Latest / In the Loop AI / Elon vs Sam Altman: The $134 Billion Lawsuit Could End OpenAI

Transcript

Elon vs Sam Altman: The $134 Billion Lawsuit Could End OpenAI · In the Loop AI

0:00
Josh: So everyone knows how messy breakups get when there's money involved.
0:03
Josh: Now imagine that breakup is between two of the most powerful people in tech.
0:06
Josh: There's $134 billion at stake and someone's personal diary just became court evidence.
0:12
Josh: This is Elon versus OpenAI, the saga that just keeps continuing to escalate.
0:17
Josh: And I guess maybe we'll start with the quick version where Elon co-founded OpenAI
0:20
Josh: in 2015 with a $38 million donation, thinking they were building a nonprofit
0:25
Josh: to save humanity from big tech AI monopolies,
0:29
Josh: Google. Flash forward to today, OpenAI is now worth half a trillion dollars,
0:33
Josh: partnered with Microsoft, about to go public, and Elon is saying, hey, I got played.
0:38
Josh: The receipts dropped. We have Greg Brockman, who's OpenAI's president.
0:41
Josh: He had journal entries from 2017.
0:44
Josh: We're going to get into it. It was really fascinating. And Greg posted a rebuttal.
0:47
Josh: The drama is kind of starting to get out of control. And there's a lot of new
0:49
Josh: updates this week that we're going to dive into. So grab your popcorn and let's get into this.
0:54
Ejaaz: Yeah. Okay. So let's start off with the drama between these two.
0:57
Ejaaz: By the way, I have to acknowledge I'm reporting from the Batcave today, Josh.
1:01
Josh: It's a little dark over there. So you're going to have to deal with me being.
1:03
Ejaaz: A bit of a silhouette. Yeah, yeah. Josh is on the West Coast for listeners here,
1:07
Ejaaz: and I'm on the East Coast right now. So it's past sunset.
1:10
Ejaaz: It's past my sunset over here. So let's jump straight into it.
1:14
Ejaaz: Elon has sued many companies and many founders in his time.
1:18
Ejaaz: But his favorite and most judicious lawsuit has been to open air. Who have wronged him?
1:23
Ejaaz: Josh. He initially invested $38 million or rather donated $38 million to what
1:30
Ejaaz: was then a non-profit OpenAI committed to building AI for the open source good
1:37
Ejaaz: so that it wouldn't get into the hands of evil.
1:39
Ejaaz: Fast forward to today and obviously OpenAI's structure has changed into something
1:45
Ejaaz: that could kind of loosely definitely be categorized as a for-profit.
1:51
Ejaaz: And so the lawsuit's been going kind of back and forth for a while,
1:54
Ejaaz: but it kind of culminated over the last two months.
1:57
Ejaaz: To give you the quick headlines, at the end of last year,
2:00
Ejaaz: the judge refused Sam Altman's request to dismiss the case completely,
2:06
Ejaaz: saying that Elon had sufficient evidence and that it was going to go to trial,
2:10
Ejaaz: which is now going to happen in April of this year.
2:13
Ejaaz: But Elon stepped up the gas over the last week and said that he's coming for it all.
2:18
Ejaaz: He's coming for what his original $38 million donation would be in today's OpenAI's
2:24
Ejaaz: valuation. Do you want to guess what that number is, Josh?
2:26
Josh: Some astronomical amount. I mean, this has to be big.
2:30
Ejaaz: It is $137 billion.
2:35
Ejaaz: He's requesting, he's going for the neck, basically. Oh, rather, $134 billion. My bad.
2:40
Ejaaz: So here are the key details. Musk's expert lawyer, he goes by the name of Paul
2:45
Ejaaz: Wazin, values the damages between $79 to $134 billion based on his original $38 million donation.
2:55
Ejaaz: So the question that becomes, well, what's really changed? Why has Moss got
2:59
Ejaaz: like the upper hand now? And it's just something that you alluded to earlier, which is,
3:04
Ejaaz: Greg Brockman, the acting president of OpenAI,
3:07
Josh: We got your diary, bro.
3:08
Ejaaz: Kept a diary of the entire sequence of events or history of OpenAI up until
3:15
Ejaaz: this day, which he had to legally give to the court for review.
3:20
Josh: Oh, that's so vile. And Josh, could you imagine your personal diary?
3:23
Ejaaz: Dude, also, why? It is like a 16-year-old girl's diary.
3:29
Ejaaz: And by that, I mean he has documented everything down to the line of whether
3:34
Ejaaz: he thinks it's morally ethical to do what he was doing back then.
3:38
Ejaaz: So to give you an idea, look at this quote, Josh.
3:42
Ejaaz: Look at this tweet right here. He goes, it'd be wrong to steal the nonprofit from him.
3:48
Ejaaz: Him he's referring to as Elon. To convert to a B Corp without him.
3:52
Ejaaz: That'd be pretty morally bankrupt.
3:55
Ejaaz: And he's not an idiot.
3:57
Ejaaz: And then he goes on to say, I cannot say that we are committed to the nonprofit.
4:02
Ejaaz: Don't want to say that we're committed if three months later we're
4:05
Ejaaz: doing the b call then it was a lie not feeling
4:08
Ejaaz: so great about all of this the true answer is that we want
4:11
Ejaaz: musk out can't see this turning into a for-profit without a very nasty fight
4:16
Ejaaz: so the long story short is elon is suing very aggressively for what his original
4:23
Ejaaz: donation is in an equity stake in the open ai is today to the tune of $134 billion.
4:28
Ejaaz: And he has pretty much a smoking gun, Josh.
4:33
Ejaaz: But there was a rebuttal from Greg Brockman himself saying that Elon Musk had
4:38
Ejaaz: let some important context or left some important context out of his claim.
4:43
Ejaaz: And I'm showing this on the screen right here where, you
4:48
Ejaaz: We've got to figure out how we transition from a non-profit to something which
4:51
Ejaaz: is essentially a philanthropic endeavor.
4:53
Ejaaz: So what you're seeing in blue on the screen here is what Elon has claimed and caught.
4:57
Ejaaz: He's saying, hey, I've always kept my notion that I wanted OpenAI to become
5:01
Ejaaz: a philanthropic endeavor, but he left out what was in red, which is him saying,
5:06
Ejaaz: I know that we need to transform this into a B Corp or a C Corp.
5:09
Ejaaz: So there's a bit of inambiguity and games being played from Elon.
