0:00

Reboot your credit card

0:02

with Apple Card, the credit card created by

0:05

Apple. It gives you unlimited daily

0:07

cash back that you can now choose

0:09

to grow in a high-yield savings

0:11

account that's built right into the Wallet app.

0:13

Apply for Apple Card now in the Wallet

0:16

app on iPhone and start growing

0:18

your daily cash with savings today.

0:21

Apple Card subject to credit approval,

0:23

savings is available to Apple Card

0:25

owners subject to eligibility

0:28

requirements. Savings accounts provided

0:30

by Goldman Sachs Bank USA, member

0:33

FDIC, terms apply.

0:59

Are you having fun in your job?

1:11

I'm loving every day of it, Steven. Most

1:16

CEOs of big technology firms are

1:18

not loving every day right now. They've

1:20

been facing all sorts of headwinds and

1:22

backlash. But you can see why Satya

1:24

Nadella might be the exception. He's

1:27

worked at Microsoft for more than 30 years, nearly 10

1:29

as CEO. At

1:31

the start of the personal computer era,

1:34

Bill Gates' Microsoft was a behemoth,

1:37

eager to win every competition and

1:40

crush every rival. But the Internet

1:42

era put the company on its heels. Newer

1:44

firms like Google,

1:45

Facebook, and Amazon were more

1:47

nimble, more innovative, and maybe

1:50

hungrier. Jeff Bezos of Amazon

1:52

would reportedly refer to Microsoft as a

1:54

country club. But under

1:57

Nadella, Microsoft has come roaring

1:59

back.

1:59

invested heavily in what turned out to be big

2:02

growth areas like cloud computing. Microsoft

2:05

has always been in the business of acquiring other

2:07

companies, more than 250 over its history, but

2:10

some of the biggest acquisitions have been Nadella's,

2:13

LinkedIn, Nuance Communications,

2:16

and if regulators allow, the gaming

2:18

firm Activision Blizzard. And

2:20

there have been many more key acquisitions like

2:23

GitHub, where computer programmers store

2:25

and share their code. Once again,

2:28

Microsoft is a behemoth, the

2:30

second most valuable company in the world, trailing

2:32

only Apple. Its stock price is up

2:35

nearly 50% since the start of 2023. But

2:39

that's not even the reason why Microsoft

2:41

has been all over the news lately. They're

2:43

in the news because of their very splashy

2:46

push into artificial intelligence

2:48

in the form of chat GPT, the

2:50

next level chat bot created by

2:52

a firm called OpenAI. Microsoft

2:55

has invested $13 billion

2:57

in OpenAI for a reported 49%

2:59

stake in the company, and

3:02

they quickly integrated OpenAI's tech

3:04

into many of their products, including

3:06

the Microsoft search engine Bing.

3:09

For years, Bing was thought

3:11

of as something between footnote and

3:14

joke, running a very distant

3:16

second to Google. But suddenly,

3:18

Bing with chat GPT is

3:21

on the move, and Google is trying to

3:23

play catch up with its own chat bot called

3:26

Bard. So how exactly

3:28

did Satya Nadella turn the country

3:30

club into a bleeding edge

3:32

tech firm with a valuation of

3:35

more than two and a half trillion dollars?

3:37

Our mission, Stephen, is to empower

3:40

every person and every organization

3:42

on the planet to achieve more. And so

3:45

as the world around us achieves

3:47

more, we make money.

3:48

I like that. I mean, I assume

3:51

you actually believe that. You're not just saying that, are you? No, 100%.

3:54

You have to have a business model that is

3:57

aligned with the world around you doing well.

3:59

Today

4:02

on Freakonomics Radio, we speak with

4:04

Satya Nadella about the blessings

4:06

and perils of AI.

4:09

We talk about Google and Heidegger,

4:12

about living with pain, and

4:14

about Microsoft's succession

4:16

plan.

4:20

No, it'll be nothing like that. Nadella

4:23

promises. We will take succession

4:25

seriously.

4:38

This is Freakonomics Radio, the

4:40

podcast that explores the hidden side

4:42

of everything with your host,

4:45

Stephen Dubner.

4:53

I spoke with Satya Nadella one afternoon

4:55

earlier this month. I was in New York and he

4:58

was in his office at Microsoft's headquarters

5:00

near Seattle. It's fantastic to

5:03

have a conversation again. We first

5:05

interviewed Nadella in 2017 for

5:07

a series called The Secret Life of

5:09

a CEO. Even then, he

5:12

was extremely excited about

5:14

AI. At the time, Microsoft

5:16

was high on a virtual reality headset

5:19

called the HoloLens. Think

5:21

about it. The field of view, what

5:23

you see is a blend of the

5:25

analog and digital. The

5:27

ability to blend analog

5:30

and digital is what we

5:32

describe as mixed reality. There are times when

5:34

it'll be fully immersive. That's called virtual

5:36

reality. Sometimes when you can

5:38

see both the real world and the artificial

5:41

world, that's what is augmented reality.

5:44

But to me, that's just a dial that you set.

