We have been watching the AI race for three full years now. At first, it was all about the opening moves, the models, the breakthroughs, the chips, GBT versus Claude, Gemini versus Deepseek.

But this part of the game is over and we're past the opening and now we're entering the middle game. We're no longer competing on the models. We're competing on the infrastructure, on who controls the compute, who controls the clouds, and how the contracts that are being signed decide who gets to play.

This year, we saw two AI mega deals. Microsoft and OpenAI and Anthropic's web of partnerships, and these are so much more than just headlines. We're watching a formidable power play and a much larger match. Today is about uncovering what is really happening on the chessboard and who controls whom. Let's dive in.

On October 28th, OpenAI has announced an extensive restructuring when they officially granted Microsoft a 26% equity stake. This move underscores how the global AI race is evolving and it's evolving from a competition of models to a contest of ecosystems, partnerships and deals.

At the core of these partnership is the fact that OpenAI's original structure became increasingly unsustainable. Their unit economics do not work. They cannot survive without extensive capital injection from the outside. They're losing $2 on every dollar of revenue and they're burning through investments faster than they can monetize.

I spoke about this at length in my earlier videos on the unit economics of Chad GBT. If you want to learn more, check them out. In 2019, OpenAI created OpenAI LP or limited partnership which is a capped profit subsidiary.

Fast forward to now and Microsoft holds approximately 30% of OpenAI and their stake in OpenAI is valued at $135 billion which makes it one of the largest technology partnerships in history. What this 30% chunk means is that Microsoft now has significant influence over OpenAI's operations and development.

On top of this, Microsoft will be taking 20% of OpenAI's direct revenue. And as a refresher, the largest chunk of OpenAI's revenue comes from Chad GBT, not API sales. The reason I'm making an emphasis on the API sales is because API sales is the heart of B2B adoption and GPT is not doing very well on API sales, meaning they're predominantly consumer.

So yeah, back to Microsoft, they're going to be taking 20% of OpenAI's revenue through 2030. And this means that they're going to have a huge influence over OpenAI's go to market and pricing decisions. But this goes both ways. Microsoft will also pay OpenAI around 20% of revenue from Azure OpenAI services and Bing AI features.

If you're building a company, you already know scaling means hiring. With every hire, you're looking at payroll setup, benefits enrollment, IT provisioning, access to 12 different tools, corporate card, compliance paperwork. Onboarding one takes your op teams about 6 hours, but you're hiring 15 this quarter. That's 90 hours of manual work.

Most companies are running on 100 plus disconnected tools. Now, each one solves a problem, but together they create silos. Ripling, one of the fastest growing HR, IT, and spend platforms out there, calls this SAD software as a disservice.

To combat this, Riplin built a unified platform across four systems that run your company. one source of truth for all employee data that allows you to manage HR, payroll, spend, and IT in one unified system. You can hire contractors in more than 185 countries and full-time employees in more than 80 countries and pay them in different currencies, which really matters because your next hire probably isn't local.

New hire gets their laptop, all accounts, correct permissions based on their role all automatically. Someone leaves, everything gets deprovisioned instantly. Real-time visibility into company spending. No more waiting for expense reports. No more budget surprises.

Rippling is trusted by more than 20,000 companies across a range of industries, including Cursor, Barry, Chaz.com, and Liquid Death. Every fragmented system in your stack is holding back execution. Rippling provides four core systems to manage your entire staff end to end. HR, payroll, IT, and spend management. All connected, all automated, helping you increase savings and make better decisions. And that's how you build a company that can actually move as fast as you scale. Check out stopsad.com to see if SAD is affecting your organization and what you can do about it.

Now, why was this partnership architected and who controls whom? The real value of this partnerships happens through several layers. The first one being the revenue share. $865 million captured in 9 months of 2025. Now look at this number. 865 million in 9 months.

The fact that Microsoft captured this much in revenue in just 9 months shows both the scale of financial potential from just one foundational AI company and the dependency, the inevitable dependency that it creates. Because first of all, it shows that OpenAI is a massive revenue driver for Microsoft and a monetization engine for Azure. because Microsoft extracts a significant cut from every dollar OpenAI makes regardless of OpenAI's own profitability.

And secondly, this flow of hundreds of millions of dollars per quarter means that OpenAI's growth directly contributes to Microsoft's cloud business and its ability to justify further infrastructure expansion which becomes their moat in AI. And thirdly, the size and cadence of these payments prove that there is a very acute pressure on OpenAI's unit economics because we already know that their inference costs already exceed revenue and now they're giving another 20% to Microsoft which means that it's going to be extremely difficult for them to achieve sustainable business economics.

This revenue share illustrates how partnerships and deals in the AI economy are a lot less about profit sharing and much more about deep control over the economics and trajectories of foundational AI companies.