5:14
Ejaaz: Which side do you take on this?
5:15
Josh: I'm trying to look at this and evaluate this as neutral as possible.
5:18
Josh: And one party, we're seeing the journal entries, which one, I have a lot of
5:22
Josh: questions how they got that journal or the diary, and they knew that it even existed.
5:25
Josh: Because that seems like a very personal thing you wouldn't really want to tell
5:27
Josh: people about. So how the lawyers, one, discovered it existed,
5:30
Josh: and then two, got access to it.
5:32
Josh: I'm sure there's some funny stuff going on behind the scenes of this case,
5:35
Josh: just to kind of provide more evidence.
5:37
Josh: But in terms of the evidence that's been provided, so far, you have one party
5:40
Josh: who is like, hey, we don't actually want this person to be at the company.
5:44
Josh: I think we just want to remove him.
5:45
Josh: But I kind of am siding with Elon in this instance for now, because it's very
5:50
Josh: clear that they wanted out.
5:52
Josh: And even though it's clear that Elon observed it was probably necessary to become
5:56
Josh: a B Corp, he still was ousted.
5:58
Josh: And in the case that it became a B Corp, he still does rightfully own those shares of equity.
6:03
Josh: So I guess for now, I'm team Elon. But more than anything, I'm team drama, man.
6:08
Josh: This is great content. And we're going to continue to follow this as we go through
6:11
Josh: this court case, because I'm sure this is just the tip of the iceberg.
6:15
Ejaaz: Dude, I saw a hilarious tweet earlier this week, which said that,
6:19
Ejaaz: My thesis for Anthropik winning the AI race is simply because they have zero drama.
6:24
Josh: No, seriously.
6:26
Ejaaz: All six co-founders are still there. No one's left. They've had the lowest employee
6:31
Ejaaz: attrition of any of the major AI labs.
6:33
Ejaaz: They've just kind of got their crap together. And OpenAIR is the complete opposite.
6:37
Ejaaz: It's an absolute wrecking ball, as we're going to find out on other things later this episode.
6:42
Ejaaz: The final thing I'll make on
6:43
Ejaaz: this point is the OpenAIR and Microsoft partnership. Josh, if Elon ends up
6:50
Ejaaz: able, like, if Elon ends up getting $134 billion, that dissolves the Microsoft
6:58
Ejaaz: OpenAI relationship completely.
7:00
Ejaaz: Because that is the equivalent of Microsoft's stake in OpenAI as well.
7:05
Ejaaz: So Elon's forcing hand basically gets ahead of him. Like Satya doesn't have
7:09
Ejaaz: the power that he originally thought he had.
7:11
Josh: And now poor Microsoft is just caught up into this, even though they had nothing
7:13
Josh: to do with the inception of the company, and they weren't involved until much later.
7:17
Josh: And I think the reason why this matters beyond the drama even is because Because
7:20
Josh: a lot of people believe that if Elon wins, I mean, it could fundamentally challenge
7:24
Josh: how AI companies are valued.
7:25
Josh: And some people are kind of whispering about whether this trial could be a catalyst
7:29
Josh: that could impact the actual AI bubble that we've been building.
7:32
Josh: If you can remove almost like $150 billion out of that and move it into another
7:38
Josh: entity, that's a huge swing.
7:40
Josh: I'm not sure if the market's going to be able to handle that because OpenAI
7:43
Josh: has so many obligations to make money and pay people back. And I guess on that
7:47
Josh: note, maybe we can get into one of the new things that they rolled out in order
7:51
Josh: to generate some revenue this week, which is their new ad model.
7:55
Josh: And Ijaz, if you remember, just last year, I guess two years technically now
7:59
Josh: because we're in 2026, but in 2024, Sam was on stage and he said,
8:04
Josh: ads are kind of the last resort.
8:06
Josh: We don't really like the ad model. We don't believe in it. We don't need it. Flash forward to today.
8:10
Josh: There are ads rolling out in OpenAI. So let's read. We'll start by reading what
8:14
Josh: they announced, which says, in the coming weeks, we plan to start testing ads
8:17
Josh: in ChatGPT free and go tiers.
8:19
Josh: We're sharing our principles early on on how we'll approach ads guided by putting
8:23
Josh: users in trust transparency first as we work to make AI accessible to everyone.
8:27
Josh: So they have like these four principles that they outlined.
8:29
Josh: What matters most? Responses in ChatGPT will not be influenced by ads.
8:33
Josh: Ads are always separated and clearly labeled. your conversations are private
8:37
Josh: from advertisers pro plus business and enterprise tiers will not have ads you
8:43
Josh: just what do you think about this.
8:45
Ejaaz: Okay, I have many thoughts. Okay, so let me give you a kind of lay of the land.
8:51
Ejaaz: OpenAI today has roughly 800 million weekly active users.
8:56
Ejaaz: Josh, guess what percentage of those users pay for their subscription?
9:00
Josh: I know this because we mentioned this in a previous episode.
9:02
Josh: It is close to single-digit percent. It's very low, surprisingly low.
9:06
Ejaaz: It is 5%. Yeah, that's not high. So the lesson that we've learned from this
9:10
Ejaaz: is you could have subscriptions, but it's not enough to keep you afloat.
9:15
Ejaaz: OpenAI is projected to blow $20 billion this year alone.
9:20
Ejaaz: You need something else to pay for it. So you need to somehow monetize the free users.
9:24
Ejaaz: And the classic model that everyone's used for decades now is ads.
9:29
Ejaaz: So they're turning it on for two specific tiers.
9:32
Ejaaz: The free users who are paying nothing to get access to ChatGPT.
9:35
Ejaaz: And this new tier, which launched at the same time that they announced ads,
9:40
Ejaaz: Josh, called ChatGPT Go.
9:43
Ejaaz: Where it's a subscription where you pay $8 a month and you get access to not
9:47
Ejaaz: the best ChatGPT models, but, you know, just below the best.
9:52
Ejaaz: So let's look at some of the napkin math here.
9:54
Ejaaz: There's roughly 600 million non-paying weekly users or monthly users for ChatGPT. Right.
10:04
Ejaaz: Each of them were to pay $2 in ad revenue for the year of 2026.
10:09
Ejaaz: OpenAI stands to make around $1.3 to $1.7 billion, which sounds like it's a lot.
10:15
Ejaaz: But again, their spending budget is $20 billion.