5:47

Just imagine if your hologram was right

5:49

here interviewing me as

5:51

opposed to just on the phone.

5:53

Back then, Nadella cautioned that there

5:55

was still a lot of work to do. Ultimately,

5:58

I believe in order to bring about some of these

6:00

magical experiences and AI capability,

6:03

we will have to break free of some of the limits we're

6:05

hitting of physics, really.

6:07

The limits of physics haven't been

6:09

broken yet, and the hollow lens has

6:11

not been the hit that Microsoft was

6:13

hoping for. But

6:15

Nadella's devotion to AI is

6:17

paying off big time in the form of chat

6:20

GPT, which quickly captured

6:22

the imagination of millions. GPT

6:25

stands for Generative Pre-trained

6:27

Transformer, and chat GPT

6:29

is what is known as a Large Language Model,

6:32

or LLM. It takes in vast

6:34

amounts of data from all over the internet so

6:36

it can learn how to read and

6:38

answer questions very much like a

6:41

human, but

6:42

a really, really smart human,

6:44

or perhaps a million smart humans.

6:47

And the more we ask chat GPT

6:49

to answer questions, or summarize

6:52

arguments, or plan itineraries,

6:55

the more finely tuned it gets, which

6:57

proves at the very least that we humans are

6:59

still good for something. The current

7:02

iteration is called GPT-4, and what's

7:05

the relationship between chat GPT

7:08

and Bing?

7:08

Basically, Bing is part of chat GPT

7:11

and chat is part of Bing, so in either way,

7:13

it doesn't matter which entry point you come to, you

7:15

will have Bing.

7:16

So Satya, I asked chat GPT

7:18

for some help in this interview. I said

7:21

I'm a journalist interviewing Satya Nadella and I want

7:23

to get candid and forthright answers.

7:25

You know, I just didn't want corporate boilerplate,

7:28

and what chat told me was to

7:30

do my homework, which you know, I did, I usually

7:32

do that, to ask open-ended questions,

7:35

which I typically try to do. But one

7:37

that hung me up a little bit was I need to build

7:39

rapport. Now, we have a relatively

7:42

short time together today. Are there

7:44

any shortcuts to building rapport?

7:46

Yeah, what's your knowledge of cricket?

7:50

Oh, I blew it. I

7:52

knew that you're a big cricketer, you

7:54

played as a kid, I knew you cared more about

7:57

cricket than schoolwork as a kid, but no,

7:59

I blew it.

7:59

That's too bad because there's a world test

8:02

championship starting tomorrow. I

8:04

was going to ask you about it, but hey, look,

8:06

your love for economics builds me an

8:09

instant ripple.

8:10

I'd like you to walk us through Microsoft's

8:12

decision to bet big on open AI,

8:15

the firm behind chat GPT. There was an early

8:17

investment of a billion dollars, but then much, much

8:19

more since then. I've

8:22

read that you were pretty upset

8:24

when the Microsoft research team came to

8:26

you with their findings about open AI is

8:28

LLM large language model. They said that they were

8:31

blown away at how good

8:33

it was and that it had surpassed

8:36

Microsoft's internal AI research

8:38

project with a much smaller

8:40

research team in much less

8:42

time. Let's start there. I'd like you to describe

8:45

that meeting. Tell me if what I've read first of all is true.

8:47

Were you surprised and upset with your

8:49

internal AI development? Yeah, I

8:51

think that this was all very recent. This

8:55

was after GPT four was very much

8:57

there. And then that was just mostly me pushing

9:00

some of our teams as to, Hey, what did we miss

9:02

you got to learn?

9:03

You know, there were a lot of people at Microsoft who got

9:06

it and did a great job of, for example,

9:08

betting on open AI and partnering

9:10

with open AI. And to me four

9:12

years ago, that

9:14

was the idea. And then as we went

9:16

down that journey, I started saying, okay, let's

9:18

apply these models for product building,

9:20

right? Models are not products. Models can

9:22

be part of products. The first real

9:25

product effort, which we started was get

9:27

up co-pilot. And, you

9:29

know, quite frankly, the first attempts on

9:31

get up co-pilot were hard because, you know, the

9:33

model was not that capable, but it is only

9:36

once we got to GPT three, when

9:38

it started to learn to code that

9:40

we said, Oh, wow, this emergent phenomena,

9:43

the scaling effects

9:44

of these transformer models are

9:46

really showing promise.

9:49

Nadella may be underplaying the tension between

9:52

Microsoft and open AI, at least

9:54

according to recent wall street journal article called

9:56

the awkward partnership leading the

9:58

AI boom.

9:59

It describes, quote, conflict and

10:02

confusion behind the scenes. And

10:05

because the OpenAI deal is

10:07

a partnership and not an acquisition, the

10:09

journal piece makes the argument that Microsoft

10:11

has influence without control

10:14

as OpenAI is allowed to partner

10:16

with Microsoft rivals. Still,

10:19

you get the sense that Nadella is excited

10:21

about the competitive momentum chat GPT

10:24

has given Microsoft, as you can tell from

10:26

this next part of our conversation.

