Hundred-billion dollar redemption OpenAI backstabs Microsoft

Hard work supporting my girlfriend through college, only for her to break up immediately after graduation, claiming she wants freedom…

How does that feel?

Microsoft knows.

Yesterday, the long-standing OpenAI and Microsoft joint announcement: revising the contract!

OpenAI finally gained the freedom they’ve been longing for.

The cost is, the funding they once received must be repaid tenfold.

The Root of the Contradiction

Rewind to 2019.

At that time, OpenAI was still an idealistic lab surviving on donations, barely able to support training GPT-3.

Microsoft’s situation wasn’t much better; Azure cloud services couldn’t compete with AWS, Bing search was hammered by Google, and they desperately needed a technological breakthrough to turn things around.

In July of that year, Microsoft officially invested $1 billion in OpenAI, marking their first collaboration: Microsoft provided Azure computing support, and OpenAI used GPT technology to help develop products like GitHub Copilot.

To outsiders, Microsoft was not only funding but also providing computing power—practically becoming OpenAI’s surrogate parent.

But for Microsoft, like many other venture investments, this was just an experiment, not expecting significant returns.

So, Microsoft even agreed to some rather outrageous conditions.

According to the details leaked by The Information, one such clause stated: when OpenAI’s non-profit board determines that AI has evolved into AGI, they have the right to immediately cut off Microsoft’s access to the technology.

This is what later became known as the “AGI clause.”

From today’s perspective, did Microsoft’s legal team back then drink fake alcohol to agree to such a condition?

But seven years ago, the so-called large models were still very primitive, and AGI was a sci-fi fantasy.

Even if it could be achieved, it was a distant future matter.

It was a grand promise, mainly used for hype at the time.

No one expected that just three years later, the sudden evolution of ChatGPT at the end of 2022 would completely change the balance of power.

OpenAI was no longer just a minor player at Microsoft’s mercy; it suddenly possessed enough influence to sway the balance.

Inevitably, cracks appeared.

Over the past few years, while OpenAI brought huge benefits to Microsoft, the “Ultraman” kept publicly hinting on social platforms, “We’re close to AGI,” “AGI could be achieved at any time”…

Showcasing technological progress was secondary; more importantly, it was about demonstrating independence.

Microsoft naturally was very unhappy with this behavior.

Because of this inexplicable clause, what could have been one of the most successful venture investments in human history turned into a ticking time bomb.

What exactly counts as AGI?

According to leaked documents from 2024: once OpenAI develops an AI system capable of generating a total profit of $100 billion for early investors (including Microsoft), AGI is considered achieved.

By October 2025, after OpenAI completed its profit-oriented restructuring, the situation became more concrete.

Post-restructuring, Microsoft held about 27% of OpenAI’s shares, valued at approximately $135 billion based on the valuation at that time.

When an investment balloons to the hundreds of billions level, this bomb called AGI must be dismantled.

The Cost of Freedom

After weeks of closed-door negotiations, both sides finally eliminated legal risks and yesterday released a brand-new agreement.

The core summary: disentangle technology bets, lock in long-term cash flow, reconstruct distribution channels.

1. Abolish the AGI clause.

Previously, the contract stipulated that AGI was to be determined solely by OpenAI, but in reality, Microsoft would never allow that.

It’s not about what you think; it’s about what I think.

Whether AGI is achieved must be verified by an “independent expert panel” to be valid.

And by the time this panel sorts out what constitutes AGI, Elon Musk’s rockets might already be on Mars—completely pointless for OpenAI, eager to commercialize.

This effectively nullifies the so-called AGI clause.

Since that’s the case, why not both sides take a step back, decouple the rights change from the vague “technology milestones,” and simply set a fixed date?

Microsoft retains a non-exclusive license to OpenAI’s models and products until 2032.

Research-level intellectual property rights will be valid until the expert panel verifies AGI or 2030, whichever comes first.

2. Microsoft no longer invests money; OpenAI single-handedly contributes.

In the old agreement, both sides charged each other.

Microsoft sold OpenAI services on Azure and took 20% of the profits; OpenAI, in turn, had to give 20% of its total revenue to Microsoft.