10:18
Ejaaz: So it doesn't really move the needle that much. And they're projecting by 2030
10:23
Ejaaz: to have made $15 per free user, which then pushes them up into the realm of $36 billion.
10:31
Ejaaz: But you can imagine that their spending budget by then, Josh,
10:33
Ejaaz: is going to be multiples of what they're spending this year.
10:37
Ejaaz: So to kind of put into context, it's not looking great. But my argument against
10:41
Ejaaz: that is I think AI is going to be the ultimate form of selling ads in the future.
10:46
Ejaaz: Everyone is going to surface their intents, be it like, I want to buy something,
10:51
Ejaaz: I want to explore something through some kind of AI chatbot.
10:54
Ejaaz: And if OpenAI or ChatGPT rather becomes the face or the doormat of the internet,
10:58
Ejaaz: then they can charge whatever they want per user.
11:02
Ejaaz: And to kind of give listeners a context of like how much money you can actually
11:05
Ejaaz: make, let's look at Meta, right?
11:08
Ejaaz: In this tweet that I have up here, in 2025, Meta made $58 per user just purely from ads.
11:16
Ejaaz: And if you want to look at like the behemoth that is Google,
11:18
Ejaaz: they made $237 billion last year from ad revenue.
11:23
Ejaaz: That makes up 77% of their entire profit. So there's a lot of money to make
11:27
Ejaaz: from ads, but OpenAI needs to figure out a way to go from that $2 projection
11:32
Ejaaz: to something where Meta's hitting like $60 or even Google at around $80 per user.
11:38
Josh: Yeah, I think it's an interesting testament to, I guess, the human nature of
11:43
Josh: how we think about sponsoring and advertising. I was very optimistic in the early days that,
11:48
Josh: advertising that the traditional ad model, the freemium model,
11:51
Josh: was going to go away in the advent of AI.
11:54
Josh: And that was just some hopeful optimism. Didn't really have any reasoning why,
11:57
Josh: but I think it's become clear that that will not be the case.
12:01
Josh: And this push to monetize free users will continue to be this durable thing
12:05
Josh: that continues into this next iteration of technology, mostly based on the fact
12:09
Josh: that people would much prefer to pay with their attention than their dollars.
12:13
Josh: And particularly when it's good information. Now, the question that I have as
12:16
Josh: it relates to this story in particular is how much data are they going to be
12:20
Josh: collecting and using in order to make these ads actually useful?
12:24
Josh: Because one of the core principles that they shared is that your data is private, your data is being held.
12:29
Josh: It's not influencing the ads that you see. But the reality is that ads can be
12:34
Josh: a good thing if they are hyper-personalized.
12:36
Josh: So where are they going to draw that line? Where are they going to take that
12:39
Josh: line in the sand in order to give people higher value?
12:42
Ejaaz: We actually have some information on that exact question that I'm showing on the screen here.
12:46
Ejaaz: They state in their official post ads do
12:50
Ejaaz: not change chat gpt answers so we know that the answer that you're
12:52
Ejaaz: going to get isn't going to be influenced in any way specifically by an
12:56
Ejaaz: advert and then it goes on to say your chats with chat gpt
12:59
Ejaaz: are not shared with advertisers ads will be clearly labeled
13:02
Ejaaz: and chats that include sensitive topics such as health mental health or politics
13:07
Ejaaz: are not eligible for ads so they're taking a kind of hybrid approach here where
13:11
Ejaaz: they're not explicitly sharing all the data or rather prompts that you're sharing
13:15
Ejaaz: within your conversations but it's taking kind of themes from your conversations
13:19
Ejaaz: and sharing them with advertisers saying,
13:21
Ejaaz: hey, these are the general vibes that our users are speaking about ChatGPT with.
13:25
Ejaaz: Maybe your product could be a well-suited fit for this.
13:28
Ejaaz: Another interesting point to make here is, well, what do the ads actually look
13:32
Ejaaz: like? Is it going to be something subtle where it says,
13:34
Josh: This was shocking to me.
13:35
Ejaaz: Right? Well, I actually like this.
13:37
Josh: I like it though. No, well, shocking in not really the most optimal way.
13:41
Josh: When you look at the amount of screen real estate this takes up for people who are listening.
13:44
Josh: It's huge. About a third of the screen is an ad.
13:47
Josh: That's a large percentage of screen real estate for a single advertising slot to go.
13:52
Josh: It's probably going to be worth a pretty hefty premium and a very strong inconvenience
13:55
Josh: to people who don't want to pay for ChatGPT.
13:58
Josh: So this, in a way, does degrade the experience, but should also make them a good bit of money.
14:03
Josh: Because man, if you're getting 25 to 30% on the screen for an impression,
14:07
Josh: that's like a pretty, you're going to see it.
14:09
Josh: Like there's no way your ad blockers, like your eyes are going to glaze over
14:12
Josh: it and not notice because it is such a profound placement on the actual display.
14:17
Ejaaz: Well, it's also not your average impression, right?
14:20
Ejaaz: Because this is a kind of like boosted impression where the conversion rate
14:26
Ejaaz: is or probability of it converting into someone purchasing your product or service
14:30
Ejaaz: is probably way, way higher.
14:32
Ejaaz: Now, to give OpenAI credit, I like that they're excluding things like health,
14:37
Ejaaz: politics, and also people that are under the age of 18.
14:41
Ejaaz: Because AI is very, very persuasive, both in a good way, but in a really, really bad way.
14:47
Ejaaz: The other kind of lens that I apply to this is the competitors are going to eat this up, Josh.
14:53
Ejaaz: Because, listen, Google is getting plenty of cash flow from all their other
14:57
Ejaaz: businesses. So they do not need to turn on ads.
15:00
Ejaaz: Today, if you are a freemium user of OpenAI and you get tired of this real estate
15:05
Ejaaz: that you just referenced, the sponsored ads, you don't want to see that crap.
15:07
Ejaaz: You can just go and use Gemini and there's no ads at all.
15:11
Ejaaz: And let me tell you, Google will be willing to subsidize no ads for as long
15:16
Ejaaz: as it takes to kill OpenAI.
15:18
Ejaaz: Anthropic, on the other hand, isn't a gigantic monopoly, but they're already
15:22
Ejaaz: making so much money from the people that want their Claude Code product.
15:25
Ejaaz: They've already produced something so valuable, Josh, from their subscription
15:28
Ejaaz: that they're making tons of money. They're projected to make $70 billion by 2028.