10:30

Google still handles about 90% of

10:33

online global search activity.

10:36

An AI search enabled model

10:38

is a different kind of search plainly

10:40

than what Google's been doing. Google is trying

10:43

to catch up to you now. How

10:45

do you see market share in search

10:47

playing out via Bing, via chat

10:49

GPT in the next five and 10 years? And

10:52

I'm curious to know how significant that might be

10:54

to the Microsoft business plan overall.

10:56

This is a very general purpose technology,

10:59

right? So beyond the specific use

11:01

cases of Bing chat or chat

11:03

GPT, what we have are reasoning

11:06

engines that will be part of every

11:08

product. In our case, they're part of Bing and chat

11:11

GPT. They're part of Microsoft 365. They're

11:14

part of Dynamics 365. And so in

11:16

that context, I'm very excited about what

11:19

it means for search. After all, Google,

11:21

as you said, rightfully, they're dominant

11:23

in search by a country mile. And

11:26

we've hung in there over the decade. We've

11:28

been at it to sort of say, hey, look,

11:30

our time will come where there will be a real

11:33

inflection point in how search

11:35

will change. We welcome Bing versus

11:38

Bard

11:38

as competition. It'll

11:40

be like anything else, which is so

11:42

dominant in terms of share and also

11:45

so dominant in terms of user habit,

11:47

right? We also know that defaults

11:49

matter. And obviously, Google controls the

11:51

default on Android, default on iOS,

11:54

default on Chrome. And so they have a great

11:56

structural position. But at

11:58

the same time, we're going to have a great conversation. whenever there is a

12:00

change in the game, it is all up

12:02

for grabs again to some degree, and I know

12:05

it'll come down to users and user choice.

12:08

We finally have a competitive angle

12:10

here, and so we're going to push it super hard.

12:13

What are some of your favorite uses,

12:15

personal or professional, for chat

12:18

GPT? The thing that I've talked about,

12:20

which I love is the cross-lingual

12:23

understanding. That's kind of my term for

12:26

it. You can go from Hindi

12:28

to English or English to Arabic

12:30

or what have you, and they've done a good job. If

12:32

you take any poetry in any one language

12:36

and translate it into another language.

12:38

In fact, if you even do multiple languages,

12:41

so my favorite query was, I said, I always,

12:43

as a kid growing up in Hyderabad, India,

12:45

said, I want to read Rumi,

12:48

translated into Urdu and translated

12:50

into English. And one

12:52

shot, it does it. But the most interesting

12:55

thing about that is it captures

12:57

the depth of poetry. So it

13:00

finds somehow in that latent

13:02

space, meaning that's

13:04

beyond just the words and their

13:07

translation. That I

13:09

find is just phenomenal.

13:11

This amazes me. You're saying, you, the CEO

13:14

of a big tech firm, is saying

13:16

that one of the highest callings of chat GPT

13:18

or a large language model is the translation of poetry.

13:21

I love it. I mean, I know you love

13:23

poetry, but what excites

13:25

you more about that than

13:28

more typical

13:30

business, societal, political, economic

13:32

applications?

13:33

I love a lot of things. I

13:36

remember my father trying to read

13:38

Heidegger in his forties

13:41

and struggling with it, and I've attempted

13:43

it thousand times and failed. And,

13:45

you know, he's written this essay. Somebody pointed

13:47

me to somebody said, Oh, you got to read that

13:49

because after all, there's a lot of talk about AI

13:52

and what it means to humanity. And

13:54

I said, let me read it. But I must say, you

13:56

know, going and asking chat GPT

13:58

or big chat to summarize.

13:59

Heidegger is the best way to read

14:02

Heidegger.

14:05

According to ChatGPT, Heidegger

14:08

himself would not have been a fan

14:10

of AI. In Heidegger's

14:13

view, Chat tells us, technology,

14:16

including AI, can contribute to what

14:18

he called the forgetting of being.

14:22

And Heidegger is hardly alone. After

14:25

all, philosophy and poetry will

14:27

likely not be the main use

14:29

cases for AI. So after

14:32

the break, we talk about potential downsides

14:34

of an AI

14:35

revolution and the degree to

14:37

which Microsoft cares. I want

14:39

all 200,000 people at Microsoft working

14:42

on products to think of AI safety. I'm

14:44

Stephen Dubner. This is Freakonomics Radio.

14:46

We'll be right back.

14:57

Freakonomics Radio is sponsored by Capital

15:00

One. Capital One offers commercial solutions

15:02

you can bank on. Now more than ever, your

15:05

business faces specific challenges

15:07

and unique opportunities. That's

15:09

why Capital One offers a comprehensive

15:12

suite of financial services, custom

15:14

tailored to your short and long-term

15:16

goals. Backed by the expertise,

15:19

strategy and resources of a top 10

15:21

commercial bank, a dedicated team

15:23

works with you to support your success

15:26

and help achieve your goals.

15:27

Explore the possibilities at

15:30

CapitalOne.com slash

15:32

commercial.