But under the new agreement, Microsoft will no longer pay OpenAI resale revenue shares from Azure; all profits go into their own pocket.

What about OpenAI?

OpenAI still needs to pay Microsoft the same revenue share—this “protection fee” until 2030.

Meanwhile, OpenAI also commits to purchasing $250 billion worth of Azure cloud services from Microsoft.

This is clearly an “unequal treaty.”

Why did Ultraman agree? Did he also drink fake alcohol?

Mainly because of the next point.

3. Give up “exclusive sales rights.”

OpenAI can now sell all its products on any cloud platform.

Although Microsoft retains the title of “primary cloud partner,” and new products will be launched first on Azure, this is just a face-saving gesture.

The agreement is cleverly written: unless Microsoft cannot or chooses not to support the required capabilities, OpenAI can go elsewhere.

So, immediately after yesterday’s announcement, Amazon CEO posted on X, announcing that AWS will provide all OpenAI models to customers via Bedrock service “in the coming weeks.”

Clearly, this is a big deal in the B2B market.

Enterprise clients hate being “locked in.”

OpenAI’s multi-cloud deployment allows B2B customers, especially government contracts requiring strict data security, to choose where to deploy models freely.

In summary, for OpenAI, the new agreement is overall more beneficial.

Although burdened with heavy financial obligations, needing to keep supplying Microsoft for the next five years,

its commercialization channels are more flexible.

Based on AWS enterprise customer scale estimates, after partnering with AWS, OpenAI’s enterprise customer count is expected to grow by 50% by the end of 2026, from 120k to 180k.

More importantly, investors no longer need to worry about “Microsoft chasing accountability after achieving AGI,” nor about “delayed AGI and unfulfilled investment returns.”

It eliminates the biggest uncertainty for IPO.

And what about Microsoft?

Microsoft has also shed many unnecessary burdens.

This time, both sides agreed to rewrite the agreement, not just because of the AGI clause.

First, external pressure.

Amazon AWS has long envied Microsoft’s monopoly in AI cloud services.

In February this year, Amazon launched a fierce attack on OpenAI, promising to invest up to $50 billion, planning to expand their existing $38 billion AWS agreement to $100 billion.

Meanwhile, AWS will also become the “exclusive third-party cloud distributor” for OpenAI’s newly launched enterprise platform Frontier in early April.

This clearly infringes on Microsoft’s exclusive rights, even considering a lawsuit for breach of contract.

But, as in past lawsuits, both sides would likely suffer losses; better to settle privately.

Of course, more importantly, is compute power! Compute power! Compute power!

Training large models is essentially burning money.

In OpenAI’s view, although Microsoft limits their freedom, its core value as a “compute cash machine” cannot be denied.

Microsoft is frantically acquiring land worldwide to build data centers and secure power quotas, yet still can’t satisfy OpenAI’s appetite—an endless pit.

Moreover, the other side can never be truly loyal; this money spent is simply not worth it.

So, why not let OpenAI go to other clouds to serve clients? Not only can this ease the infrastructure pressure on Azure, but it also allows OpenAI to find more big spenders in the market to share the infrastructure costs.

At least until the contract expires, no matter how much OpenAI earns on AWS, as long as OpenAI’s valuation rises, Microsoft’s reported investment gains will also increase.

This also ensures that their core products like Office Copilot and Windows OS won’t face a “mindless” technical supply cutoff crisis.

Lying back and earning passive income—what’s wrong with that?

As for the future, while OpenAI flirts with AWS, Microsoft is secretly working with Anthropic, even using the Claude model to drive some of its Agent products.

These two have long been scheming, each other’s backup plans.

The current outcome is quite favorable.

Epilogue

This event is not just about the titans splitting and merging.

Once, AGI was revered as a silicon deity, even written into contracts as a sacred doctrine.

But now, its most devout followers have personally stripped it of its “sacredness.”

It has become something that can be defined by an “independent expert panel,” a set of financial caps, intellectual property years (2032), and revenue sharing deadlines (2030), a product that must be distributed through multiple channels to earn a meager profit…

After three years of rapid progress, the entire AI industry has shed its early idealism and entered the deep waters of commercialization.

And yesterday, it officially marked a new chapter filled with pragmatic realism.