Microsoft Still Has the Leverage After $135B OpenAI Deal, Anthropic Avoids the Same Fate by Expanding Google Partnership
Microsoft and OpenAI make a deal, but the power dynamics remain unchanged. Anthropic partners with Google, ensuring they don't have the same problems with Amazon.
Microsoft & OpenAI Agree on For-Profit Structure, But Microsoft Still Holds the Keys
The tech world’s most consequential partnership just got a major rewrite. OpenAI transformed into a for-profit public benefit corporation, with Microsoft securing a 27% stake valued at approximately $135 billion.
Here are the details:
The OpenAI Foundation will hold 26% of the for-profit entity
Microsoft will hold 27%
Employees and other investors will hold the remaining 47%
Microsoft will have access to OpenAI technology until 2032, including models deemed as being AGI, but not for consumer hardware.
OpenAI can develop products with 3rd parties.
Microsoft will still be entitled to 20% of OpenAI revenue until an expert panel deems OpenAI has reached AGI.
Microsoft gives up its right of first refusal on new OpenAI cloud infrastructure.
OpenAI will make an additional $250 billion commitment to use Azure infrastructure.
This probably looks like a huge win for Microsoft, and many are saying as much on social media, but remember they held all of the cards in this negotiation.
And rightfully so! GPT-3 was trained using Microsoft money and servers. ChatGPT was originally released on top of GPT-3.5 and runs on Microsoft servers. GPT-4 and all other models were trained using Microsoft’s money and servers.
If you back out Microsoft’s money and infrastructure, all OpenAI has is GPT-2, which is worthless today.
And more importantly, I think this new partnership can still get contentious. Imagine if OpenAI tried buying Cursor, like they did with Windsurf. According to the new agreement, Microsoft still gets access to that IP, which would be value destructive for OpenAI. This ultimately sunk the Windsurf deal, and you could imagine any deal outside of consumer hardware would have the same issues.
Also, don’t expect OpenAI to spend $250 billion on Azure cloud services tomorrow, since they have only raised $78 billion so far.
So yes, this deal lets OpenAI become a for-profit company, but it doesn’t wipe the slate clean between Microsoft and OpenAI.
Anthropic Bets Tens of Billions on Google TPUs to Avoid OpenAI’s Fate
While Microsoft and OpenAI were finalizing their new agreement, Anthropic struck a massive deal with Google for access to up to one million tensor processing units (TPUs), valued in the tens of billions of dollars.
This arrangement is massive, but also not knew. We wrote about Anthropic being one of the first companies partnering with Google’s TPUs back in 2023, and explained why this partnership could be a massive win-win.
Google’s TPUs could be a competitive advantage in the AI race. They can work quicker than GPUs while using fewer resources, and it makes them less dependent on GPU powerhouse Nvidia.
Now Anthropic is going much bigger, deploying more than a gigawatt of computing power by 2026. Industry experts believe the average cost of that much computing power at around $50 billion per year.
For Google, this reiterates that their TPUs are a credible, large-scale alternative to Nvidia’s dominant GPUs. Analysts have suggested that if Google were to spin off the TPU unit and its AI research organization DeepMind, those two entities could be worth as much as $900 billion.
Anthropic keeps winning the enterprise, and they’re building a diversified compute strategy across Google’s TPUs, Amazon’s Trainium chips, and Nvidia’s GPUs. In other words, they are commoditizing their compliments just like OpenAI.
The real question is what this move means for their AWS partnership, as Amazon has been their primary backer. My guess is that their accelerating revenue has emboldened Anthropic, knowing they can raise from venture without relying Amazon’s cloud credits.
Essentially, Anthropic is ensuring they don’t end up in a relationship with Amazon like OpenAI has with Microsoft.