15:33

Hey there, Stephen again. Communication

15:36

is key to success in business and in

15:38

life, but things like speaking off the cuff,

15:41

negotiating a raise and giving feedback

15:43

can take specific skills and a lot of

15:45

practice to do well. Think Fast,

15:48

Talk Smart is a podcast that teaches

15:50

these skills in a fun, engaging way. Sponsored

15:54

by Stanford University lecturer, Matt Abrahams,

15:56

each episode covers all aspects of

15:58

communication, including... actionable

16:00

tools. Hone your communication

16:03

skills and learn valuable career insights.

16:05

Listen to Think Fast, Talk Smart

16:08

wherever you get your podcasts.

16:18

Last month, a group of leaders from across

16:21

the tech industry issued a TERS one-sentence

16:24

warning. Understanding the risk

16:26

of extinction from AI should

16:29

be a global priority alongside

16:31

other societal-scale risks such

16:33

as pandemics and nuclear war. The

16:36

extinction they are talking about is

16:39

human extinction. Among

16:41

the signatories were Sam Altman, the

16:43

CEO of OpenAI, and

16:45

two senior Microsoft executives.

16:48

Altman, Satya Nadella, and

16:50

other executives from firms working on AI

16:53

recently met with President Biden to

16:55

talk about how the new technology should

16:57

be regulated. I asked Nadella

16:59

where he stands on that issue. I

17:02

think the fact that

17:04

we are having the conversation

17:07

simultaneously about both the

17:10

potential good that can come

17:12

from this technology in terms of

17:14

economic growth that is more

17:16

equitable and what have

17:18

you. At the same time that

17:20

we are having the conversation on

17:22

all the risks, both here and now

17:24

and the future risk, I think it's a super

17:27

healthy thing. Somebody gave me this analogy,

17:29

which I love. Just imagine when the steam engine

17:31

first came out if we had a conversation

17:34

both about all the things that the steam engine

17:36

can do for the world and the industrial production

17:38

and the industrial revolution and how it will change

17:41

livelihoods. At the same time, we were

17:43

talking about pollution and factory

17:46

filth and child labor. We

17:48

would have avoided whatever, 100 years plus

17:50

of terrible history. So

17:52

then it's best to be grounded on

17:55

what does the risk framework look like. If

17:57

AI is used to create more and more people, then we can do that.

17:59

more disinformation, that's a problem

18:02

for our democracy and democratic institutions.

18:05

Second, if AI is being

18:07

used to create cyber attacks or bioterrorism

18:10

attacks, that's a risk. If there is

18:12

real world harms around bias, that's

18:14

a risk. Or employment displacement,

18:16

that's a risk. So let's just take those four.

18:19

In fact, those were the four even the White

18:21

House was upfront on and saying, hey,

18:23

look, how do we really then have

18:25

real answers to all these four risks?

18:28

So in terms of, for example, take

18:29

this information, can we have techniques

18:32

around watermarking that help verify

18:35

where the content come from? When

18:37

it comes to cyber, what can we do to

18:39

ensure that there is some regime

18:42

around how these frontier models are being

18:44

developed? Maybe there is licensing,

18:46

I don't know, this is for regulators to decide.

18:49

Microsoft

18:51

itself has been working on provisions to

18:53

best govern AI. For instance, safety

18:56

breaks for AI systems that control

18:58

infrastructure like electricity or

19:01

transportation. Also a certain

19:03

level of transparency so that academic

19:05

researchers can study AI systems.

19:08

But what about the big question?

19:11

What about the doomsday scenario wherein

19:13

an AI system gets beyond

19:16

the control of its human inventors?

19:18

Essentially, the biggest unsolved

19:21

problem is how do you ensure

19:23

both at sort of a scientific understanding

19:26

level and then the practical engineering level

19:29

that you can make sure that

19:31

the AI never goes out of control? And

19:33

that's where I think there needs to be a

19:35

certain like project where both

19:38

the academics, along with

19:40

corporations and governments all

19:42

come together to perhaps solve

19:44

that alignment problem and accelerate the

19:46

solution to the alignment problem.

19:48

But even a certain like project after

19:50

the fact, once it's been made available

19:52

to the world, especially without watermarks

19:55

and so on, does it seem a little

19:57

backwards? Do you ever think that you're a good scientist?

20:00

excitement over the technology led you

20:02

and others to release it publicly too early?

20:04

No, I actually think first of all, we're in very early

20:07

days and there has been a lot of work. See,

20:09

there's no way you can do all of this

20:12

just as a research project. And we

20:14

spent a lot of time, right? In fact, if anything,

20:16

that, for example, all the work we did in launching

20:19

Bing Chat and the lessons learned in

20:21

launching Bing Chat is now all available

20:24

as a safety service, which, by the way, can

20:26

be used with any open source model. So

20:29

that's, I think, how the industry and the

20:31

ecosystem gets better at AI safety.

20:34

But at any point in time, anyone

20:36

who's a responsible actor does need

20:39

to think about everything that they can do for

20:41

safety. In fact, my sort of mantra internally

20:43

is the best feature of AI is AI

20:46

safety.

