Latest / In the Loop AI / This Week in AI: NVIDIA Wins CES With Rubin and Alpamayo, ChatGPT Health, Anthropic Raise

Transcript

This Week in AI: NVIDIA Wins CES With Rubin and Alpamayo, ChatGPT Health, Anthropic Raise · In the Loop AI

0:00
Josh: Welcome back to the AI Roundup, the first of the year in 2026.
0:04
Josh: And the first year of every year always comes with a lot of excitement,
0:07
Josh: for me at least, because it is the biggest consumer electronic show in the world.
0:11
Josh: In fact, that's what it's called.
0:12
Josh: CES 2026 has come, it is on the way out. And there have been some pretty amazing
0:18
Josh: announcements that have come out of it.
0:19
Josh: But through all of these amazing announcements, it has become increasingly clear
0:22
Josh: that almost every CES now should just be rebranded to the NVIDIA show.
0:26
Josh: NVIDIA is releasing the coolest, most badass
0:29
Josh: technology every year jensen shows up with this like really
0:32
Josh: fancy shiny alligator leather jacket and he just drops the
0:35
Josh: most honestly the most groundbreaking new technology that
0:38
Josh: we've ever seen and this year is no different there are two major announcements
0:42
Josh: he made one around full self driving the other around their new chip architecture
0:45
Josh: and then also this week on the roundup we have a few other updates one of them
0:49
Josh: being open ai's new health application are you going to give them all of your
0:53
Josh: health data do you want them to know everything about you we'll get into that but Ijaz,
0:57
Josh: I want to start with the autonomous car situation because as a Tesla diehard,
1:00
Josh: as someone who has been driven around autonomously for the last X amount of years,
1:04
Josh: what's going on with nvidia and their attempt at doing this because it feels
1:08
Josh: like if nvidia puts their mind to something normally they can do it and this
1:11
Josh: is the first real attempt i've seen at someone else who doesn't actually manufacture
1:15
Josh: the car trying to build some sort of full self-driving technology i
1:18
Ejaaz: Don't think you need to start selling your tesla shares just yet but it's something
1:22
Ejaaz: to be uh aware of um okay what is it just an yeah so they announced this uh
1:27
Ejaaz: new self-driving ai model called Alper Mayo.
1:32
Ejaaz: And in Jensen's words himself, it's the world's first thinking,
1:36
Ejaaz: reasoning, autonomous vehicle AI.
1:39
Ejaaz: And he's launching it on US roads as soon as Q1 of this year with Mercedes CLA
1:46
Ejaaz: cars, which is going to be like an insane turnaround time.
1:49
Ejaaz: He says it's trained end to end, literally from camera in to actuation out.
1:55
Ejaaz: It reasons what action it's about to take, the reason by which it came about
1:59
Ejaaz: that action and the trajectory.
2:00
Ejaaz: If this sounds similar to end-to-end neural net, which is exactly what Tesla
2:06
Ejaaz: FSD is, you would be correct. That's basically what he's built with Alpameo.
2:11
Ejaaz: It's a vision language action model, which takes in a lot of data via different
2:15
Ejaaz: types of sensors, such as cameras, kind of barometers, stuff like that.
2:20
Ejaaz: And it feeds it into the neural net and judges what to do in a particular scenario.
2:25
Ejaaz: But it's not just one model, Josh. There's a few other aspects to it,
2:29
Ejaaz: such as simulation testing,
2:30
Ejaaz: which is also what Tesla FSD does to kind of generate different types of hypothetical
2:35
Ejaaz: ways to deal with a particular scenario on a car so that it can judge and reason
2:40
Ejaaz: to kind of do the right type of action, whether to decelerate,
2:44
Ejaaz: accelerate, change steering, etc.
2:46
Ejaaz: Now, the main core difference between what NVIDIA has built here versus Tesla
2:52
Ejaaz: FSD is that he's not building the actual car.
2:56
Ejaaz: In fact, he doesn't want to go anywhere near it. He's not going to be building
2:58
Ejaaz: any of the sensors either.
2:59
Ejaaz: What he's proposing here is the end-to-end model itself that you can retrofit into a car.
3:05
Ejaaz: And the first brand that he's working with is Mercedes to achieve this.
3:09
Josh: What I like about this is that NVIDIA, in a way, they're kind of acting like
3:13
Josh: Apple, where Apple normally kind of sits on their hands, lets the fields play
3:17
Josh: out, decides what the winning strategy is, and then goes all in on that.
3:20
Josh: We had really Waymo and Google and Tesla kind of competing on these different
3:26
