An Update on the Messy Microsoft/Open AI Breakup

Microsoft and OpenAI had an exclusive relationship for OpenAI to provide Chat GPT services through Microsoft Bing and Office 365. However, the relationship soured almost immediately. ChatGPT 5 launched a few months ago and was buggy and unreliable. Plus, OpenAI temporarily disabled access to ChatGPT 4, which worked. This broke applications that were using ChatGPT calls.

At the same time, OpenAI was going behind Microsoft's back to make direct deals with enterprise customers rather than let Microsoft do the selling. This is especially true as pertains to OpenAI cozying up to Oracle, a major competitor of Microsoft's. The most recent example was a $300 billion deal to provide generative AI services to Oracle Cloud Infrastructure. This helped boost the perception of OCI from also-ran to genuine competitor to Microsoft Azure.

It was also no secret that Microsoft was investing heavily in its own AI products. The Verge reported that Microsoft is making "significant investments" in its own models and has already made some modest models available.

Microsoft certainly has the resources to catch up and close the gap with OpenAI despite a multi-year lead, but it won't be addressed overnight. The fact is OpenAI does have a multiyear lead on Microsoft and there's just not enough resources to throw at the problem to close the gap expeditiously.

Now comes word that Microsoft has added Claude from Anthropic to its chatbot options alongside ChatGPT. Claude is different from other leading chatbots because of its extensive context windows, highly natural and expressive language and advanced ethical safeguards. This makes it especially well-suited for nuanced writing, creative exercises and coding tasks.

It is known for offering a conversational style and has the ability to maintain context over very long interactions.  It supports up to a million tokens per model, letting users process the equivalent of an entire book or a large code base in a single session, something other chatbots can't do. It has some shortcomings; it isn't as good at image generation as ChatGPT, for example. But for the business users Microsoft is targeting, it is a versatile service.

Add to that the buggy, problematic launch of ChatGPT 5 and you have the makings for a breakup, which appears to be what's happening. By partnering with Anthropic, Microsoft has a more reliable, stable partner for enterprise customers that is not likely to go around making deals behind its back and is relatively cheap if Microsoft wants to acquire it.

The question then becomes what does Microsoft do for generative AI? The Verge article made it abundantly clear Microsoft is well behind the leaders and they won't be closing the gap anytime soon. What's worse, this isn't even something within Microsoft's control. It needs the hardware to match its AI ambitions and that is not easing up any time soon.

Microsoft's preview model was trained on a cluster of 15,000 H100 GPUs. The H100 is Nvidia's prior generation and already out of date compared to the current generation of GPUs. You know things have gotten ridiculous when 15,000 GPUs from last generation is considered "tiny" in the grand scheme of things.

But if Microsoft wants to build a cluster of 40,000 GPUs using current Blackwell generation processors, it has to get online and take a number. Meta, xAI, Amazon and Oracle are all ahead of it with massive orders that have the TSMC factories in Taiwan running 24/7 and still not able to meet demand (and NVIDIA boss Jensen Huang tapping his fingers together like Mr. Burns, saying "eeeeeexcellent").

In between now and whenever Microsoft can get all the GPUs it needs, Anthropic will continue to advance Claude and hopefully stay ahead of the game. Microsoft may decide that the gap is too great to make up on its own and simply buy out Anthropic in the process.

All of this comes on top of the fact that there are Wall Street analysts starting to wave the flag of concern over an AI bubble, that things have gotten too out of hand, far too much is being spent and there has been no measurable return on investment. Some brave souls are even raising the specter of a 2000-like dot.com crash, knowing full well that they will be at the very least ignored.

One thing is for certain: the AI wars have only just begun and there is a lot ahead of us.

Posted by Andy Patrizio on 09/26/2025