20:47

I did read, though, Satya, that as part

20:49

of a broader, a much broader, layoff earlier

20:51

this year that Microsoft laid off its entire

20:53

ethics and society team, which presumably

20:56

would help build these various guardrails

20:58

for AI, from the outside,

21:00

that doesn't look good. Can you explain that?

21:02

Yeah, I saw that article too. At the same time,

21:04

I saw all the headcount that was increasing

21:07

at Microsoft because it's kind of like saying,

21:09

hey, should we have a test organization

21:12

that is somewhere on the side? I think

21:14

the point is that work that AI

21:16

safety teams are doing are now become

21:19

so mainstream, critical

21:21

part of all product making that

21:24

we have actually, if anything, doubled down

21:26

on it. So I'm sure there was

21:28

some amount of reorganization and any reorganization

21:31

nowadays seems to get written

21:32

about, and that's fantastic. We love that.

21:35

But

21:35

to me, AI safety is

21:37

like saying performance or quality

21:40

of any software project. You can't separate

21:42

out. I want all 200,000 people

21:44

at Microsoft working on products to think

21:46

of AI safety.

21:49

One particular

21:49

concern about the future of AI is

21:52

how intensely concentrated the technology

21:54

is within the walls of a relatively

21:57

few firms and institutions. The economy

21:59

is a little bit

21:59

Economists Daron Acemoglu and

22:02

Simon Johnson recently published a book on

22:04

this theme called Power and Progress,

22:06

Our 1,000-Year Struggle Over

22:09

Technology and Prosperity. And

22:11

here's what they wrote in a recent New York Times

22:14

op-ed. Tech giants Microsoft

22:17

and Alphabet Google have seized a large

22:19

lead in shaping our potentially AI-dominated

22:22

future. This is not

22:24

good news. History has shown

22:27

us that when the distribution of information

22:29

is left

22:29

in the hands of a few, the result

22:32

is political and economic oppression.

22:34

Without intervention, this history

22:37

will repeat itself. Their

22:39

piece was called Big Tech is Bad,

22:42

Big AI Will Be Worse.

22:45

You could argue we are fortunate to have

22:47

a CEO as measured as Satya

22:50

Nadella leading the way at Microsoft. But

22:52

of course he won't be there forever. After

22:55

the break, what does a Microsoft

22:58

succession look like? I'm

22:59

Stephen Dubner. This is Freakonomics Radio.

23:02

We'll be right back.

23:06

No matter how strong

23:08

your portfolio is or how well you budget

23:10

your spending, the truth is life can

23:12

be unpredictable and anything could happen.

23:14

So how can you protect yourself from anything happening?

23:17

The answer might surprise you. An annuity or

23:19

life insurance policy from Lincoln Financial

23:21

might be what you need. Their annuities offer

23:23

options to help you safeguard your money against unpredictable

23:26

market swings, to help you manage sudden dips,

23:28

and can provide you with a steady income during

23:30

retirement. Their life insurance policies have

23:32

a death benefit for your loved ones. Plus, they

23:35

offer some access to the cash in your policy

23:37

in case of an emergency. Because anything happening

23:39

sometimes happens. Go to LincolnFinancial.com

23:42

slash Get Started Now to learn how to plan,

23:44

protect, and retire today. Lincoln

23:47

annuities and life insurance are issued by the Lincoln National Life

23:49

Insurance Company, Fort Wayne, Indiana. Products sold

23:51

in New York are issued by Lincoln Life & Annuity Company

23:53

of New York, Syracuse, New York. Distributed

23:55

by Lincoln Financial Distributors, Inc., a broker-dealer.

23:58

Freakonomics Radio.

23:59

is sponsored by Whole Foods Market. Nothing

24:02

says summer fun like grilling outside

24:05

with help from Whole Foods Market. From

24:08

family reunions to hanging with friends,

24:10

you will find the perfect spread, like

24:12

animal welfare certified bone-in

24:15

ribeye, top sirloin, and chicken

24:17

kebabs. You can just imagine those

24:19

perfect grill marks. Keep your

24:21

wallet happy with sides and fixings from 365

24:24

by Whole Foods Market, plus beer,

24:27

pie, fresh produce, the list goes

24:29

on.

24:29

Can't take the heat? Let them cater

24:32

your event. Make summer happen at

24:34

Whole Foods Market.

24:37

["The Whole

24:50

Foods Market"] ["The

24:56

Whole Foods Market"] ["The

25:26

Whole Foods Market"] ["The

25:56

Whole Foods Market"]

25:59

Rather than try to come

26:02

up with long lists of ways

26:04

of vilifying predecessors,

26:07

what Nadella did is he

26:09

was able to be on a frontier

26:12

at this exact same moment as the early investors

26:14

in open AI as well as in

26:17

reinventing their own artificial

26:19

intelligence opportunities so

26:22

that Bing, surprised at all, might

26:24

soar past everybody. He got

26:26

people excited about building a new

26:28

future, investing $25 billion in R&D each year.