15:32
Ejaaz: So they don't need to pull off the stunt for ads.
15:34
Ejaaz: So OpenAI really seems like they've got their back against the wall There's
15:37
Ejaaz: no other way for me to describe it, to be honest.
15:39
Josh: Yeah, they're a small fish in a very, very big pond. I mean,
15:42
Josh: they're competing with companies like Google that have an infinite balance sheet
15:45
Josh: relative to theirs, and they have the ability to subsidize and absorb any additional
15:49
Josh: cost required to win additional market share.
15:51
Josh: And that's what we're seeing as we look at these charts of weekly active users,
15:54
Josh: decline for ChatGPT, and increase for all the others across the board.
15:58
Josh: While Anthropix sits cozy and soundly with their incredible coding model and
16:02
Josh: their business-to-business service.
16:03
Josh: Now, I think it's time to get back into the drama news. This is the gossip part
16:08
Josh: of the show because we have some more.
16:10
Ejaaz: Dude, I wish Sam versus Elon was the only catfight going on.
16:14
Ejaaz: But do you remember that angel from OpenAI that couldn't do no wrong and actually
16:19
Ejaaz: called Sam out for being moralistically unethical?
16:24
Ejaaz: Miro Murati, who left and started her own company, her own AI lab called Thinking Labs.
16:30
Josh: Thinking Machines.
16:32
Ejaaz: Thinking Machines, rather. They just lost...
16:35
Ejaaz: Total 50% of their founders or their founding team this week,
16:42
Ejaaz: they lost their CTO, Barrett Zoff, due to, as this tweet says,
16:46
Ejaaz: unethical conduct, according to two sources familiar with the matter.
16:50
Ejaaz: Those two things that were unethical was an in-work relationship,
16:54
Ejaaz: which he didn't disclose.
16:55
Ejaaz: And also, apparently, he was leaking information to competitors.
16:59
Josh: Yikes.
17:01
Ejaaz: Josh, do you want to guess which competitor, if it was true,
17:04
Josh: Who he might have been leaking that to? take a guess. Considering where they
17:07
Josh: all came from, I'm going to guess they were returning back to OpenAI,
17:10
Josh: which seems very clear to me.
17:12
Ejaaz: Absolutely. Look at this tweet from Fiji Simo, the CEO of OpenAI Applications,
17:18
Ejaaz: who at the same time Mira Murati announced that these guys were leaving,
17:23
Ejaaz: she announced welcome Baratsoff, Luke Metz, and Sam Sheldon.
17:27
Ejaaz: So basically, the three exact people that left Thinking Machines this week joined OpenAI.
17:31
Ejaaz: So I have a feeling that they were already getting tired of their work.
17:35
Ejaaz: They didn't really feel focused under Miro Moracci's leadership and decided
17:39
Ejaaz: that the best place to go back was OpenAI and Fiji welcomed them back with open arms.
17:44
Josh: Hey, I got a question for you, actually. What does the key machines do?
17:48
Ejaaz: Dude, I have no idea.
17:51
Josh: Two billion seed raise. I'm not sure anybody has an answer. Two billion dollar
17:53
Josh: seed raise. I'm not sure anybody has an answer.
17:57
Josh: This company, as far as I'm concerned, they have shipped one small product that
18:00
Josh: I don't think, it was just kind of a research product, wasn't very impressive, and nothing else.
18:04
Josh: But they've raised two billion dollars in their seed. They're looking to raise even more money.
18:08
Josh: And there is clearly a lot of drama going on in-house. There's this funny,
18:11
Josh: funny post, EJs. I actually dropped it in our agenda, if you don't mind opening it.
18:15
Josh: It's titled Jim. and there is a post that
18:18
Josh: someone saw looking into the offices and it has like these weights
18:21
Josh: that are custom branded with thinking machines just as a testament
18:24
Josh: to how kind of outrageous the spending is yeah this
18:27
Josh: this photo it's so funny if you scroll down a little bit you could see the plates
18:30
Josh: on the gym are thinking machines plates and ironically there's no 45s in this
18:34
Josh: picture because i mean maybe ai researchers don't don't really trust me like
18:37
Josh: they need to be who knows but i think it's a it's a funny testament to the fact
18:41
Josh: that um there are bubbly things happening.
18:44
Josh: And one of which seems like it's thinking machines. They've raised a ton of
18:47
Josh: money. They haven't really come up with the product.
18:49
Josh: There's clearly a lot of drama where half of the co-founders now have returned
18:52
Josh: back to where they're left from.
18:54
Josh: So clearly they're not liking it. And it seems like there's this lack of direction.
18:57
Josh: And you can give them the benefit and the doubt in the case that they are a
19:01
Josh: research lab and they're hoping to discover some novel algorithmic improvement.
19:04
Josh: But we've discussed this in previous episodes where an algorithmic improvement
19:08
Josh: is very difficult to capture value from because it very quickly becomes commoditized
19:13
Josh: and democratize throughout the stack.
19:14
Josh: So even in the best case scenario, it seems difficult to see thinking machines.
19:20
Josh: Anywhere else besides either being acquired or just kind of fizzling out.
19:24
Josh: And I mean, at this point, it wouldn't even surprise me to see OpenAI acquire them again.
19:28
Josh: The mirror will come back to OpenAI. They'll get the remaining three co-founders
19:30
Josh: and they'll just continue to move on as they were previously.
19:33
Josh: But that is to be determined.
19:36
Josh: In other AI news, perhaps we'll go off the drama and get back to the hard tech,
19:40
Josh: the people who are actually building stuff, not talking about stuff.
19:43
Josh: And that is XAI, who has just released the first, or has just made it to the
19:48
Josh: first gigawatt scale coherent training cluster in the world.
19:51
Josh: And they have not been around. They've been around the least amount of time
19:55
Josh: out of any other company in the world.
19:56
Josh: So it's pretty ironic the fact that the youngest company is now the first to
20:00
Josh: reach this critical scale of one gigawatt worth of energy.
20:04
Ejaaz: Yeah, this is a big deal. So Colossus 2 and Colossus 1 are XAI's specific data
20:12
Ejaaz: centers that they use to gather all their GPUs and train their AI models and
20:17
Ejaaz: inference their AI models.
20:18
Ejaaz: And the unique part around Colossus specifically is Elon is an infrastructure scaling genius.
