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. You may ask what is openp?

Openp is a subsidiary created by openai

the purpose of which was to balance

fundraising with the nonprofit mission

using a capped profit structure through

open AI LP. They could access

significant capital that they need to

develop advanced AI models without

abandoning the core mission to benefit

humanity. This was the first step OpenAI

took to balance commercial viability

with ethics. 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.