26:33

That's perhaps twice as much as the average pharma company

26:35

invests, and that's amazing for an

26:37

IT company to do that.

26:39

A big part of Nadella's success

26:41

came from expanding Microsoft's

26:43

footprint in cloud computing with their

26:46

Azure platform.

26:47

Their footprint across the board at Enterprise

26:49

Software was flourishing, where he knew

26:52

how to invest in Azure and a commercial

26:54

cloud business where his revenues

26:56

grew 42% over the past year.

26:58

I asked Nadella himself if he

27:01

had been surprised by how valuable

27:03

cloud computing has become for Microsoft.

27:05

Both surprised and not surprised in

27:08

the following sense. We were leaders in client

27:10

server, but while we were leaders

27:12

in client server, Oracle did

27:14

well, IBM did well, and

27:17

so in fact it shaped even my

27:19

thinking of how the cloud may sort

27:21

of emerge, which is that it will have

27:23

a similar structure. They will be at least two

27:26

to three players who will be at scale,

27:28

and there will still be many other smaller

27:30

niche players perhaps. So in that

27:33

sense, it is not that surprising. What

27:35

has been surprising

27:35

is

27:37

how big and expansive

27:40

the market is. Let's think about

27:42

it. We sold a few servers in India,

27:44

but oh my god, did I think that

27:47

cloud computing in India would be this

27:49

big? No. The market is

27:51

much bigger than I originally thought.

27:54

I have a fairly long and pretentious

27:56

question to ask you. There are economists

27:59

and philosophers.

27:59

and psychologists who argue that most

28:02

of us still operate under a scarcity

28:04

mindset that might have been appropriate on

28:06

the Savannah a million years ago, but now

28:09

we live in an era of abundance. So

28:12

rather than competing for scarce resources,

28:15

we should collaborate more to grow

28:17

the overall resource pool. From

28:19

what I know about your time

28:22

as CEO at Microsoft, it seems you

28:24

have embraced the collaborative model over the

28:26

competitive model. One example being how nicely

28:28

Microsoft now plays with Apple devices,

28:30

whereas the previous administration didn't even

28:33

want Microsoft employees owning Apple

28:35

devices. So I'd like to hear your

28:38

thoughts generally on

28:40

this idea of collaboration versus

28:43

competition and scarcity versus

28:45

abundance.

28:47

That's

28:48

a very deep question. I

28:50

mean, at the macro level, Stephen,

28:53

I actually do believe that

28:55

the best technique humanity

28:58

has come up with to create,

29:00

I would say, economic

29:03

growth and growth in our well-being

29:05

as humanity is through cooperation.

29:08

So let's start there, right? So the more

29:11

countries cooperate with countries, people

29:13

cooperate with people, corporations cooperate

29:16

with other corporations, the better off we are.

29:19

And then at a micro level,

29:21

I think you want to be very careful

29:24

in

29:24

how you think about zero-sum

29:27

games, right? I think we overstate

29:30

the number of zero-sum games that

29:32

we play. In many cases,

29:34

I think growing your overall

29:37

share of the pie is probably

29:40

even more possible when the pie itself

29:42

is becoming bigger. So I've always approached it

29:44

that way. That's kind of how I grew up, actually,

29:46

at Microsoft. And so

29:49

all of what we have done in the last whatever,

29:51

close to 10 years, has been to look

29:54

at the

29:54

opportunity set first

29:57

as something that expands the opportunity

29:59

for all. players and in their being competitive.

30:02

Were there people within the firm though who

30:05

said or felt, wait a minute, I

30:07

know you're the new CEO and I know you have

30:09

a new way of doing things, but Google is our enemy.

30:12

Apple is our enemy. We can't do that.

30:14

Did you have pushback?

30:15

Yeah, I mean, look, it's a very fierce

30:17

competitive industry. And even

30:19

if we didn't think of them as our competitors,

30:22

our competitors probably think of us

30:24

as competitors. But I think at the end of the day, I think

30:27

it helps to step back and say, you

30:29

know, it doesn't mean that you back away from some

30:31

real zero sum competitive

30:34

battles, because after all, that's kind of what fosters

30:36

innovation. And that's what creates consumer surplus

30:38

and opportunity. And so that's all fine.

30:41

But at the same time, leaders in

30:43

positions like mine have

30:45

to also be questioning

30:50

what's the way to create economic

30:52

opportunity. And sometimes,

30:55

you know, construing it as zero sum is probably

30:57

the right approach. But sometimes it's not.

30:59

So

31:00

Microsoft is a huge company and huge

31:02

companies get bigger by acquisition.

31:05

Typically, let's go through a couple. I know you

31:07

tried a few times to buy Zoom. You

31:09

haven't succeeded yet. You're

31:11

still in the middle of trying to acquire Activision. That's

31:14

tied up in the US at least in an FTC

31:16

lawsuit. A few years

31:18

ago, I read you tried to buy TikTok.

31:20

You called those negotiations the

31:22

strangest thing I've ever worked on. What

31:25

was so strange about that?

31:26

At least let me talk to all the acquisitions

31:29

that we did that actually have succeeded.

