Argues this isn't a breakup but a renegotiation from strength. Microsoft now runs multiple model families through Azure — GPT, Llama, Mistral, Phi, and hundreds of open-weight alternatives — so the original revenue-sharing structure that locked in OpenAI as the sole frontier supplier no longer reflects the competitive landscape.
Notes that unlike typical per-model or per-token licensing, Microsoft gave OpenAI an estimated 20-30% cut of the entire Azure OpenAI Service revenue — meaning every enterprise GPT customer generated revenue for both companies. This arrangement was always anomalous and reflected a moment when Microsoft needed to lock in the only credible frontier model provider at any cost.
Reports that OpenAI has been restructuring from its capped-profit model toward a traditional for-profit entity, and Microsoft negotiated modified terms as part of that process. The end of revenue sharing appears to be one of the concessions that emerged from those restructuring negotiations.
Microsoft is ending its revenue-sharing arrangement with OpenAI, according to Bloomberg reporting on April 27. Under the existing deal — forged during the roughly $13 billion investment Microsoft poured into OpenAI starting in 2019 — Microsoft shared a portion of Azure AI services revenue with OpenAI as compensation for exclusive access to the lab's models. That arrangement is now being unwound, marking the most significant structural change to the partnership since it began.
The revenue-sharing mechanism was always unusual. Unlike a typical licensing deal where you pay per-model or per-token, Microsoft's agreement effectively gave OpenAI a cut of the broader Azure AI pie — meaning every enterprise customer running GPT-4, GPT-4o, or any OpenAI model through Azure was generating revenue for both companies. The exact percentages were never publicly disclosed, but estimates from analysts put OpenAI's share in the range of 20-30% of Azure OpenAI Service revenue, which Microsoft's cloud division has been reporting as a major growth driver.
The timing is notable. OpenAI has been in the middle of a complex restructuring from its original capped-profit model toward a more traditional for-profit entity. Microsoft reportedly negotiated for modified terms as part of that restructuring, and the end of revenue sharing appears to be one of the concessions that emerged.
This isn't a breakup — it's a power shift. Microsoft isn't walking away from OpenAI's models. It's walking away from a financial structure that made sense when OpenAI was the only credible supplier of frontier models and Microsoft needed to lock them in at any cost. That calculus has changed dramatically.
Microsoft now runs multiple model families through Azure: OpenAI's GPT series, Meta's Llama models, Mistral, Phi (Microsoft's own small models), and a growing catalogue of open-weight alternatives. The Azure AI Model Catalog lists hundreds of options. When you're a platform with model diversity, paying a premium revenue share to a single supplier looks less like a strategic investment and more like an expensive legacy contract.
The financial implications are substantial. Morgan Stanley estimated Azure OpenAI Service revenue at roughly $4-5 billion annualized by early 2026. If Microsoft was sharing even 20% of that, the end of revenue sharing represents $800 million to $1 billion in annual savings — or margin improvement that drops straight to the cloud division's bottom line. For a company fighting to show Wall Street that AI spending generates returns, recapturing that margin matters enormously.
For OpenAI, the picture is more complicated. The company has been burning cash at a staggering rate — reportedly over $5 billion annually on compute alone — while racing to build out its own consumer and enterprise products (ChatGPT, the API, enterprise plans). Losing a recurring revenue stream from Azure customers doesn't kill OpenAI, but it increases the pressure on its direct sales channels to compensate. OpenAI's own API revenue and ChatGPT subscriptions need to grow even faster now.
The Hacker News community response (808 points) reflects the significance. The top-line reaction splits into two camps: those who see this as Microsoft asserting dominance over a partner it no longer needs to subsidize, and those who see it as the natural maturation of a relationship where OpenAI was always going to need to stand on its own commercial feet.
If you're building on Azure OpenAI Service, the immediate impact is likely positive. Microsoft recapturing that revenue share creates room for price reductions on Azure AI endpoints — and Microsoft has been under competitive pressure from Google Cloud and AWS to lower inference costs. Don't expect an announcement tomorrow, but watch for Azure AI pricing adjustments in the next two quarters.
More strategically, this accelerates the trend of Azure positioning itself as model-agnostic. Microsoft has less financial incentive to steer customers toward OpenAI models specifically and more incentive to optimize for whatever model best serves the workload — including its own Phi family and open-weight alternatives. If you've been building exclusively against the `gpt-4o` endpoint, now is the time to abstract your model calls behind a routing layer that can swap providers.
For teams evaluating their AI vendor strategy, the lesson is clear: platform owners always recapture value from their suppliers eventually. The companies that built moats by being the exclusive model provider to a hyperscaler are discovering that exclusivity is a depreciating asset. Google invested in Anthropic, Amazon invested in Anthropic, Microsoft invested in OpenAI — and all three are simultaneously building in-house capabilities and diversifying their model offerings. The hyperscalers want optionality, not dependence.
If you're on OpenAI's direct API, this doesn't change your costs today. But it does change the competitive dynamics. OpenAI now has stronger incentive to differentiate its direct API offering from what's available through Azure — expect features, rate limits, or early access windows that favor direct customers over Azure-mediated ones.
The Microsoft-OpenAI partnership was the defining deal of the AI boom's first wave. Its restructuring marks the transition to the second wave, where the question shifts from "who has the best model" to "who controls the distribution." Microsoft's answer is unambiguous: we do, and we're done paying rent for the privilege. The next 12 months will reveal whether OpenAI can build a commercial engine that thrives without the Azure revenue backstop — and whether the model-as-a-moat thesis that justified $100B+ valuations for AI labs still holds when the platforms decide they'd rather own the models themselves.
This agreement feels so friendly towards OpenAI that it's not obvious to me why Microsoft accepted this. I guess Microsoft just realized that the previous agreement was kneecapping OpenAI so much that the investment was at risk, especially with serious competition now coming from Anthropic?
This gives OpenAI the ability to goto AWS instead of exclusively on Azure. I guess Azure really is hanging on by a thread.https://news.ycombinator.com/item?id=47616242
Nadella had OpenAI by the short and curlies early on. But all I've seen from him in the last couple of years is continuously acquiescing to OpenAI's demands. I wonder why he's so weak and doesn't exert more control over the situation? At one point Microsoft owned 49% of OpenAI bu
A wise man from Google said in an internal memo to the tune of: "We do not have any moat neither does anyone else."Deepseek v4 is good enough, really really good given the price it is offered at.PS: Just to be clear - even the most expensive AI models are unreliable, would make stupid mist
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.
Opinions are my own.I think the biggest winner of this might be Google. Virtually all the frontier AI labs use TPU. The only one that doesn't use TPU is OpenAI due to the exclusive deal with Microsoft. Given the newly launched Gen 8 TPU this month, it's likely OpenAI will contemplate using