The Gilded Cage: How Microsoft Almost Killed OpenAI's Enterprise Future

The Gilded Cage

7:00

I want to say something that might sound provocative but is actually just historically accurate: Microsoft has a gift for acquiring or partnering with the most promising things in tech and then slowly, methodically, making them worse.

Nokia. Skype. LinkedIn — still alive but institutionalized into corporate mediocrity. GitHub — functional, but ask any developer how they feel about the direction since the acquisition. Mixer — Microsoft's Twitch competitor that they launched, poured money into, signed exclusive deals with top streamers, and then shut down two years later with a three-paragraph blog post and a shrug.

They don't do it on purpose. That's almost the more terrifying part.

So when OpenAI's Chief Revenue Officer Denise Dresser sent a memo to staff last Sunday — a memo that was promptly leaked to CNBC, because of course it was — and wrote that their Microsoft partnership has been "foundational to our success, but has also limited our ability to meet enterprises where they are," I read that as a polite corporate way of saying: we almost let them do it to us too.

See the full CNBC breakdown of the leaked memo: OpenAI touts Amazon alliance, says Microsoft 'limited our ability'

What the Memo Actually Said

Let me be precise here, because the X posts are doing what X posts do — compressing nuance into hot takes. Dresser wasn't burning Microsoft to the ground. She was acknowledging a structural reality: Microsoft's Azure exclusivity meant OpenAI couldn't go where enterprise customers actually live, and increasingly, enterprise customers live on AWS Bedrock. Amazon invested $50 billion in OpenAI in February. Since the partnership was announced, Dresser says inbound demand from enterprises wanting OpenAI on Bedrock has been "frankly staggering." Staggering. From a CRO who used to run Slack and spent years at Salesforce. That's not hyperbole — that's a person who's seen enterprise sales cycles and knows what staggering actually looks like.

The Microsoft deal had to be renegotiated to even allow this. Think about that. OpenAI — the company that created ChatGPT, sparked the modern AI race, and is currently valued at $850 billion — had to ask permission to sell its own product to customers who weren't using Azure. That's the gilded cage in plain English.

The Anthropic Problem

Here's where the memo gets spicier. OpenAI is losing the enterprise race. Not losing badly, not losing catastrophically — but losing. Anthropic's Claude has become the default AI model for serious enterprise deployments. At the HumanX conference in San Francisco last week, Arvind Jain — CEO of enterprise AI startup Glean — described the dynamic in terms that stuck: "It has become a religion, that's the level of that mania."

Claude mania. In the enterprise. While OpenAI has been stuck behind an Azure tollbooth.

Dresser's response in the memo is to go after Anthropic on three fronts simultaneously: The brand attack: Anthropic's strategy is built on "fear, restriction, and the idea that a small group of elites should control AI." That's a deliberate reframing of Anthropic's safety-first positioning as paternalism. It's smart. It's also a little rich coming from the company whose board famously fired and rehired its own CEO in 48 hours, but we'll let that one slide.

The accounting attack: Anthropic announced $30 billion in annualized revenue. Dresser says that number is inflated by roughly $8 billion due to how they gross up revenue share deals with Amazon and Google. OpenAI reports the same type of deals net. It's a legitimate accounting distinction — and it's also the kind of thing you say publicly when you're getting your lunch eaten and need to put an asterisk on the scoreboard.

The compute attack: Anthropic made a "strategic misstep to not acquire enough compute." OpenAI says their own compute ramp is "materially ahead and widening." In the AI race, compute is infrastructure, and infrastructure is destiny.

The Microsoft Breakup That Isn't Quite a Breakup

To be fair to all parties — and I'm going to try, briefly — both Microsoft and OpenAI continue to describe their relationship as "core and strategic." And technically it is. Microsoft has $13 billion invested. Azure still runs a massive portion of OpenAI's workloads.

But consider the trajectory:

Mid-2024: Microsoft quietly added OpenAI to the list of competitors in its annual report. Not partners. Competitors.

2025: OpenAI started routing workloads to CoreWeave, Google, and Oracle — Microsoft's direct competitors. Microsoft started publicly testing its own in-house AI model. The partner is building a replacement.

Early 2026: OpenAI announces it's building a GitHub competitor after repeated outages frustrated their engineering teams.

Now: A CRO memo saying the partnership limited them, landing in the press within 24 hours of being sent.

That's not a partnership. That's a divorce in slow motion, with both parties still sharing the apartment and insisting to mutual friends that everything is fine.

What Actually Matters

The enterprise AI market is the real prize. Consumer AI is loud and visible — ChatGPT has 500 million users, everyone has an opinion about it. But enterprise is where the contracts are. The multi-year deals. The integrations so deep they become impossible to rip out. The revenue that builds companies into institutions. Claude got there first and built a following that one of its own critics is calling a religion.

OpenAI spent five years tethered to Microsoft's infrastructure while Anthropic was free to go wherever the enterprise customer was. Now OpenAI has $50 billion from Amazon, access to Bedrock, and a CRO who used to run one of the most widely deployed enterprise communication tools in history. The cage is open. The race is actually starting now. Whether OpenAI can break the Claude religion in enterprise before Anthropic closes its own IPO — that's the story of the next 18 months. And Microsoft? They'll probably be fine. They always are. It's just that everything they touch tends to be a little less fine afterward.