31:31

And we feel thrilled about it, right? Whether it's

31:33

LinkedIn or GitHub or Nuance

31:36

or ZeniMax or Minecraft.

31:39

These are all things that we bought.

31:41

I feel that these properties are better

31:44

off after we acquired them because

31:46

we were able to innovate and then make

31:48

sure that we straight through to the core

31:51

mission of those products and those customers

31:53

who depended on those products.

31:55

What about TikTok though? What was so strange

31:57

about that negotiation or those conversations?

32:00

everything. First of

32:02

all, I mean, just to be straight

32:04

about it, TikTok came

32:06

to us because they at that time

32:09

sort of said, hey, we need some help

32:11

in thinking about our structure

32:14

and given what at that time at least

32:16

was perceived by them as some

32:18

kind of a restructuring that the

32:20

United States government was asking.

32:22

They needed a US partner, in other words,

32:24

yes. Yeah, so at that point we

32:27

said, look, if that is the case that you want

32:29

to separate out your US operations or worldwide

32:31

operations, we would be interested in being engaged

32:34

in a dialogue and it is just

32:36

that's just say an interesting

32:38

summer that I spent on it. Okay,

32:41

so not long ago Satya, you became the chair

32:43

of the Microsoft board in addition to CEO.

32:46

Now, a lot of corporate governance people

32:48

hate the idea of one person having

32:51

both jobs. I

32:52

asked chat GPT about it. What's

32:54

the downside? One potential

32:56

conflict of interest, chat GPT told me

32:58

is the roles of CEO and board chair can

33:01

sometimes be at odds. The CEO is typically focused

33:03

on the day to day yada yada, but there can be potential

33:06

conflicts of interest. Can you give

33:09

an example of one

33:11

conflict that you've had or maybe you haven't,

33:14

which would give the Corp governance people even

33:16

more headache?

33:17

The reality is we have a lead independent

33:20

director, a fantastic lead independent

33:22

director in Sandy Peterson. She

33:24

has the ultimate responsibility of hiring

33:26

and firing me. That said, I think

33:28

the chair role as I see it is

33:30

more about me being able to sort of,

33:33

you know, having been close to 10 years in my role

33:35

to use my knowledge of what it is that

33:37

Microsoft's getting done in the short and the long

33:40

run to be able to coordinate the board

33:42

agendas and make sure that the topics

33:44

that we're discussing are most helpful

33:47

for both the board

33:47

and the management team. And so

33:50

it's kind of as much about, you know, program

33:52

managing the board versus being responsible

33:54

for the governance of the board. And the governance

33:57

of the board is definitely with the independent directors.

33:59

Can you name a time when the board

34:01

voted down a big idea of yours?

34:04

I don't know. There is a particular vote that

34:06

they voted me down, but I take all of

34:08

the board feedback on any idea

34:11

that I or my management team

34:13

has. We have a good format where

34:15

every time we get together, we kind of do a left

34:17

to right, I'll call it overview

34:20

of our business. And we have a written doc,

34:23

which basically is a living document

34:25

which captures our strategy and performance

34:27

and having that rich discussion where

34:29

you can benefit from the

34:32

perspective of the board and

34:34

then change course based on that perspective

34:36

is something that I look forward to and I welcome.

34:39

Now the last time we spoke, which was several

34:41

years ago, you talked about how the birth of

34:44

your son Zane changed you

34:46

a great deal. He was born with cerebral palsy

34:48

and you said that empathy didn't

34:51

come naturally to you, certainly

34:53

not compared to your wife, but

34:55

that over time being a parent to a

34:57

child with a severe handicap was a powerful

35:00

experience for you on many levels. I

35:03

was so sorry to read

35:05

that Zane died not long ago in

35:07

just his mid 20s. So my

35:09

deepest condolences on

35:12

that Satya. I'm also

35:14

curious to know if or how

35:17

his death has changed

35:19

you as well. No,

35:20

I appreciate that, Stephen. It's

35:23

probably it's hard, Stephen, for me

35:25

to even reflect on it

35:28

that much. It's been for both

35:30

my wife and me.

35:31

In some sense, he was the one sort of

35:35

constant that gave us a lot of purpose,

35:37

I would say, in his short life.

35:39

And so I think, you know, I think we're still

35:42

getting through it and it'll, I think,

35:44

take time. But I just

35:47

say the thing that I perhaps

35:49

have been more struck by is

35:52

what

35:53

an unbelievable support

35:56

system that got built

35:58

around us in.

35:59

even the local community around Seattle.

36:02

At his memorial, I look back at it,

36:04

all the people who came, right? All

36:06

the therapists, the doctors, the

36:08

friends, the family, the colleagues

36:11

at work. I even was thinking about it, right?

