Latest / Elon Musk Podcast / Nadella says Musk never raised concerns to him about Microsoft

Transcript

Nadella says Musk never raised concerns to him about Microsoft · Elon Musk Podcast

0:00
Elon Musk is suing Sam Altman, Greg Brockman and Microsoft for
0:04
$150 billion, claiming they stole a charitable organization
0:08
to create an $852 billion artificial intelligence empire.
0:14
That financial demand is, I mean, it is staggering.
0:18
And the accusation really strikes at the very foundation
0:21
of corporate law because, you know, this entire dispute
0:24
centers around an organization that originally presented itself
0:27
to the world as a strict nonprofit.
0:29
The stated goal was purely altruistic, right?
0:32
The founders declared they were dedicating the organization
0:35
entirely to protecting humanity from the potential dangers of
0:38
artificial intelligence. But well, to survive the brutal
0:42
economic realities of tech development, that organization
0:46
completely restructured, they adopted A capped profit model,
0:49
took in billions upon billions of dollars in corporate
0:51
investment, and transferred all of their valuable intellectual
0:54
property into a commercial entity.
0:56
So can a mission to protect humanity actually survive the
0:59
financial gravity of a trillion dollar technology market?
1:03
That is the exact tension driving this entire courtroom
1:06
battle. Yeah.
1:07
And we have to look at how this started to understand the
1:09
magnitude of that shift. Elon Musk, the plaintiff here,
1:14
provided the critical early funding.
1:16
He donated 10s of millions of dollars to start this artificial
1:19
intelligence lab. He did, yeah.
1:20
And he personally helped recruit the initial engineering and
1:23
research talent too. Right, because the fundamental
1:26
agreement, the entire premise of the organization, was that this
1:29
would be a nonprofit open source operation designed specifically
1:32
to serve as a counterweight to Google's dominance in the tech
1:35
sector. Exactly.
1:37
The idea was that if one massive, profit driven
1:41
corporation controlled the most powerful technology ever
1:43
created, well, that would be incredibly dangerous for
1:46
society. So the solution was to create an
1:49
open research lab that effectively belonged to
1:51
humanity. You fund it with donations.
1:53
You publish the research freely for anyone to use, and you keep
1:56
the profit motive entirely out of the equation.
1:59
So the engineers can focus on safety rather than quarterly
2:02
earnings, right? But that philosophical ideal, it
2:05
immediately crashed into a fundamental physical roadblock.
2:09
I mean, the astronomical cost of computing power.
2:11
Greg Brockman, the president of the company, actually testified
2:14
in court that their internal financial projections showed
2:17
their computing bills were on track to rise to $50 billion.
2:20
Wait. Backup.
2:22
Why does creating this software cost $50 billion?
2:26
I mean I know no hardware is expensive and semiconductor
2:28
companies are making a killing right now, but 50 billion is the
2:32
gross domestic product of a small country.
2:35
Are they actually buying the physical hardware or just
2:38
renting server time? How does the math on that even
2:41
work? Well, it requires looking at
2:43
what artificial intelligence actually is at a physical level.
2:47
OK, traditional software is highly efficient.
2:49
A human being writes specific lines of code, compiling logic
2:52
into a program that a standard computer can run easily.
2:55
Yeah, you could run complex traditional software on a basic
2:58
consumer laptop because the computer is just following
3:00
predefined instructions. Exactly.
3:03
But the system this lab was building is fundamentally
3:05
different. They were building massive
3:06
neural networks. You do not write the logic for a
3:09
neural network. You build a blank architecture,
3:12
and then you force that architecture to process
3:15
unimaginable amounts of data until it recognizes patterns and
3:19
forms its own internal logic. So you are basically throwing an
3:23
entire library of information at a blank wall until the computer
3:27
learns how to read the books on its own.
3:29
That is a great way to put it, yeah.
3:31
And throwing that data requires immense, sprawling physical
3:35
infrastructure. We are talking about data
3:37
centers the size of massive shopping malls.
3:39
Inside those buildings are thousands upon thousands of
3:42
highly specialized, incredibly expensive servers.
3:46
Each server contains multiple graphics processing units.
3:49
These units are designed to perform simple mathematical
3:52
calculations simultaneously. Millions of calculations
3:55
happening in parallel billions of times a second.
