Author: Ada, Shenchao TechFlow

Recently, Microsoft and OpenAI jointly announced a revision of their cooperation agreement. The exclusive cloud restrictions have been lifted, IP licensing has been reduced to non-exclusive, and the AGI escape clause has been removed.

After the news broke, almost all Chinese media were asking the same question: Who won? But this is not the core of the issue.

This "divorce" truly buries the competitive logic of an era in the entire AI industry: whoever binds the best model wins.

In the new game rules, the stakes have shifted from models to something else.

Models are no longer scarce

First, let's look at a set of numbers.

OpenAI currently discloses a total infrastructure commitment of $680 billion, having signed $250 billion with Microsoft Azure, $300 billion with Oracle's Stargate project, and $138 billion with Amazon AWS (including the original $38 billion plus an additional $100 billion, over 8 years).

Combined, this exceeds $680 billion, while OpenAI's annual revenue is only about $25 billion.

A company with an annual revenue of $25 billion has signed a computing bill exceeding $680 billion. OpenAI has sold itself to computing power suppliers, making it a key customer for the three major cloud providers.

Anthropic is in a similar situation. Last week, it just signed an expanded cooperation with Amazon, committing to spend over $100 billion on AWS over the next decade in exchange for 5 gigawatts of computing power. Four days later, it signed a 3.5 gigawatt TPU capacity agreement with Google and Broadcom, expected to be launched in 2027. Coupled with Google’s announcement last week of an investment of up to $40 billion, Anthropic is now simultaneously locked in by two major cloud giants.

Both leading AI companies are exchanging their futures for computing power.

Looking back, what did Microsoft invest $1 billion in OpenAI for in 2019?

It was for the exclusive distribution rights for the model. Azure had exclusivity over the GPT series; other cloud providers' customers wanting to use OpenAI's models? Sorry, move to Azure.

That era was when models were scarce. GPT was the only significant large language model available; whoever owned it had pricing power.

But the reality in 2026 is: models are no longer scarce.

Anthropic's Claude, Google's Gemini, Meta's open-sourced Llama—all run on multiple cloud platforms. Data from Ramp’s enterprise spending shows that 79% of enterprises paying to use Anthropic are also paying to use OpenAI. Enterprise clients don't want to be locked into one platform.

OpenAI has recognized this internally as well. Chief Revenue Officer Denise Dresser wrote clearly in a March internal memo: "The partnership with Microsoft laid our foundation, but also limited our ability to meet the actual needs of enterprise clients."

In other words, exclusive binding was once an advantage; now it is a shackle.

The model layer is rapidly commodifying. When all mainstream models can run on all mainstream clouds, the value of exclusive distribution rights for models approaches zero.

So what is increasing in value? Computing power.

The data makes this clear. Amazon has poured hundreds of billions of dollars into both OpenAI and Anthropic in just two months. Google invested $40 billion in Anthropic while continuing to invest in its own Gemini. Microsoft loosened its grip on OpenAI while allowing Mustafa Suleyman to lead the development of independent superintelligence research.

At the core of each transaction are computing power, chips, and data centers. Models have become a side product.

Electricity is the new oil

Returning to the revised agreement between Microsoft and OpenAI.

On the surface, OpenAI gained freedom to sell models on AWS and Google Cloud. Microsoft lost its exclusivity, but retained a 27% equity stake and non-exclusive IP licensing until 2032.

Exclusive to non-exclusive sounds like OpenAI won, but the $250 billion Azure procurement commitment remains, and OpenAI products still launch on Azure first, unless Microsoft chooses not to support them. This hasn't changed. This isn't uncoupling; it's swapping one chain for a pipe. Previously it relied on contracts, now it relies on infrastructure to lock you in.

OpenAI's current situation is that it has simultaneously signed computing contracts worth $250 billion with Azure, $138 billion with AWS, and $300 billion with Oracle. Each contract is long term, tied to specific chip architectures and deployment plans. Technically, it has achieved "multi-cloud freedom," but financially, it is locked in by three cloud providers. This looks more like changing from one landlord to three landlords.

Zooming out further.

In 2023, ChatGPT emerged, and everyone was saying: models are the new oil. Whoever controls the best model controls the future.

Two and a half years later, oil has turned into tap water. Models are still important, but they are no longer scarce. What is truly scarce are the electricity, chips, and physical space needed to run models.

This mirrors the early evolution of the internet. In the 1990s, everyone was scrambling for content and traffic entry points. In the end, the winners were those who built the pipelines: Cisco, AT&T, AWS.