The second layer of this is guaranteed sale of infrastructure because OpenAI has committed to purchasing $250 billion in Azure services. What this means is that OpenAI is not just a big client that Microsoft managed to land. Microsoft's whole business model is transforming because they used to sell software. That's what they've been known for and they're now becoming a renting compute infrastructure business.

It's not good or bad. It's just fascinating to observe how AI race is becoming the race of computing power. When OpenAI makes up 50% of Azure's revenue growth, Microsoft becomes dependent on a partner that is losing $2 on every dollar of revenue and the same partner is burning through compute resources faster than it can monetize.

And if you add the whole AGI narrative on top of this and all the effort and the money that is being spent on achieving this magic entity of AGI, which if you're curious about AGI, watch the previous video. There is a lot of capital burning happening here. This creates a risk that cloud businesses do not typically face. Microsoft cannot easily replace OpenAI's demand.

OpenAI is the largest AI client on the planet and the one with the biggest computing needs. Signing them as a client is a double-edged sword. Yes, you get the largest AI client on the planet. You're selling more than you ever have, but you're also scaling your operations to cater to this client. And if this client leaves, you're screwed. To find the next OpenAI level customer, for example, Anthropic or Deepseek, you would need years for them to scale to the needs of OpenAI.

And now, let's turn this around because OpenAI is actively seeking diversification and they're seeking it through Core Weeave and Oracle. And they're doing so to reduce their dependency on Microsoft. And the reason they're looking to diversify is because this extreme financial strain makes it risky for OpenAI to depend solely on Microsoft. If Microsoft raises prices or changes terms of this partnership, OpenAI's whole survival could be at risk. And this risk keeps growing.

But the same risk goes for Microsoft. If OpenAI's usage drops or diversifies to other clouds or the company completely collapses under its own financial structure, which yes, extremely unlikely, but nevertheless not impossible, Microsoft will lose both the revenue stream and the growth narrative that supports its $4 trillion valuation.

Yep, don't forget that Microsoft's valuation went up because of this partnership as well. You may ask, but why did Microsoft do it given such a massive risk? They did it because they were catastrophically behind Google and Amazon in the AI race.

At the time of the initial $1 billion investment in 2019, Microsoft held only 29% of the cloud market versus AWS's 37%. And they did not have any competitive AI research capability to match Google's DeepMind or Amazon's Alexa. The partnership with OpenAI delivered an instant 10-year leap because Microsoft got exclusive access to frontier models that they couldn't build internally and a chance to embed AI into its Microsoft 365 and Teams.

By doing this, they create distribution advantage that neither Google or Amazon would be able to replicate. And to top it off, they get the ability to position Azure as the only cloud with OpenAI API access. This was a gamble, and I applaud everyone who has architected this gamble because it pays off.

Azure grew 40% annually through 2025 compared to AWS's 19%. Microsoft stock gained over $2 trillion in market cap since this partnership began. Microsoft now controls the enterprise AI distribution layer through co-pilot and Microsoft now controls the entire enterprise distribution layer and as a B2B product manager who works at an enterprise and the vast majority of enterprises are on Azure cloud and using Microsoft stack it is hard to overestimate the reach that Microsoft has into the enterprise tech.

And finally number four the competitive mode which is arguably the highest currency of all they got exclusive API rights through Azure until AGI is verified, which means that they're locking enterprise customers into Microsoft's cloud. Microsoft's exclusive API rights to OpenAI mean that any enterprise that wants to use OpenAI's models in production must route all traffic through Azure.

And this creates a dependency loop almost an unbreakable loop where purchasing access to GBT5 for example for your business automatically makes you an Azure customer regardless of whether you're on AWS or Google Cloud or onrem. This is beyond billing. This is technically an architectural lockin because when you purchase access to enterprise level GBT, you have no choice but to accept Azure's private network, Azure's compliance, Azure's data residency rules, Azure's authentication, and Azure's pricing because OpenAI's API literally cannot be run anywhere else until AGI is verified by an independent expert panel.

And on top of everything I just said, once an enterprise builds applications on Azure OpenAI service, it automatically integrates with Azure Cognitive Search, Azure Functions, Azure Key Volt, and Azure Rulebased Control and the switching costs become enormous because you'd need to rewrite your entire application. You would need to rebuild your data pipelines.

The problem with the vendor switching even as is with the current combo of Microsoft 365 copilot and teams and when it needs to be done companies hire full teams of people just to do the vendor switch. I mean I was recently doing a switch from Apple to Google Workspace and I wanted to shoot myself in the head.

Microsoft used OpenAI's virality to its advantage and they used it to force cloud migration to Microsoft and so much so that there are a bunch of companies that picked AWS as their cloud provider back in 2020 and are now running significant Azure workloads because their engineering teams at some point said that they wanted GBT and the only compliant enterprisegrade path to access GBT requires full Azure adoption.