3:57
And I am assuming pushing that much data through that many
4:01
chips requires A tremendous amount of power.
4:03
Oh, the electricity demands are staggering.
4:07
A single data center running these training models can
4:09
consume more electricity than a mid sized city.
4:13
You have to pull power directly from the grid at an industrial
4:16
scale. That is wild.
4:18
But the electricity used to power the chips is only half the
4:21
equation. Actually, when you run thousands
4:24
of chips at maximum capacity, they generate an enormous amount
4:27
of physical heat. Just to keep the machines from
4:29
literally melting down requires massive industrial cooling
4:33
systems. You have complex plumbing
4:35
systems pumping thousands of gallons of chilled water through
4:39
the server racks continuously. So the infrastructure required
4:42
just to prevent the building from Catching Fire is a multi
4:45
$1,000,000 engineering project on its own.
4:47
Precisely. That makes sense when you were
4:49
trying to simulate human reasoning.
4:51
You are pushing electricity through microchips at an
4:54
unprecedented scale. Right.
4:55
And during the early days of this lab, one of their major
4:58
projects involved training an artificial intelligence model to
5:01
play and beat human champions in highly complex strategic video
5:05
game tournaments. Oh right, I remember that.
5:08
Yeah, training a system to process that much visual
5:11
information, to learn strategy from millions of simulated
5:15
matches, It requires A continuous, uninterrupted flow
5:18
of massive computing power. The model plays against itself
5:22
millions of times a day to learn from its mistakes.
5:25
You are essentially renting supercomputers by the minute,
5:28
and the meter is always running. Every single second of training
5:32
burns through cash. And this physical reality
5:35
completely limits the viability of purely donor funded research.
5:39
You simply cannot fund $50 billion in server costs through
5:43
charitable galas and philanthropic grants.
5:45
I mean, even the most generous billionaires in the world are
5:48
not going to write a recurring $50 billion check just to cover
5:51
a nonprofit's electricity and server rental bills.
5:54
You hit on the exact mechanism that forced the change.
5:56
This changes the organization's trajectory entirely.
5:59
It opens up the absolute necessity for corporate backing.
6:02
The leadership team looked at the math and realized that to
6:05
achieve their stated goal of building advanced artificial
6:09
intelligence diligence well, they had to abandon the pure
6:12
nonprofit model. The computational requirements
6:16
forced a transition into a capped profit entity.
6:20
They needed a structure they could offer a massive financial
6:23
return to private investors in order to attract the billions of
6:27
dollars required to rent the service.
6:29
It's like starting a neighborhood community garden to
6:31
feed the poor, only to realize that the soil is bad so you need
6:34
fertilizer. Then the local well runs dry so
6:38
you need to build a water pipeline.
6:39
Before you know it, you need to purchase a global agricultural
6:43
conglomerate just to afford the water bill for your little
6:45
community plot. That is a very accurate analogy.
6:48
You start with this pure localized intention to do good,
6:52
but the physical resource requirements force you to become
6:54
the exact type of corporate machine you were initially
6:57
trying to provide a safe alternative to.
6:59
And, well, the necessity for endless computing power leaves
7:02
directly into the arms of the one entity that could actually
7:05
provide it on a global scale. Right, Microsoft.
7:08
Yes, Microsoft enters the picture.
7:10
They offer a multibillion dollar investment to keep the
7:14
artificial intelligence lab alive.
7:16
However, the structure of this investment is incredibly
7:19
specific. How so?
7:21
Microsoft did not just wire billions of dollars of liquid
7:24
cash into the lab's checking account.
7:27
The investment was heavily structured around providing
7:29
cloud computing credits on Microsoft's own proprietary
7:33
servers. Wait, so Microsoft wasn't just
7:35
handing over cash, they were basically paying the lab in
7:38
company store tokens? They give the lab billions in
7:41
credits, but those credits can only be spent renting
7:44
Microsoft's own hardware. That is the core of the
7:47
arrangement. Yeah, the Artificial
7:49
Intelligence Lab gets the computing power they desperately
7:51
need to train their models, but they are entirely locked into
7:54
Microsoft's ecosystem. Oh wow.
7:56
They tune their software to run efficiently on Microsoft's
7:59
specific hardware setup. This creates vendor lock in.