20:26
Josh: When he started Colossus 1.
20:27
Ejaaz: He scaled it to 400 megawatts of compute capacity in 122 days.
20:34
Ejaaz: Do you know how long it should have taken him, technically?
20:37
Josh: I mean, this is like a year, years-long project.
20:40
Ejaaz: Dude, four years. A year to just set up the infrastructure, about four years to get at life.
20:45
Ejaaz: He did all of that in 122 days. Jensen Huang called him something like short
20:51
Ejaaz: of an absolute genius that no one can compete with.
20:54
Josh: That's one-tenth the time.
20:55
Ejaaz: Four months, right? Yeah. And so he did that kind of midway last year,
20:59
Ejaaz: and he thought, you know what?
21:00
Ejaaz: I'm going to put my foot on the gas even more. And he started with Colossus 2.
21:04
Ejaaz: And by the start of this year, or rather not by the start of this year,
21:10
Ejaaz: but January 17th specifically, he got one gigawatt of compute total live across
21:17
Ejaaz: Colossus 1 and Colossus 2.
21:19
Ejaaz: He is officially the quickest AI startup to get to one gigawatt's worth of compute,
21:24
Ejaaz: beating OpenAI, who started this stuff years ago. This is a two-year startup.
21:29
Josh: Oh my God, we got to show the chart. Please, go to the chart at the bottom.
21:32
Ejaaz: Yeah, please, please, please, please. let me go to the let me go to this where
21:35
Ejaaz: is the wait where is the chart oh it
21:36
Josh: Was at the bottom of the post.
21:37
Ejaaz: Oh yeah there you go sorry right look at this scale it
21:40
Ejaaz: is just absolutely insane to kind of like put this into context for
21:43
Ejaaz: people um we're talking about 555 000 gpus that's half a million gpus worth
21:50
Ejaaz: 18 billion dollars he deployed it that quickly and if you want to understand
21:54
Ejaaz: why he got the edge over every other lab that's been working on this for multiple
21:59
Ejaaz: years and somehow elon's just kind of like swept the rug with them
22:02
Ejaaz: He thought outside of the box. He played in the gray area. And what I mean by
22:06
Ejaaz: that is, of course, he flew in gas turbines, Josh.
22:09
Ejaaz: He flew in gas turbines to power his data centers.
22:12
Ejaaz: If he couldn't get access to electrical grids, he would set it up himself.
22:16
Ejaaz: He would literally thread the wires himself.
22:19
Ejaaz: And if he couldn't get any energy from the national electric grid or the state
22:22
Ejaaz: electric grid, he would fly in Tesla megapacks to power this.
22:26
Ejaaz: He's thinking way outside the box where other AI labs are just kind of waiting
22:30
Ejaaz: for regulatory burdens to be overcome. which is going to take years, to be honest.
22:34
Josh: Yeah, and so much so that these two data centers, Colossus 1 and 2,
22:38
Josh: actually sit in different states, but right on the state line.
22:41
Josh: So they could lobby in two states at once. They have access to double the amount
22:44
Josh: of senators to help get the legislation passed that they need in order to build these things.
22:48
Josh: They have really, like, if you think about how to build a data center from first
22:51
Josh: principles, you want to optimize all of these traits that they have actually been doing.
22:55
Josh: And they have that unique advantage that you mentioned where they have partnerships
22:58
Josh: with companies like Tesla to give them access to these mega packs.
23:02
Josh: And a lot of people don't realize when you do these training runs,
23:04
Josh: there's a couple hundred thousand GPUs that spin up very quickly and then power down.
23:09
Josh: And there's a huge variance in energy that comes through the grid in order to
23:13
Josh: make this work. And when there's a physical combustion motor that's actually
23:17
Josh: spinning, it's very difficult to speed it up and slow it down that fast.
23:19
Josh: So you need something quicker. You need things like Megapack batteries.
23:22
Josh: And the only company that makes batteries this size at this scale with this
23:26
Josh: type of software stack ready to go is Tesla, is the Megapack,
23:29
Josh: and they have this competitive advantage. And in a world in which
23:32
Josh: the winner is the person who can deploy resources the fastest and most efficiently,
23:37
Josh: it very clearly seems that XAI is going to continue to take this lead as they
23:42
Josh: move forward just because of the sheer rate of improvement.
23:44
Josh: And if you look at the chart, I mean, it's a vertical line to get to Colossus 2.
23:48
Josh: And the next person who's coming after this is OpenAI, but that chart looks
23:52
Josh: like it's going to beat them sometime in 2027.
23:55
Josh: So, I mean, granted, there will be variants here, but it seems like XAI finally
24:00
Josh: has taken the lead, And I don't see any world in which they don't continue to dominate this lead.
24:04
Ejaaz: Yeah, I agree with you. Elon's taking a very large gamble here,
24:08
Ejaaz: which is the more compute I have, the better AI model I can build.
24:13
Ejaaz: And that was in contention last year, but Google proved it after they spun up
24:17
Ejaaz: a bunch of TPUs that they produced a leading model.
24:20
Ejaaz: And I have a feeling that Grok 5, whenever that releases in this quarter or
24:24
Ejaaz: first half of the year, is going to be an absolute beast.
24:28
Josh: I can't wait.
24:29
Ejaaz: I can't wait. To put it to context, by the way, because we throw around a lot
24:33
Ejaaz: of these numbers, like one gigawatt, what the hell does that mean?
24:36
Ejaaz: That is the total power that is needed to power San Francisco City,
24:42
Ejaaz: but not just during normal hours.
24:44
Ejaaz: I'm talking about peak hours. If you turn on all the lights,
24:48
Ejaaz: use up all the energy, the most
24:49
Ejaaz: energy, that's how much power is coursing through Colossus 2 right now.
24:54
Ejaaz: And I've got news for you guys. he set up Colossus 3 already and he has raised
24:59
Ejaaz: another $20 billion $5 billion oversubscribed to buy even more GPUs and he is
25:05
Ejaaz: the largest purchaser of NVIDIA's Blackwells and their latest Vero Rubens as
25:10
Ejaaz: well so again it's a big gamble if it pays off
25:13
Ejaaz: Grok's going to be the best, dude.
25:15
Josh: Man, when these Vera Rubin chips come online and they have clusters this size
25:19
Josh: at this scale, there is an impossibility we don't reach AGI at that point.
25:23
Josh: There is so much intelligence, so much energy and compute power solving problems.