This is the reason why Azure captured 62% of the Genai cases despite only 29% of cloud share. They're not just betting on the infrastructure. They're betting on a legal absolutely legal distribution for the world's most in demand AI models. And that exclusivity doesn't end until someone declares AGI. And now coming back to AGI, Microsoft has every reason and every incentive to delay the declaration of AGI indefinitely.

So this was the OpenAI Microsoft situation and now let's look at OpenAI's rival Anthropic and see what they've come up with. Anthropic went in a completely different but nevertheless brilliant direction. I personally have huge respect for Anthropic as a company and I prefer their models to any other LLM. And it's interesting how they're balancing the power game while trying to maintain independence.

They secured investments and compute commitments from all three hyperscalers, Google Cloud, AWS, and Microsoft. With Google, they became their major investor and cloud partner. Through Google, they got access to 1 million of custom tensor processing units or TPUs that will be coming online by 2026 and 1 gawatt of power.

To put this in perspective, 1 gawatt of power is comparable to the output of a large nuclear power plant, which means an enormous scale of AI compute capacity dedicated to powering anthropics AI models.

Now for the AWS, they made an $8 billion total investment in Athropics, becoming their lead financial backer. The motive of AWS was well, first of all, securing Anthropic as one of the largest clients for the cloud business, but also to strengthen AWS AI service with Anthropics Claude.

And on top of this, their deal included a multi-billion dollar cash infusion and commitments for AWS to supply 1 million Tranium 2 processors, which is the Amazon's custom chip specifically designed for AI training, which means that in practice, this made AWS Anthropic's principal cloud provider.

And lastly, Microsoft and Nvidia. In November 2025, Microsoft and Nvidia announced a new partnership with Enthropic that involves several moves. Enthropic agreed to a $30 billion commitment to use a Microsoft's Azure cloud for future compute needs, meaning that Anthropic will spend at least $30 billion on Azure Infra over multiple years. This guarantees Azure a large and long-term stream of revenue.

And now they're not only OpenAI's cloud provider, they're also Anthropic's provider. And this in combination means that Anthropic is the only foundational model company available on all three of the world's most used cloud services. And as we continue rotating this board, Anthropic has committed to spend $50 billion in future compute spending, which means that they're locking themselves into these guys for years to come.

There is a fundamental divergence in the anthropic and multicloud approach and the Microsoft OpenAI partnership. And the delta is in the revenue model, who they sell to and how they make money. OpenAI has a consumer first business model. 73% of OpenAI's revenue comes from consumer subscriptions, Chad GPT plus and Chad GPT Pro with 27% and some sources say 15% from API and enterprise.

They have 800 million weekly users. They love the weekly user metric, but only 5% of them are paid. This consumer oriented business model automatically implies dependency on one cloud provider and not just any provider. It must be always on because they're serving 8 billion weekly users and you cannot run a consumerf facing product at this scale on multiple clouds without catastrophic user experience fragmentation. OpenAI has locked itself into complete Azure dependence.

Anthropic on the other hand is enterpriseoriented. Anthropic's revenue model is the inverted version of open AIS. They get around 85% of their revenue from enterprise API calls and only 20% from consumer subscriptions. Enterprise customers access claude through AWS bedrock, Google's vertex AI, Azure AI foundry or direct API.

Point is it doesn't matter which cloud runs the inference. Cloud is everywhere. This is why the multicloud model is viable for anthropic enterprise workloads are batch oriented, latency tolerant and already distributed across clouds based on a company's IT infrastructure. It's a lot easier for a business or an enterprise to use claude than chajbt.

And as Ben Thompson observed, anthropic's lack of a strong consumer play means that it is much more tenable, if not downright attractive, for them to have a supplier type of relationship with AWS. Only Anthropic can pull this off right now. Open AAI cannot replicate this model because their consumer success is their blessing and their curse.

Now, if you paid attention to these numbers, a question that may have formed in your brain is seems like Anthropic is doing better than OpenAI. Anthropic's multicloud partnership despite complexity and despite operational overhead does deliver superior unit economics. Their multicloud model optimizes costs and once again remember they're serving 8 billion users weekly and this volume requires a unified infrastructure that cannot tolerate delays from crosscloud routing which forces them to accept Azure's premium pricing even though their inference costs run at 200% revenue. This means that Enthropic's multicloud partnership is the key to profit margin.

The real war and the real race in AI is control over the infrastructure. The big boys are no longer competing for the best models. Microsoft and AWS do not have their native models at all. And while Google is investing quite a bit into Gemini, they're still competing to own the compute layer where all models must run.

Anthropic's committed compute spending across all three hyperscalers is the only strategy that keeps them independent. And even that locks Enthropic into infrastructure dependencies for the next decade. We're getting into the middle game now where all players are making their moves and fighting for who owns the board. The models are just the pieces. The board or the infrastructure decides who is going to win. As always, we hope this was helpful. Till next time. Bye.