8:03
Which is fascinating because an internal e-mail from a Microsoft
8:06
executive surfaced during the trial and it shows they knew
8:09
exactly what was happening behind the scenes.
8:11
Oh, absolutely. This executive explicitly
8:14
questioned if the original donors knew that an open source
8:17
effort to gather top talent was being used to build a closed
8:20
commercial product on its back. That internal communication is a
8:24
crucial piece of evidence for the plaintiff.
8:27
It highlights the ethical and financial realities of this
8:30
partnership. Microsoft secured a massive
8:32
equity stake in the newly formed Capped Profit subsidiary, Right?
8:36
And through this arrangement, Microsoft has already recognized
8:39
billions of dollars in actual revenue from the partnership
8:43
because, well, when the lab uses those cloud credits, it
8:47
registers as consumption on Microsoft's balance sheet,
8:50
boosting their cloud divisions metrics.
8:52
So Microsoft gets to report massive usage numbers to Wall
8:55
Street, which pumps up their own stock price while simultaneously
8:59
owning a huge chunk of the most valuable artificial intelligence
9:02
startup in the world. Exactly.
9:04
Furthermore, expert financial testimony presented in court
9:08
revealed the true scale of the windfall.
9:10
The value of Microsoft's cloud credits has gained over $100
9:13
billion in value. As the Artificial Intelligence
9:17
Labs valuation skyrocketed, Microsoft provided the
9:20
infrastructure and in return, their investment multiplied
9:24
exponentially. So the cloud credits were
9:26
essentially a casino chip that paid out $100 billion.
9:29
Pretty much. And this dynamic completely
9:31
limits the artificial intelligence labs independence.
9:34
It changes them from a sovereign open Research Institute into a
9:38
dependent entity tied exclusively to 1 corporate
9:42
infrastructure. They rely on Microsoft servers
9:44
to exist. Yes, if Microsoft turns off the
9:47
servers, the lab ceases to function.
9:49
This opens up the plaintiff's specific legal claim that
9:53
Microsoft actively aided and abetted a breach of charitable
9:57
trust. How does that specific legal
9:59
claim work? Because aiding and abetting a
10:02
breach of trust sounds like a criminal charge, but this is
10:05
corporal litigation. Right.
10:06
So in the context of a charitable trust, the argument
10:09
is that Microsoft knowingly facilitated the dismantling of a
10:12
charity to secure an exclusive commercial advantage.
10:15
OK, I see the plaintiff is arguing that Microsoft looked at
10:18
A5O1C3 charity, saw the incredible technology the
10:23
researchers were building, and use their massive server
10:26
infrastructure as leverage to pry that technology out of the
10:30
public domain. And into a closed commercial
10:32
structure that Microsoft could control and profit from.
10:35
Exactly. So the money gets huge and the
10:37
people get complicated. Very complicated.
10:39
You go from a group of idealistic researchers trying to
10:42
save the world to executives managing $100 billion valuation
10:47
swings. And the trial exposed a wealth
10:50
of internal documents detailing this internal shift.
10:53
One of the most revealing pieces of evidence was Greg Brockman's
10:56
private diary. Oh right, the diary.
10:58
Yeah. In an entry written during the
11:00
period when the company was exploring the restructuring from
11:03
a pure nonprofit to the capped profit model, he asked himself
11:07
what it would take for him to reach $1 billion financially.
11:10
Wow. And today, his personal stake in
11:12
the capped profit entity is valued at $30 billion.
11:16
Hold on. $30 billion. And he did not invest his own
11:21
personal funds to acquire that equity, right?
11:25
He acquired that through his role as a founder when they
11:29
restructured the company. Correct.
11:31
He acquired founder shares in the new capped profit entity.
11:35
He contributed his labor and leadership essentially sweat
11:38
equity rather than fronting billions in cash.
11:41
You look at a $30 billion personal stake and you contrast
11:45
that with the initial donors who funded the labs creation.
11:49
The people who donated the original cash got a charitable
11:52
tax deduction and nothing else. The founders who managed the
11:55
transition ended up with 10s of billions of dollars in personal
11:59
wealth. That alone makes the jury look
12:01
very closely at their motivations.
12:03
Absolutely. But the testimony regarding the
12:04
internal culture is even more intense.