25:28
Josh: And this is happening quickly. I mean, this is by next year. We're done.
25:31
Josh: So things are moving quickly. But maybe on that note, particularly the Tesla
25:35
Josh: one, we have updates. It's not that...
25:37
Ejaaz: Well, come on. We're talking about XAI. This is one company,
25:40
Ejaaz: one of six that Elon manages. And you're telling me that he's had a banger week for Tesla as well.
25:45
Josh: Absolutely incredible. So there's a funny thing. I'm in LA right now,
25:48
Josh: and I've been taking Waymos around all week. And this obviously is Tesla's largest
25:51
Josh: competitor. And they're fantastic.
25:53
Josh: They work very well, but they work in this geofenced area where I wanted to go to Santa Monica Pier.
25:57
Josh: And I actually couldn't get there, even though it was only four miles away,
26:00
Josh: because it wasn't in the geofenced area. I wasn't able to get there.
26:04
Josh: FSD is solving this problem. FSD is going to be fully autonomous everywhere.
26:08
Josh: And in fact, just this week, they announced that they are removing the ability
26:12
Josh: to purchase a license February 14th on Valentine's Day.
26:15
Josh: Which is kind of sad because that's a heartbreak thing. And on Valentine's Day
26:20
Josh: of all days, it's not a cool thing, but it signals a few things to me.
26:23
Josh: One is since the beginning of time, it was made very clear that the price of
26:27
Josh: full self-driving would only go in one direction and that's up because the value
26:30
Josh: of having a license to basically grant you unlimited full self-driving miles
26:34
Josh: for the remaining value of the car is a very high premium.
26:38
Josh: And as they get closer and closer, that premium will get higher.
26:41
Josh: Now we've gotten so close to the finish line that they've removed it entirely.
26:45
Josh: And that license no longer exists.
26:47
Josh: So if you want to buy unlimited lifetime full self-driving chauffeur miles,
26:52
Josh: you have about a month to do it.
26:55
Josh: And if you don't do it in that time, that's it, you're done for. Wait, question for you.
26:59
Ejaaz: So let's say you buy a new Tesla after this deadline, Josh.
27:03
Ejaaz: Can you move your lifetime FSD subscription to your new car?
27:08
Josh: There is a month grace period after the window until sometime in mid-March,
27:12
Josh: where if you are a current owner, you can transfer the license to the new vehicle
27:15
Josh: but after that i suspect it's done and it's not coming back but.
27:20
Ejaaz: Didn't people pay like eight thousand dollars
27:22
Josh: To yes i owned the license so it's it's disappointing,
27:27
Josh: But it's been expected for a very long time and the cars last forever.
27:32
Josh: But now the car becomes much more valuable because it does have lifetime free
27:37
Josh: full self-driving miles.
27:39
Josh: And that amount of inference power is really impressive on the back of this
27:42
Josh: post that we're sharing right now, which is the hardware plan as they move forward
27:46
Josh: and as they're actually powering these systems.
27:48
Josh: So AI4 is currently what's on the cars right now.
27:51
Josh: That's what is driving full self-driving. that is what is going to get to full
27:55
Josh: level five autonomy where the cars will drive themselves. But AI5 is coming next.
27:59
Josh: And AI5 is a new chip design that they've been working really hard on basically
28:04
Josh: every week for the last three months.
28:05
Josh: And AI5 is built entirely for full self-driving in a way that AI4 was not.
28:10
Josh: So traditionally with GPUs, the way they're built is for large,
28:14
Josh: I think it's floating point matrix math, some like very elaborate data set.
28:18
Josh: And it's good for general purpose. But basically what Tesla and the AI team
28:21
Josh: are doing is designing, you could think of it like a TPU, but for full self-driving
28:26
Josh: vehicles and for autonomy on machines.
28:28
Josh: So that includes things like Optimus and things like any of the Tesla vehicles.
28:32
Josh: And I'm sure it will find its way into the data centers.
28:35
Josh: In fact, I was listening to an episode with one of the XAI employees who I guess is ex,
28:39
Josh: XAI now because he's no longer with the company after that podcast.
28:43
Josh: But he was mentioning how the plan is actually to take these AI5 chips and all
28:47
Josh: the chips moving forward and actually place them into data centers because they're
28:50
Josh: going to be so effective,
28:51
Josh: so efficient relative to others that Tesla will now soon become not only a large
28:56
Josh: chip manufacturer, but will do the Google thing and will vertically integrate
29:00
Josh: the chips into its own technology stack as it trains these models and starts
29:04
Josh: to build its quest for AGI.
29:06
Ejaaz: He's already working on the future iterations of this chip as well.
29:12
Ejaaz: AI-6, AI-7, and then Dojo 3. He's resuming that project as well.
29:16
Ejaaz: He mentioned that he's going to be doing these iterations in nine-month cycles.
29:21
Ejaaz: So what has typically taken two years, he's accelerating chip design to a new
29:27
Ejaaz: one every nine months, which is just an insane cadence and kind of reminds me
29:31
Ejaaz: of what Jensen is doing at NVIDIA, which brings me to,
29:35
Ejaaz: because Jensen's like doing a new GPU model every year now. That's what he's targeting.
29:39
Ejaaz: Which leads you to my point. We had a conversation, you and I,
29:42
Ejaaz: Josh, months ago last year, where I said that I think Elon might be going down
29:47
Ejaaz: the path where he's not exactly trying to compete with NVIDIA,
29:51
Ejaaz: but he wants to become independent from them eventually.
29:54
Ejaaz: I'm not saying this is a direct hit towards it. I think your comparison to the
29:58
Ejaaz: AI models, the AI-5 chips, rather, being kind of better compared to Google's TPUs is accurate.
30:06
Ejaaz: But if he's not just going to put these in cars, but he's going to put these
30:09
Ejaaz: in humanoid robots and whatever robots Tesla builds in the future,
30:13
Ejaaz: maybe even his spaceships, if they do a cross-collaboration,
30:15
Ejaaz: we already know that there's a lot of synergies between SpaceX and Tesla.
30:20
Ejaaz: This ends up becoming one of the most valuable GPU providers there could be.
30:24
Ejaaz: And I wonder if Tesla starts to get valued similarly to an NVIDIA-esque for
30:29
Ejaaz: maybe custom specialized chips for robots in the future.
30:33
Josh: Yeah, their plan is to develop a TerraFab factory.
