Microsoft CEO Nadella\’s latest interview: Whoever owns computing will dominate the world

Microsoft CEO Satya Nadella recently said in an interview with technology website Stratechery founder Ben Thompson that whoever has computing can dominate the world. This is why Microsoft chose to cooperate with OpenAI. Because OpenAI is more powerful than OpenAI. Everyone believes in calculation.
Nadella also said that sometimes cooperation is the only way to achieve success.
In the past five years, Microsoft\’s cooperation with OpenAI has been effective. He wants to ensure that it will continue to be effective in the next five years and the next ten years.
If we want long-term and stable cooperation, we must ensure that both parties can succeed.
When asked whether there is or should be some kind of ‘anti-Google alliance’ in the AI ​​field? Nadella said there is always room in the market for someone to vertically integrate.
In the long run, he believes more in horizontal specialization.
Take the chip industry as an example. Nvidia hopes that leading AI models can be trained on its chips.
But in fact, Google has its own chip TPU. AMD is also actively innovating in this area. The same is true for Microsoft.
In terms of capital expenditure, Nadella said that even if AI is excluded, Microsoft is a knowledge-intensive and capital-intensive enterprise.
Over the past seven years, Microsoft\’s capital expenditures have increased from 13% to 26% of gross profit.
The following is a summary of the interview: Cooperation with OpenAI Q: When it comes to the integration of Microsoft, how do you solve this problem for cooperation with OpenAI? Are you worried about this? For example, Microsoft has done a good job in this area. Both companies are moving in the right direction. But there seems to be a dependency in this cooperation. Does Microsoft have control? How do you deal with these issues? Nadella: For us, the partnership with OpenAI is like the partnership with Intel back then. Or it’s like the partnership with SAP when we were developing SQL.
These collaborations are industry-defining. They\’re Microsoft-defining. So we\’re very invested in this partnership.
It\’s a simple logic. It\’s related to calculations.
Question: Who owns computing? Who can dominate the world? Nadella: Yes.
This unconventional bet (partnering with OpenAI) was back in 2019. At that time we said, wow, maybe we should invest in a lot of computing.
At the time, OpenAI believed in computing more than anyone else. Even people inside Microsoft.
So. That\’s why we made this bet.
It\’s worked for the past five years. I want to make sure it continues to work for the next five years. And the five years after that.
This was a very critical period. Both of us achieved success.
How to ensure the long-term stability of this cooperation is to ensure that both parties can continue to win. At least that\’s what I think.
I think that for them (OpenAI), we are the infrastructure and they are the model developers.
They develop apps. We develop apps. Third parties are developing apps. And so on.
This will create competition. There will be some completely vertically integrated competition.
Vertical integration works great until one layer of your business loses its competitiveness.
If you want to check out Microsoft, you don\’t have to keep an eye on history.
You have to keep an open mind. Sometimes collaboration is the only way to achieve success.
Integration and Modularization of AI Q: You mentioned that OpenAI has confidence in computing. This is what Microsoft is convinced of.
Given Google\’s leadership in models and especially infrastructure, do you think there is or should be some kind of anti-Google alliance in the AI ​​field? Do we see this coming? Not just Microsoft and OpenAI. But potentially Apple as well? Nadella: I think there\’s always room in the market for someone to vertically integrate.
I often think back to the past. I think of the Gates/Grove (representing Microsoft/Intel) model. And then there\’s the Apple or Google model. That\’s the vertical integration model.
I would say that if you really look at it in the long run, I believe more in horizontal specialization.
Take the chip industry, for example. Nvidia CEO Jensen Huang is sitting there, really aggressive in executing on some incredible roadmaps.
Today, his foundation is ensuring that leading AI models are trained on Nvidia chips.
But guess what? Google doesn\’t do this. It trains on its own chip TPU.
At the same time, AMD CEO Su Zifeng is also actively innovating in this area.
In addition, we are also building our own chips.
So everyone says, OK. Let\’s do silicon innovation. Let\’s do model innovation. And that includes OpenAI and Meta. And Mistral. And there will be many more in the future.
Anyway, any of our applications, including Copilot, will use GPT-4o and mix it with Phi (Microsoft\’s AI model) and other products.