12:06
Former executives, including Mira Morati, testified about a
12:10
deeply ingrained culture of lying at the executive level.
12:13
Yeah, Morati's testimony provided a look inside the daily
12:16
operations during the period of hyper growth.
12:19
She claimed the chief executive officer actively hid safety
12:22
reviews from the staff and from the nonprofit board of
12:24
directors. She testified that he pushed a
12:27
massive, highly aggressive revenue target that was
12:30
explicitly mandated by Microsoft in order to secure further
12:34
rounds of funding. OK, I have to push back a little
12:36
here though. If you look at the reality of
12:39
the tech industry, these executives were simply doing
12:43
what was required to survive and ship products in a hyper
12:47
competitive market. How do you mean?
12:49
Well, we just established that they were burning millions of
12:52
dollars a day on server compute. If they do not hit those revenue
12:56
targets, they do not get the next tranche of server credits
12:59
from Microsoft. If they do not get the server
13:02
credits, the company dies, the research stops entirely, and a
13:06
competitor wins the race to build the technology.
13:08
That is the practical business argument, yes.
13:10
Right, so hiding safety reviews might be terrible management,
13:13
but pushing hard for revenue is just how a startup stays alive
13:17
when the burn rate is that high. The problem with that defense,
13:20
though, is that you cannot separate the management tactics
13:23
from the legal structure they agreed to operate under.
13:26
Actively hiding safety reviews directly contradicts the
13:29
founding charitable mission of protecting humanity.
13:32
When they incorporate it as a nonprofit, they legally bound
13:35
themselves to that specific mission.
13:38
That mission is not just a marketing slogan, it is the
13:40
entire legal basis of the lawsuit.
13:43
But surely the law allows a nonprofit to generate revenue to
13:47
sustain its operations. I mean, hospitals and
13:49
universities are nonprofits and they charge for their services
13:52
to keep the lights on. They do, but the generation of
13:54
revenue must serve the charitable mission, not
13:57
superseded. If the leadership is ignoring
14:00
safety protocols specifically to chase commercial revenue
14:03
mandated by a corporate partner while they are violating the
14:06
fiduciary duty of the nonprofit. I see.
14:09
Fiduciary duty in a charitable trust means the directors must
14:12
put the mission of the charity above all other concerns,
14:15
including financial survival. If the only way to survive is to
14:19
abandon the mission and build unsafe products for profit, the
14:22
legal requirement is to let the organization fail.
14:25
Wow, really? Yes, you do not pivot to a
14:27
commercial enterprise and enrich yourself.
14:29
So by hiding the safety reviews to hit the revenue targets, they
14:33
were effectively proving that the commercial demands of
14:36
Microsoft had taken priority over the charitable mission to
14:40
protect humanity. Exactly.
14:42
This internal behavior completely changes the
14:45
perception of the leadership team from altruistic researchers
14:48
to ruthless corporate executives.
14:51
It limits their ability to claim in front of a jury that they
14:54
were acting purely for the public benefit, and this opens
14:57
up severe legal vulnerability regarding the unjust enrichment
15:01
charges. An unjust enrichment is that
15:04
legal concept which states that if you use assets that belong to
15:07
a charity to make yourself rich, the law requires you to return
15:11
those financial gains. Correct, you cannot use a
15:14
charity as an incubator for your own personal wealth.
15:17
That makes total sense. The legal obligation to a
15:19
charity does not just disappear because the business environment
15:22
gets tough or the computing costs get too high.
15:25
You made a promise to the public and you have to honor it.
15:28
Exactly. But while the Labs leadership
15:30
faces immense scrutiny for their greed, the plaintiff's own
15:33
motives are dragged right into the light during cross
15:36
examination. Yes, the defense strategy relies
15:39
heavily on exposing the plaintiff's actions to dismantle
15:42
his moral high ground. The plaintiff's own allies and
15:45
former associates testified against his purely altruistic
15:49
narrative. Oh yeah, this part is wild.
15:51
They revealed internal discussions and emails showing
15:54
that during the early struggles to fund the compute costs, he
15:56
actively wanted to merge the nonprofit artificial
15:59
intelligence lab into his own publicly traded car company to
16:02
use it as a cash cow. Hold on, the guy's suing them
16:06
for becoming a for profit. The guy claiming he is the
16:09
righteous defender of the charitable trust wanted to
16:12
absorb the charity into his own for profit car company.
