The most interesting development in AI this week isn’t a new model or a benchmark score. It’s Microsoft telling Wall Street it’s done playing second fiddle to OpenAI.
On Wednesday’s earnings call, Microsoft pitched its own homegrown AI models, tooling, and infrastructure as distinct products, not just wrappers around OpenAI’s tech. The shift is striking because Microsoft has poured over $13 billion into OpenAI since 2019. Now it’s openly competing with them.
The company highlighted its own Phi and Orca models, positioned Azure AI Foundry as a place to build AI apps (not just deploy GPT variants), and even teased a Mythos competitor. That last one matters because Mythos, if you’re not tracking it, is OpenAI’s attempt to build agentic coding tools that compete directly with GitHub Copilot, which Microsoft owns.
Microsoft’s pitch to investors is clear: we’re not just reselling someone else’s models. We’re building our own stack.
Microsoft’s earnings also revealed something else: its $4 billion investment in Anthropic is working out nicely. The company logged $3.2 billion in gains from that stake, which it acquired through a convertible note structure in 2024.
Meanwhile, OpenAI was “a mixed bag.” Microsoft didn’t elaborate, but the implication is obvious. OpenAI’s restructuring, leadership drama, and pivot to a for-profit model created uncertainty. Anthropic, by contrast, has been stable, shipped Claude Opus 5, and kept enterprise customers happy.
The Anthropic gains aren’t just financial. They gave Microsoft leverage. When you’re backing two competing labs, neither can strong-arm you on pricing or exclusivity. That’s useful when you’re also building your own models.
If you’re building on Azure OpenAI Service, nothing changes immediately. Microsoft still resells GPT models and will continue to do so. But the long-term signal is clear: don’t assume Microsoft’s roadmap is tied to OpenAI’s.
Azure AI Foundry is positioning itself as a multi-model platform. You can already deploy Llama, Mistral, Cohere, and Microsoft’s own models alongside OpenAI’s. That optionality matters if you’re planning infrastructure for the next two years.
For developers on OpenAI’s direct API, this is also a reminder that your biggest customer is now your biggest competitor. Microsoft has the data, the chips, the enterprise relationships, and increasingly, the models. OpenAI still has the brand and the research talent, but the gap is narrowing.
Mark Zuckerberg also chimed in on Wednesday, telling investors that Meta sees a “large enterprise opportunity” in AI that goes beyond agents. He name-checked APIs, compute, and internal software as revenue drivers.
That’s notable because Meta has mostly played the open-source card with Llama. Now they’re signaling they want to sell infrastructure and tooling to enterprises, not just give away model weights. If Meta, Microsoft, and Google are all pitching enterprise AI stacks, OpenAI’s moat gets narrower.
The pattern here is consolidation. The companies with hyperscale cloud infrastructure are realizing they don’t need to license models from startups when they can train their own. OpenAI and Anthropic are still ahead on capabilities, but the gap between frontier labs and big tech is shrinking fast.
Microsoft’s next move will likely be tighter integration between its own models and enterprise products. Expect Phi and Orca to show up in more Office and Dynamics workflows, especially for tasks where GPT-4 is overkill.
Also watch how OpenAI responds. They’ve been expanding into infrastructure (like the Mythos coding tools), but that puts them in direct competition with Microsoft’s core business. The partnership was always awkward. Now it’s openly adversarial.
For developers, the takeaway is simple: build for portability. If you’re locked into one provider’s API format or tooling, you’re making a bet on a relationship that’s increasingly unstable. Use abstraction layers, test against multiple models, and don’t assume today’s partnerships will hold.
The AI landscape just got more competitive. That’s probably good for everyone except the companies that thought they had a locked-in customer base.
One email at dawn. The five stories that mattered, with the bits removed and the meaning kept. Free, for now.