a year or two ago as Microsoft plowed billions of dollars into the AI darling and inked a deal that seemed to tie the two together inextricably for years to come. But there have already been whispers in recent months that Microsoft has been interested in loosening its reliance on OpenAI, in part out of concern over ChatGPT’s price and speed.
Microsoft Copilot has thus far received poor reception in enterprises due to its high cost and limited results. It can be useful at completing some easier tasks, like placing prepared text into a slideshow. Limiting a chatbot to a fixed dataset can often yield better results than one searching the open web, but language models still create a lot of errors that require manual review and correction that can be easy to overlook when using a chatbot.
Microsoft, in January, decided to let OpenAI out of a contract that required it to use Azure for all its hosting needs. The startup recently announced an ambitious $500 billion plan (or $100 billion initially) to open massive new data centers with Oracle and SoftBank, a sign that Microsoft was unwilling to provide more resources.
One interesting wrinkle in the whole story is that, as part of its investment in OpenAI, Microsoft retains the right to use the startup’s intellectual property. The Information story says, however, that OpenAI has been unwilling to turn over documentation explaining how it built its o1 reasoning model:
Last fall, during a video call with senior leaders at OpenAI and Microsoft, [Mustafa] Suleyman—who leads Microsoft’s in-house artificial intelligence unit—wanted OpenAI staffers to explain how its latest model, o1, worked, according to someone present for the conversation and two other Microsoft employees who were briefed on it. He was peeved that OpenAI wasn’t providing Microsoft with documentation about how it had programmed o1 to think about users’ queries before answering them.
It seems that both sides understand that the two partners are quickly becoming competitors. Perhaps Microsoft CEO Satya Nadella overestimated how much control it would have over OpenAI. During an interview in late 2022, Nadella questioned why Microsoft would develop its own foundational models when it can just use those from OpenAI. It seems the tune has changed since OpenAI CEO Sam Altman appeared at the White House to announce his new infrastructure deal with Oracle.
Microsoft is reportedly planning on selling access to its in-house reasoning model, called MAI, to other developers, directly encroaching on OpenAI’s space.
It makes sense that Microsoft would want to detach from OpenAI. If they are going to remain independent companies, and artificial intelligence is going to be the next major platform shift, as some believe, Microsoft has an interest in controlling the technologies and writing its own destiny. OpenAI is currently a non-profit but has been working for some time to convert into a for-profit entity so it can raise more cash and presumably break free from obligations of a non-profit to operate solely in the public interest.
In a sense, Nadella and Microsoft are making strategic bets and strategically hedging in order to capture the value of the AI boom wherever possible. Maybe Microsoft’s models win; maybe OpenAI’s do. Or maybe DeepSeek, the open-source model from China, is the future of models, and all that matters is the application layer, where Microsoft dominates in enterprise. The company is clearly not going to take any chances and instead place a lot of chess pieces on the board.
Read the full article here