16:16
That is exactly what the testimony confirmed.
16:18
He recognized the immense value of the research team at the
16:22
time. His car company was trying to
16:24
solve the problem of autonomous driving, which is fundamentally
16:27
an artificial intelligence problem.
16:28
Right, because you need a system that can understand and navigate
16:31
the physical world. Yes.
16:32
So he saw a natural synergy. He demanded majority ownership
16:36
and the chief executive role over the Artificial Intelligence
16:39
lab to facilitate this merger. And what was his justification
16:43
for taking over the charity? His stated reasoning, according
16:46
to the testimony, was that he needed massive funds to build a
16:49
city on Mars. Wait, really, Mars?
16:53
Yes, he viewed the artificial intelligence technology as an
16:56
engine to generate the trillions of dollars in capital required
16:59
for interplanetary colonization. He intended to commercialize the
17:03
technology through his car company to fund his space
17:06
exploration goals. That is a stunning level of
17:09
contradiction. You cannot sit in a courtroom
17:12
and claim you're defending a charity from corporate greed
17:15
when your own proposed alternative was to take it over,
17:18
monetize it to build rockets and make yourself the CEO.
17:21
It is a massive contradiction, and on top of that, he admitted
17:25
on the witness stand that his own entirely separate competing
17:28
artificial intelligence company currently uses distilled
17:31
versions of the exact models he is suing the defendants over.
17:35
What does that mean technically using distilled versions of the
17:37
model models? Distillation is a process in
17:40
artificial intelligence where you take a massive, highly
17:43
capable model that costs billions of dollars to train,
17:46
and you use it to train a smaller, cheaper model.
17:49
Oh OK. The smaller model learns from
17:51
the outputs of the larger model. It is a way to shortcut the
17:54
expensive training process. The plaintiff admitted that his
17:58
new startup relies on the outputs of the commercial models
18:01
built by the defendants, so he is directly benefiting from the
18:05
very commercialization he is suing to stop.
18:08
Precisely. That just completely undermines
18:10
his entire argument. He is using the commercial
18:12
products to build his own commercial products while suing
18:15
them for being commercial. And it gets worse for him.
18:18
The defense introduced text messages he sent to the founders
18:21
during the dispute, threatening to make them the most hated men
18:23
in America if they did not comply with his demands to hand
18:26
over control. Wow.
18:28
Yeah, this behavior severely limits the plaintiff's moral
18:31
high ground in the eyes of the jury.
18:33
It changes the narrative from a righteous defense of a betrayed
18:36
charity to a bitter personal dispute between rival
18:40
billionaires fighting for control over the future of
18:43
technology. It suggests to the judge and
18:45
jury that his core objection is not actually that the
18:48
organization became a for profit entity, but rather that it
18:52
became a highly lucrative for profit entity that he did not
18:56
personally control and profit from.
18:58
Exactly. And the question of who truly
19:00
controls the company comes to a head during a crisis involving
19:04
the nonprofit board. I mean, you have this bizarre
19:08
structure where a nonprofit board of directors legally
19:12
controls a massive capped profit it.
19:16
Is a highly unusual structure. We have to look at the dramatic
19:19
weekend when the nonprofit board actually exercised its power and
19:22
fired the chief executive officer, citing a lack of candor
19:26
in his communications. For a brief moment, the
19:28
nonprofit board did exactly what it was designed to do.
19:31
It rained in the commercial leadership to protect the
19:33
mission. Right, because the legal
19:35
structure of the organization gave the nonprofit board
19:37
absolute authority. They did not need shareholder
19:40
approval to remove the CEO. They felt he was lying to them
19:43
about safety metrics and commercial agreements, so they
19:46
terminated his employment. In theory, this proved that the
19:49
nonprofit was still in charge. In theory, yes.
19:52
But the immediate aftermath of that firing demonstrated where
19:55
the true power resided. Microsoft staged an overwhelming
19:59
intervention. The chief executive of Microsoft
20:03
publicly offered to hire the recently fired CEO and the
20:06
entire staff of the Artificial Intelligence Lab.
20:09
That is a brilliant and brutal corporate maneuver.
