Will OpenAI Risk a Legal Fight With Microsoft?

$13 Billion Alliance Can Be Fickle

It is a weird feature of the modern artificial intelligence industry that a startup can be funded to the tune of $13.75 billion by a trillion-dollar behemoth and then, a few years later, reportedly consider suing that same behemoth for anticompetitive behavior. It is like complaining that the person who bought you a private jet now owns the only airport you can land at. But this is where we are with OpenAI and Microsoft, a relationship that started as a beautiful symbiosis and has now morphed into a very public, very high-stakes sibling rivalry.

The latest news, via the Wall Street Journal, is that OpenAI executives have weighed lodging an antitrust complaint against their most important partner, a move that would have been unthinkable just a short while ago. But things have changed. What was once a simple arrangement—OpenAI had the brains, Microsoft had the data centers and the checkbook—has become much more complicated.

What caused the first major cracks? Well, success, of course. After ChatGPT exploded in November 2022, Microsoft found itself in the slightly awkward position of playing catch-up to its own investment. Then came the "November OpenAI sort of situation," that infamous weekend where OpenAI’s board fired its CEO Sam Altman, Microsoft hired him, and then he was un-hired from Microsoft and re-hired at OpenAI after a staff revolt. That sort of thing can strain a relationship.

Since then, both sides have been building fortresses. Microsoft has its own in-house AI models (called MAI) and poached most of the team from AI startup Inflection. OpenAI, for its part, is cozying up to Microsoft's biggest rivals, announcing a massive $500 billion "Stargate" data center project with Oracle and, more recently, tapping Google for its cloud infrastructure.

The core of Microsoft’s strategy seems to be that if AI models are going to become a commodity, it wants to be the supermarket that sells all of them. Here is Bloomberg on CEO Satya Nadella’s thinking:

Nadella’s primary allegiance now isn’t to OpenAI’s very expensive skunkworks. His ultimate objective is to sell whatever AI his customers might want through Microsoft’s platforms. Nadella had spent three years running various parts of the company’s cloud business, called Azure, before he became CEO in 2014, and that’s now central to his AI strategy. Customers can choose from over 1,900 different models on Azure, including ones made by Meta and OpenAI, and upstarts such as Cohere, Mistral, Stability AI and now DeepSeek. … Whether a model’s usage costs a customer $10 through OpenAI or 90¢ via DeepSeek, Microsoft gets paid for the cloud computing, cybersecurity protections, data storage and other upsold services.

Meanwhile, OpenAI is trying to restructure itself into a for-profit entity, but the negotiations are reportedly stuck. OpenAI wants to reduce Microsoft’s future cut of the profits and get rid of the exclusive cloud hosting deal. Microsoft, having put in the money, is presumably less enthusiastic about these new terms. The threat of an antitrust complaint looks a lot like a negotiating tactic to get Microsoft to agree.

It’s a strange way to run a partnership, but maybe it’s the only way to run an AI company. You need the scale and capital of a Microsoft, but you also need the focus and agility of an OpenAI. The problem is that eventually, the startup you nurtured grows up to become your biggest competitor. And then things get messy.

Apple's Intelligence

There is a simple rule about Apple, which is that Apple controls everything. Apple makes the hardware, Apple makes the software, Apple runs the App Store. It is a beautiful, seamless, and fantastically profitable walled garden, and the primary rule of the garden is that Apple is the gardener.

So it is a little weird that the company is now integrating its biggest rivals directly into its core products. To power the more advanced AI features for its devices in the West, Apple is relying on OpenAI's ChatGPT. And in China, it appears a deal with Alibaba is imminent after the tech giant announced this week that its Qwen3 AI model is now compatible with Apple’s own MLX software framework. [1]

This is a little like a Michelin-starred chef, famous for a secret sauce he’s perfected for decades, suddenly announcing that while he'll still make his own sauce, the most complex dishes on the menu will be finished with a drizzle of sauce from the powerhouse restaurant across the street.

Perhaps, for Apple, this is a sign of desperation. The company is so far behind in the generative AI race that it has no choice but to partner with its rivals. Its own efforts have been slow and plagued by engineering problems. There was the iPhone 16 ad featuring a powerful new Siri feature that, embarrassingly, never actually shipped. And the full, ambitious overhaul of Siri itself has now been delayed until at least the spring of 2026.

Here’s Bloomberg on Apple's struggle:

“This is a crisis,” says a senior member of Apple’s AI team. A different team member compares the effort to a foundering ship: “It’s been sinking for a long time.”

... What’s notable about artificial intelligence is that Apple has devoted considerable resources to the technology and has little to show for it. The company has long had far fewer AI-focused employees than its competitors, according to executives at Apple and elsewhere.

This is all bad. Apple’s traditional strength has been in heavily integrated, non-generative AI—things like Face ID or predictive traffic alerts. But it has stumbled badly on the new generative AI that consumers now expect. To plug the capability gap, Apple is doing two things at once.

First, it is repeating the playbook that made the iPhone a hit: opening up its on-device models to its vast community of developers, hoping they will invent the "killer app" for AI that Apple itself has so far failed to produce. Second, it is finding partners for the heavy lifting.

But, in China, you can’t just plug OpenAI’s ChatGPT into iPhones and call it a day. The Chinese government has very specific ideas about what an AI model is allowed to know and say. Foreign models are not welcome. So if you are Apple, you need a Chinese partner with a powerful, locally-developed model that is already compliant with Beijing's rules. Enter Alibaba. Its Qwen models are competitive with the best Western AIs, and they are, crucially, government-approved.

For a company that built an empire on the premise that it alone knows best, this is a profound admission. The company's intelligence, for now, will not be built entirely in Cupertino.

[1] MLX is Apple's own open-source machine learning framework, optimized to run efficiently on Apple Silicon. By making its AI model compatible, Alibaba is essentially creating a version of its software that is tailor-made to run on Apple's latest hardware.

The Scoreboard

• AI: Why Is Microsoft-OpenAI Partnership Cracking? (ARPU)
• Semiconductor: AWS’ custom chip strategy is showing results, and cutting into Nvidia’s AI dominance (CNBC)
• Social Media: Trump to extend TikTok deadline for third time, pushing decision out another 90 days (CNBC)
• E-commerce: Amazon's corporate workforce may shrink as AI takes over routine tasks (Reuters)