So. I think any enterprise application, what they\’re actually most excited about is models as a service.
So I think it\’s going to be more diverse. At least my history tells me. There are rarely winner-take-all games.
Be very aware of this. Then work toward playing winner-take-all.
But everything else has to take that broad-tent platform approach.
Q: It does sound reasonable. And you are talking about the idea of ​​the model being commercialized.
Microsoft has hired a lot of talent from Inflection AI. It looks like you want to make sure there\’s diversity in the models you offer.
But if the model is going to be commoditized, why would the dynamics in the cloud computing space be different than they have been over the past 12 to 15 years? Is this really something new and different? Nadella: That\’s a very good question.
I think Hyperscaler (the technical term in the field of very large-scale computing) has a fundamental structural advantage in this regard. In a sense, if you say what else the world will need in five years, I would say Hyperscaler computing tools are everywhere.
If you think about it. I think the new model for economic growth is as clear as ever. That is you need more electricity from renewable sources. A better grid. Better computing.
If you have these, then other sectors of the economy can really benefit from these two things.
Any country or community can occupy a place in economic growth as long as it is at the best frontier in terms of efficiency.
So if you accept this high-level premise, then it\’s definitely (something new).
Question: But what will be the competitive dynamics in the future? Because Amazon got a head start. They basically got the entire SaaS enterprise of the company you started on Amazon.
Microsoft has moved to cloud computing as their enterprise customer base has moved.
And Google\’s answer is look. Ours is the best. Just give it a try. They are currently a distant third.
Will this play out in a similar way in the future? Will data gravity still dominate? Maybe AI is a big new thing. But the actual competitive dynamics still exist.
Nadella: I haven\’t seen at least one single-cloud enterprise customer.
I remember when I first started cloud computing. Everybody would talk about it. Like this was going to be. Oh my gosh. It was winner-take-all.
And what I think about a lot is, I grew up on servers, and when people say we won, I don\’t even really understand it.
In fact, every type of server, whether it is an operating system, a database, a Web server, or all these middle-tier things, has two or three participants.
So, fundamentally speaking, I think there is definitely room for two companies in hyperscale, if not three.
Steve (former Microsoft CEO) used to always tell me that in a multi-player market, revenue share and market share are two different things. The same is true in the cloud computing market.
Regardless, I do think the market can accommodate three companies.
Remember, when we started, Amazon had no competition for six or seven years.
Guess what. The competition is coming. Here we are. I feel really, really good about the next level of competition.
I didn\’t start from behind.
In fact, if anything, we have a beginning. And things have changed.
Take B2C customers as an example. Whether it\’s Shopify, Spotify or others.
This also benefits from the OpenAI API. None of these were previously Azure customers.
This is also the first time. They are not all on Azure. But they are Azure customers. This is a huge change in our fortunes.
Capital Expenditure and the Future Q: You have said that you have visibility into revenue expenditures.
There\’s no doubt about it. You have to invest in AI. But your capex as a percentage of gross profit has gone up from 13% to 26% over the last seven years. That\’s a big increase.
What makes you so confident that you\’re going to get a return; or does it not matter because you\’re investing in competitive dynamics anyway? Nadella: I think about the laws of economics. I think you pointed out correctly that we are a capital expenditure-heavy enterprise.
The reason most people are focused on our capital expenditures is simply because of AI.
In fact, even if AI is removed, we are still a knowledge-intensive and capital-intensive enterprise. This is what hyperscale requires.
If you can\’t put $50, $60 a year into capital expenditures right now, then you can\’t say, Hey, I want to get into the hyperscale market.
This is what is necessary to enter the market.
But at the same time, it\’s always going to be governed by what\’s happening in the market. It can\’t be much faster than your revenue growth.
Therefore. There is an absolute regulator. That\’s for sure.
For example. How much does (AI) training cost? The distribution of training calculations has step function changes. But the final result is driven by demand.
So. If you put this combination together, I feel like if something happens in cycles, it\’s not that difficult to adjust it.
As a pure business management. I wouldn\’t just manage it for a quarter. It doesn\’t scare me.