36:13

After all, Zane was born when I was working at

36:15

Microsoft and he passed when I was working at Microsoft

36:19

and everything, even from the benefits

36:21

programs of Microsoft to the managers

36:23

who gave me the flexibility. I think that sort

36:25

of was a big reminder to me that all of us

36:27

have things happen in our lives. Sometimes

36:30

things like pandemics or the passing of a

36:32

loved one or the health issues of elderly

36:35

parents. And we get by

36:38

because of the kindness of people

36:40

around us and the support of communities

36:42

around us. And so if anything, both

36:44

my wife and I have been super, super thankful

36:47

to all the people and

36:49

the institutions that were very much part

36:52

of his life and thereby part of our lives.

36:56

You

36:56

are a young man, still 55 years

36:58

old, but you've been at Microsoft

37:00

a long time now, been CEO almost 10 years.

37:02

I'm curious about a succession plan, especially,

37:06

I don't know if you watched the HBO show, Succession.

37:09

Do you watch Succession, Satya or no? I

37:13

watched, I think, the first season a bit

37:15

and I was never able to get back to it. Okay,

37:18

so I'll give you a small spoiler. It doesn't go

37:20

well. And there are succession plans

37:23

turns out to be, I think the technical term

37:25

is total show, okay?

37:26

So I am curious if your succession

37:29

plan will be somewhat more orderly than

37:31

the succession plan on succession. Obviously,

37:34

the next CEO of Microsoft is going to

37:36

be appointed by the lead independent directors of

37:39

Microsoft and not by me. But

37:42

to your point, it's a board topic when

37:45

we have a real update on it every year

37:47

as it should be. And I take that as a serious

37:50

job of mine. Like one of

37:52

the things that I always say is long

37:55

after I'm gone

37:56

from Microsoft, if Microsoft's doing

37:58

well,

37:59

then maybe I did a decent

38:02

job, because I always think about the strength

38:04

of the institution long

38:06

after the person is gone is the

38:08

only way to measure the leader. I'm

38:11

very, very suspicious of people who

38:13

come in and say, before me, it was horrible,

38:15

and during my time, it was great, and after

38:17

me, it is horrible. I mean, that's, first of all,

38:19

means you didn't do anything to build institutional

38:21

strength. So yes, I take that

38:24

job that I have in terms of surfacing

38:26

the talent and having the conversation

38:28

with the board of directors

38:29

seriously. And when the time

38:32

comes, I'm pretty positive that they will

38:34

have a lot of candidates internally, and

38:36

they'll look outside as well. And so

38:38

yes, we will take succession seriously.

38:41

That was Satya Nadella, CEO of Microsoft.

38:46

His

38:50

intelligence, I think you will agree, doesn't

38:52

feel artificial at all.

38:57

Coming up next time on the show. Most

38:59

people, when they think about marriage, they think

39:02

about it in terms of preferences

39:04

and in terms of love. But economists

39:08

aren't most people. So this

39:10

idea is what encapsulates

39:12

the idea of the marriage market. Is

39:14

marriage really a market?

39:16

I think people truly misunderstand

39:19

these dating services. Why

39:22

did you marry that person? That's

39:24

next time on the show. Until then, take

39:26

care of yourself. And if you can,

39:29

someone else too. Freakonomics

39:32

Radio is produced by Stitcher and Renbud Radio.

39:34

You can find our entire archive on any

39:37

podcast app or at freakonomics.com,

39:39

where we also publish transcripts and show

39:42

notes. This episode was produced

39:44

by Zach Lipinski with research

39:46

help from Daniel Moritz-Rabson. It

39:48

was mixed by Greg Rippin with help from

39:50

Jeremy Johnston. Our staff also

39:52

includes Alina Kullman, Eleanor Osborne,

39:59

Catherine Moncure, Lyric Bowditch, Morgan

40:02

Levy, Neil Caruth, Rebecca Lee Douglas,

40:04

Ryan Kelly and Sarah Lilly. Our

40:06

theme song is Mr. Fortune by the

40:08

Hitchhikers. All the other music was composed

40:11

by Luis Guerra. As always,

40:13

thanks for listening.

40:15

I blew an opportunity here. I need to

40:17

ask ChatGPT how to get over intense

40:20

disappointment at myself.

40:28

The Freakonomics Radio Network.

40:31

The hidden side of everything.

40:40

Old Dominion Freightline was built

40:42

on keeping promises. With an industry leading

40:44

on-time delivery record and low claims rate,

40:47

we keep promises better than any other LTL

40:49

freight carrier because we treat every shipment

40:51

like it's our most important one, which means

40:54

we do the little things right so that we can

40:56

keep our promises and you can keep yours

40:58

too. That's what drives us. To

41:00

learn how OD can help your business keep

41:02

its promises, visit odfl.com.

41:05

Old Dominion, helping the world keep promises.

41:10

As the weather warms, Ohioans

41:13

are finding more ticks. Avoiding

41:15

tall grass and wearing light-colored,

41:17

long pants and long-sleeved shirts

41:20

are just a few things you can do to protect

41:22

yourself. After being outdoors, be

41:24

sure to check for ticks and if you find

41:27

any, remove them and watch

41:29

for any symptoms if you find any bites.

41:32

For a full list of ways to prevent tick

41:34

bites, visit ohio.gov

41:37

slash ticks.