20:13
They didn't try to buy the company, which would trigger
20:15
antitrust reviews and lawsuits. They just offered to buy the
20:18
people. It was a highly calculated move
20:21
to prevent competitors from acquiring the specialized
20:23
talent. An artificial intelligence
20:25
company without its researchers is just a pile of expensive
20:28
servers. By offering to absorb the entire
20:31
workforce and match their salaries, Microsoft effectively
20:34
neutralized the nonprofit board's decision.
20:37
If the board did not reverse course and reinstate the CEO,
20:41
they would be left ruling over an empty shell of a company with
20:44
no employees. While Microsoft instantly
20:46
acquired the best artificial intelligence team in the world,
20:49
and the text messages revealed during the trial show exactly
20:53
how hands on Microsoft was during the reinstatement process
20:56
over that weekend, the messages show Microsoft executives
21:00
actively vetting potential new board members for the nonprofit.
21:04
When certain independent names were suggested for the board,
21:07
Microsoft executives responded with phrases like strong, strong
21:11
No. And this event completely limits
21:13
the defense's argument that the nonprofit board remained fully
21:17
in charge of the organization. It changes the legal
21:20
understanding of corporate control, opening up intense
21:23
regulatory scrutiny. It proves that a corporate
21:26
partner with absolutely no official voting rights on the
21:29
board, no legal authority to appoint directors, could
21:32
completely overrule the independent directors of a
21:35
charity by using sheer financial weight and the threat of a mass
21:39
talent exodus. The corporate partner held the
21:42
server keys in the capital, which meant the nonprofit boards
21:45
authority was a complete illusion.
21:46
This single lawsuit threatens to unravel the legal foundation of
21:50
the entire tech sectors latest gold rush.
21:53
We're looking at demands that could dismantle the most
21:56
important corporate alliance in the industry.
21:58
The exact demands submitted to the court are unprecedented in
22:01
modern corporate litigation. What exactly is he asking for?
22:05
The plaintiff wants the return of $150 billion to the charity.
22:09
He is demanding the complete removal of the current
22:12
leadership team and most significantly, he is demanding
22:15
the total unwinding of the for profit structure, forcing the
22:19
organization to revert entirely to its original nonprofit
22:22
status. The judge categorized this
22:25
incredibly complex technological dispute simply as a case of
22:28
promises and breaches of promises.
22:30
It allows the ancient established legal doctrine of
22:33
charitable trust to be wielded directly against a modern tech
22:36
giant. If you solicit money by
22:38
promising to build an open charity, you cannot secretly use
22:42
that money to incubate a closed commercial monopoly.
22:45
And you should care about this because state attorneys general
22:49
and international regulators are watching this trial incredibly
22:52
closely. The company's future initial
22:55
public offering and its ability to raise capital from global
22:58
markets are entirely dependent on surviving this legal
23:02
challenge. Oh, absolutely.
23:03
If a judge rules that transferring intellectual
23:05
property from a nonprofit to a capped profit subsidiary is a
23:09
fundamental breach of trust, the entire valuation of the company
23:13
collapses instantly. You cannot take a company public
23:16
if a court might order you to dissolve the corporate structure
23:20
and give all the assets back to a charity.
23:22
This changes the risk profile for every startup trying to
23:25
blend philanthropic missions with venture capital.
23:28
It limits the ability of founders to use nonprofit status
23:31
to attract elite talent and early donations by promising to
23:34
work for humanity only to pivot to commercial profits once the
23:38
technology is viable. Precisely.
23:40
You cannot use the moral high ground as a stepping stone to a
23:42
trillion dollar valuation without facing the legal
23:45
consequences of the promises you made to get there.
23:48
Well, the battle over this artificial intelligence empire
23:51
proves that lending a charitable mission with endless computing
23:54
costs inevitably forces a choice between original ideals and
23:59
commercial survival. The Court's decision will
24:02
ultimately define whether for the benefit of humanity is a
24:05
legally binding contract or just an obsolete marketing slogan for
24:09
the world's most valuable companies.
24:10
If you're not subscribed yet, take a second and hit follow on
24:13
whatever app you're using. It helps us keep making this.
24:16
We appreciate you being here. Also, check out our YouTube
24:18
channel for more business and tech updates.
24:20
There's a link in the description.