30:36
Josh: So their plan is to actually make these chips in-house to become someone like
30:40
Josh: NVIDIA, but do so in a way that's vertically integrated.
30:42
Josh: And you can see in that post, he's kind of sharing what each one of those chips
30:46
Josh: are going to be used for. And we could almost reverse engineer this to create
30:49
Josh: a timeline of when we'll get to AI7, which he says will be the space-based AI compute.
30:54
Josh: If you assume about nine months per iteration, that takes us to 27 months,
30:59
Josh: so two years and change after AI5 releases.
31:01
Josh: So maybe AI5 comes out sometime later this year.
31:04
Josh: After that, we have about two years. So you're looking at like 2028,
31:09
Josh: 2029, end of the decade for space-based AI compute to become really a superpower.
31:16
Josh: And you have to assume that by that time, starships will be working very,
31:19
Josh: very well. They'll get the cost of kilogram down.
31:21
Josh: And that's probably when we'll start to see this new AI in space-based narrative
31:25
Josh: actually literally taking off and getting into outer space. So I found that interesting too.
31:30
Josh: So now we kind of have this loose estimate for timelines as well.
31:32
Josh: And one final bonus on the full self-driving part of this show is the real live
31:39
Josh: implementation of this right now, where we're seeing this post,
31:41
Josh: this person drove 13,000 miles fully autonomously from the West Coast to the
31:46
Josh: East Coast and then partially back.
31:47
Josh: This was all done using a production Tesla. There was no early access information
31:51
Josh: and it was actually verified through the software that's on the car,
31:54
Josh: that 100% of these 13,000 miles were driven autonomously.
31:58
Josh: That includes parking, that includes charging, that includes driving,
32:01
Josh: detours, anything that needed to happen, it was done 100% autonomously.
32:05
Josh: And I think that's why you're seeing this premium for these licenses.
32:08
Josh: And the second they go away, you're on a monthly plan, and that monthly plan
32:11
Josh: is going to change price.
32:13
Josh: Right now it's $100 a month, but I suspect it will continue to go up as the cost per mile goes down.
32:18
Josh: And that's basically it for the autopilot section. I mean, it's exciting.
32:21
Josh: It's happening quickly it's out there in production you could go sit in them
32:25
Josh: today and it really does work you have your own chauffeur.
32:28
Ejaaz: To be honest, the most shocking thing from this news update,
32:31
Ejaaz: Josh, is that you spent the week driving around in Waymo's, dude. You know what?
32:36
Josh: Keep your friends close and your enemies closer, man. I got to know what's going on over there.
32:40
Ejaaz: There's 2,500 of them out there right now, right?
32:43
Josh: Oh, and it looks like it. It's unbelievable. Just a sidebar on LA.
32:46
Josh: There are robots everywhere.
32:47
Josh: There's robots rolling down the street delivering food. There's Waymo's that
32:50
Josh: are driving in the streets. It feels like a futuristic place relative to New York City.
32:54
Ejaaz: Did you make it out to the Tesla diner this time?
32:55
Josh: Yes, and enjoyed every moment of it. In fact, I was there the week that it opened
33:00
Josh: up because I'm such a fanboy.
33:03
Josh: It's a cool experience. I would recommend anyone who is in LA go check it out
33:06
Josh: because it's really fun.
33:07
Josh: It is very good. It's all locally sourced. It's high quality ingredients.
33:11
Josh: It tastes great. It's a really fun and novel experience. So it's worth making
33:15
Josh: a pit stop if anyone's around.
33:17
Ejaaz: Okay, well, if you guys are based in the West Coast or better yet in LA,
33:21
Ejaaz: definitely go check out the diner or get a ride in a Waymo. Let us know. Is it safe?
33:25
Ejaaz: Do you get into a car crash? I personally don't know.
33:28
Josh: And share some pics. It's fun. You get an opportunity to actually sit in the
33:31
Josh: future and live that before a lot of people do.
33:33
Josh: So would highly advise. Now, final topic of the week here is a new partnership
33:37
Josh: between OpenAI and Cerebris.
33:39
Josh: I know, Ejaz, you're a Cerebris fan.
33:42
Josh: You're a Cerebris guy. What's going on over here?
33:44
Ejaaz: I'm more kind of concerned that OpenAI is spending money again,
33:49
Ejaaz: despite them burning so much money.
33:51
Josh: How much money are we talking?
33:53
Ejaaz: $10 billion over the next three years. Just $10 billion. By the way,
33:57
Ejaaz: they don't have that $10 billion. Lord knows what it is for.
34:00
Josh: They don't have any dollars. They haven't made a profit yet.
34:02
Ejaaz: They haven't made any profit, but with ads, of course. But yes,
34:04
Ejaaz: this is OpenAI's latest and greatest partnership, or rather investment,
34:09
Ejaaz: which is to the tune of $10 billion in this company called Cerebrus.
34:14
Ejaaz: Now, when I read this headline, Josh, I had a flashback because I'd seen the name Cerebrus before.
34:20
Ejaaz: And I remember I'd seen their name in the context of them moaning about another
34:26
Ejaaz: investment, which was NVIDIA's $20 billion investment in a company called Grok, which makes custom,
34:34
Ejaaz: let's call them custom GPUs, but they're not exactly GPUs.
34:37
Josh: Grok with a Q. Grok with a Q, yeah. $20 billion acquisition that just happened.
34:41
Ejaaz: $20 billion acquisition.
34:42
Josh: $20 billion acquisition.
34:43
Ejaaz: Licensing acquisition, exactly. And NVIDIA made this acquisition so that they
34:49
Ejaaz: would get access to Grok with the queues
34:52
Ejaaz: LSUs, they're language-specific units or whatever the hell it stands for.
34:57
Ejaaz: And basically, it allows them to process inference at a much cheaper rate.
35:04
Ejaaz: And the reason why that's interesting is, okay, trading the model is all well
35:08
Ejaaz: and done and you need GPUs to do that.
35:10
Ejaaz: But afterwards, you have a ton of people like Josh and I just sending prompts every single hour.
35:15
Ejaaz: And it requires a different type of chip that you can make more performant,
35:20
Ejaaz: save you more money, and also make you more money, right?
35:23
Ejaaz: If you can lower the cost of inference, you end up making money on the back
35:26
Ejaaz: end, right? That's what we all want.
35:28
Ejaaz: And so opening, I thought, hmm, I don't have a Grok.