Now the AI industry is undergoing the same transition. Model companies think they are the protagonists, but after signing computing contracts and looking back, they find they have become long-term users of cloud providers. Those hundreds of billions of dollar contracts are not empowering agreements but rather entrapment agreements.

What is the cost for Microsoft in giving up OpenAI's exclusive distribution rights? A $250 billion Azure revenue commitment.

On the business level, did Microsoft lose?

According to a CNBC report, Barclays analysts believe this is marginally beneficial for Microsoft. It will no longer bear the financial pressure of building all of OpenAI's data centers and can release the funds to Copilot and other cloud services.

Microsoft traded "exclusive rights" for "certain revenue." From the logic of venture capital to the logic of public utilities.

The entire AI industry is undergoing this shift. The rate at which leading model companies burn cash is accelerating, while bills received by cloud providers are increasing. Model company valuations are highly volatile, but cash flow for cloud providers is growing steadily.

Axios mentioned a detail in a report last week: OpenAI had just written to investors the previous week, stating that the scale of computing power was its core competitive advantage compared to Anthropic and claiming that Anthropic had made a "strategic mistake by not acquiring enough computing power."

Days later, Anthropic signed two new computing agreements totaling over 8 gigawatts.

This is the AI competition of 2026: it's not about whose model is smarter, it's about who locks in more electricity.

In this restructuring, there is a rarely discussed beneficiary: Amazon.

Amazon now holds significant stakes in both Anthropic and OpenAI. Both cutting-edge AI labs have committed to spending over $100 billion on AWS.

Investing $50 billion in OpenAI in exchange for $138 billion in AWS revenue. Investing $33 billion in Anthropic in exchange for over $100 billion in AWS revenue.

Amazon doesn't care who wins. It cares that no matter who wins, the electricity bills will be sent to me.

The truth of contracts

Just the day after Microsoft and OpenAI announced their "decoupling," The Wall Street Journal published a report stating that OpenAI had not met internal revenue targets for several consecutive months in the first quarter of 2026, and user growth was also below expectations.

CFO Sarah Friar warned internally that if revenue growth does not accelerate, the company may not be able to afford future computing contracts.

The reality is, revenue is still stuck at $25 billion, while computing contracts have been signed for over $680 billion.

The market's reaction is more honest than any commentary. On the same day as the WSJ report, Oracle's stock fell 7.7%, CoreWeave dropped 7.4%, SoftBank in Tokyo fell nearly 10%, and Nvidia, AMD, and Broadcom fell between 2% and 6%. Investors were not selling OpenAI but rather all the companies that were counting on OpenAI to fulfill computing bills.

John Belton, a fund manager at Gabelli Funds, told CNBC that OpenAI's growth had clearly slowed from late 2025 to early 2026, with market share being eaten away by Anthropic and Gemini. OpenAI, having signed too many computing contracts, could not afford the bills.

This is the true picture of "the end of the exclusive era."

OpenAI has gained the freedom to sell models across three clouds. The cost is being bound by computing contracts from all three clouds at the same time. It has transitioned from being Microsoft's exclusive partner to becoming a long-term paying customer of Azure, AWS, and Oracle, with each contract being long term, each attached to specific chip architectures and deployment plans, and each assuming revenue will continue to grow rapidly.

OpenAI thought it had gained bargaining power, but in the computing power shortage of 2026, the bargaining power does not lie with the model companies. Whoever has electricity, chips, and physical space calls the shots. Those hundreds of billions of dollar contracts signed by model companies are not procurement agreements but rather servitude contracts. Once signed, the cost of moving becomes unacceptable; after models have been running on Trainium for two years, migrating to another chip architecture requires re-optimizing the entire training process—it's not as simple as changing a cloud account.

The "breakup" between OpenAI and Microsoft may appear to be a declaration of independence for the AI industry, but looking into the contract details reveals that the $250 billion commitment to Azure remains, the CFO is warning internally about potentially unpaid bills, revenue has continuously fallen short of targets, competitors are capturing market share, and all these issues assume that revenue in 2030 will be 11 times what it is today.

Those pipeline builders never talk about ideals. They only discuss contract durations, delivery schedules, and default clauses.

The ultimate winner of this AI arms race may not be the one with the best model, nor the one with the most funding. It is those infrastructure providers who collected deposits, signed long-term contracts, and can collect rent regardless of who wins. Just like the story of gold mining that always repeats, the ones who ultimately get rich are always the ones selling shovels.