35:32
Ejaaz: What's the next best company to invest in? And that was Cerebrus.
35:36
Ejaaz: Cerebrus, at the time of the NVIDIA acquisition, complained,
35:40
Ejaaz: saying, Grok's not good enough, our chips are better, basically,
35:44
Ejaaz: because they wanted this kind of salty that they didn't get the investment.
35:47
Ejaaz: Now they're getting it from OpenAI, Josh.
35:50
Ejaaz: And so why would OpenAI do it at this time? Well, they received a big bit of
35:55
Ejaaz: criticism for one of their products, Josh, and that is their coding model called Codex.
36:00
Ejaaz: They were told that it was too slow and Cloud Code not only was smarter, but a lot quicker.
36:06
Ejaaz: Well, Josh, a bit of news dropped in the last week, which actually you and I
36:10
Ejaaz: didn't even catch. We caught it at the last minute. um
36:14
Ejaaz: OpenAI dropped Codex Max High.
36:18
Ejaaz: God knows why they called it that, but it's their latest coding model,
36:21
Ejaaz: which is a lot smarter than it was.
36:23
Ejaaz: It's almost at parity with Cloud Code, but not quite as good.
36:26
Ejaaz: But most importantly, it's quicker.
36:28
Ejaaz: And the rumors state that the
36:29
Ejaaz: reason why it's 15 times quicker is because they're using Cerebrus chips.
36:34
Ejaaz: So they have this specific custom chip. It's huge. I've seen a picture of this
36:38
Ejaaz: thing. It's the size of a dinner plate.
36:39
Ejaaz: It looks absolutely ridiculous, but it's more performant and will allow OpenAI
36:45
Ejaaz: to charge more on the backend whilst also delivering enough compute for anyone
36:49
Ejaaz: and everyone that wants to use it for coding.
36:52
Josh: Yeah, it's a 10 times faster than GPU-based inference performance boost.
36:56
Josh: So this is going to be a huge increase in performance.
36:59
Josh: Now they have the performance. Do they have the actual intelligence?
37:03
Josh: That's the next question that remains to be seen. So is it going to be valuable
37:06
Josh: if it's faster, even though the tokens that it's generating are a bit inferior to Anthropic?
37:11
Josh: I don't know. But again, this is kind of a testament to the strategy of open
37:16
Josh: AI, which is being fighting basically every war on all fronts.
37:19
Josh: They're trying to win consumer.
37:20
Josh: They're trying to win institutional and business.
37:22
Josh: They're trying to compete on coding while also doing image generation,
37:26
Josh: while also doing video generation.
37:28
Josh: So they're trying to be the best across the board. And it seems like as a result,
37:32
Josh: they're becoming kind of...
37:34
Josh: Not the best at anything. And I'm not sure where that strategy leads,
37:39
Josh: but it appears as if they're going to continue doing that based on their ad
37:43
Josh: strategy and now this acquisition of Cerebra.
37:45
Josh: So time will tell how this actually pans out. Do you want to hear.
37:49
Ejaaz: My tinfoil hat
37:50
Josh: Conspiracy, Josh?
37:52
Ejaaz: I also read earlier this week, Josh, what is the Neuralink competitor that Sam soldered?
37:57
Ejaaz: Is it Merge Labs or something like that? It's Merge, right?
38:01
Josh: I think that's probably right. Yeah, I know he has a competitor.
38:05
Ejaaz: Guess which AI lab made a massive investment in Merge Labs this week as well? Who? OpenAI.
38:11
Josh: No. So he's just buying his own bags across the board. It's kind of annoying.
38:15
Ejaaz: It's annoying, right? Because he also secretly has equity in this thing.
38:19
Josh: Is he using investor money as exit liquidity? Yes, that's exactly what he's doing.
38:23
Ejaaz: Yeah, it's sus, right?
38:24
Josh: On the back of a $135 billion lawsuit, I think this is probably going to be
38:29
Josh: a big trend for the year, is just evaluating the...
38:32
Josh: Ethics and strategy of OpenAI as they have gone from non-profit to for-profit
38:37
Josh: and now continue to acquire the same companies that are on the CEO's balance sheet.
38:43
Josh: It's interesting. There's a lot of good lore here that we still have to unpack.
38:48
Josh: But I think that's enough unpacking for today. We covered quite a bit.
38:51
Josh: That's going to be the end of the episode. We have another roundup coming later
38:54
Josh: this week. There's so much stuff to talk about.
38:56
Josh: But yeah, that was another.
38:58
Ejaaz: Well, I was going to say, you might have noticed if you're listening to this,
39:01
Ejaaz: that we didn't mention two of the hottest topics this week.
39:04
Ejaaz: Anthropic releasing Claude Cowork and Google taking over the entire AI world
39:10
Ejaaz: with four product releases, including personal intelligence.
39:13
Ejaaz: We actually made dedicated episodes to both of those. And if you don't know
39:17
Ejaaz: what we're talking about, you missed it. And you should definitely go and watch those.
39:21
Ejaaz: In fact, Josh wrote a banger of a newsletter essay in our newsletter.
39:26
Ejaaz: So you should sign up for that as well.
39:28
Ejaaz: Turn on notifications wherever you listen or watch these things.
39:31
Ejaaz: I know some of you can't bear to look at our faces, so you just listen to our voices.
39:35
Ejaaz: That's totally fine. Turn on notifications, subscribe. It helps us out so much.
39:39
Ejaaz: And yeah, we'll see you on the next episode. We're filming a pretty interesting one.
39:42
Josh: Can we link a little bit of alpha here? Yes, please. For anyone who subscribes
39:45
Josh: to the newsletter, the piece that's coming out this week, EJES,
39:48
Josh: I believe, is going to be on XAI and about why XAI is undervalued,
39:52
Josh: how valuable it can get, kind of a continuation of the conversation we had today.
39:55
Josh: So if that seems interesting, that's going to drop on Wednesday of this week.
39:59
Josh: So subscribe in order to get access to that and see a little bit more detail
40:03
Josh: as to why we suspect XAI is going to have the success that we believe it will
40:08
Josh: be, which will be very, very high.
40:09
Josh: But yeah, that concludes the episode. Thank you all so much for watching. We appreciate it.
40:14
Josh: Appreciate you sharing with your friends, doing all the good things.
40:15
Josh: And we will see you guys in the next one.
40:17
Ejaaz: See ya.