Josh: types of way that you could train it. One was hard-coded, one was neural nets.
3:29
Josh: It became clear that the convergence was on neural nets. And that's what NVIDIA
3:32
Josh: did is they built this architecture just for that.
3:34
Josh: What I find interesting is that this is kind of like Apple CarPlay,
3:39
Josh: but for full self-driving, where Apple doesn't make any cars and yet their software
3:43
Josh: is prevalent in every single car that's basically made in the US today because everyone has an iPhone.
3:48
Josh: Well, NVIDIA is doing the same thing, but for full self-driving where they're
3:51
Josh: not going to actually try to manufacture these cars.
3:53
Josh: Like you said, they're just going to license them to other dealers like Mercedes
3:56
Josh: using this thing called Drive Hyperion, which is this reference architecture.
4:00
Josh: And it's built on the standardized sensor set. And one thing that I found interesting
4:04
Josh: with the sensor set is how
4:07
Josh: Detailed it was, how much there was going on in this, included but not limited
4:11
Josh: to 14 cameras, nine radars, one LiDAR, 12 ultrasonic sensors, and a microphone array.
4:18
Josh: So this is a full stack suite of tools that they're giving to car companies
4:23
Josh: like Mercedes to do whatever they want with it.
4:25
Josh: And notably, it's funny that the ratio of LiDARs to cameras is like 0.75.
4:31
Josh: There's still a lot of LiDAR built into this. So it's interesting to see someone
4:35
Josh: who feels like they actually have a chance at competing with Tesla in the world of autonomy.
4:39
Josh: This is the first real threat that I think exists.
4:42
Josh: Not to say that it's a real threat just yet, because they have a long way to
4:45
Josh: go before they reach the 84 billion miles or whatever it is that Tesla has of training data.
4:50
Josh: But they're on their way, and this is the correct approach to do it.
4:53
Ejaaz: Yeah, it's interesting. We were watching the CES Q&A after this announcement,
4:58
Ejaaz: and someone actually asked him, like, is this a direct competitor to Tesla FSD?
5:03
Ejaaz: And Jensen starts of the response with saying like, hey, listen,
5:06
Ejaaz: Tesla is an amazing company, blah, blah, blah.
5:08
Ejaaz: And then he says Alpameo was designed around a different kind of idea.
5:12
Ejaaz: The first difference being that NVIDIA doesn't want to actually build the self-driving
5:15
Ejaaz: cars, which is the point that we just mentioned, but he wants to build the full
5:19
Ejaaz: stack of technology for everyone else that is trying to build self-driving cars.
5:23
Ejaaz: So reinforcing the point that he just wants to be around the software and maybe
5:26
Ejaaz: some of the sensor stack, but otherwise it's on the car producers themselves
5:32
Ejaaz: to kind of manufacture the actual vehicle itself.
5:35
Ejaaz: And this is an interesting strategy because you could argue that like Tesla
5:39
Ejaaz: has easy scalability because they've built their product end to end,
5:43
Ejaaz: right? And they're churning it out of their factories.
5:46
Ejaaz: Kind of refactored the dye architecture so that they can kind of pump these
5:49
Ejaaz: cars out once an hour or whatever.
5:52
Ejaaz: And NVIDIA is kind of taking a different approach where they're like,
5:54
Ejaaz: okay, we'll just let every single other car manufacturer do this and we'll retrofit
5:58
Ejaaz: with our own kind of model. So I like the approach that they're taking.
6:01
Ejaaz: And I don't know if you saw this as well, Josh, but kind of within all the CES
6:06
Ejaaz: announcements, Uber and Lucid announced that they were also going to launch
6:10
Ejaaz: an autonomous vehicle fleet or network system.
6:13
Ejaaz: So every single car company is kind of veering towards doing this.
6:17
Ejaaz: And I think NVIDIA kind of pitching the software play, like the Apple CarPlay
6:22
Ejaaz: analogy that you just used, it's actually a really, really smart one.
6:25
Ejaaz: Because if this ends up kind of like, if they end up signing like all the right
6:28
Ejaaz: partnerships with the right kinds of car makers, they could end up with the
6:31
Ejaaz: same amount of data and therefore as good a model as Tesla does in fewer years
6:36
Ejaaz: that Tesla's taken to get to the standard that they are at right now.
6:39
Josh: And it's funny, it's not like companies haven't tried to do this.
6:41
Josh: I mean, Rivian's tried full self-driving Ford, Chevy, a whole bunch of companies have tried.
6:45
Josh: They just can't figure it out. So NVIDIA is like, okay, well,
6:48
Josh: clearly there's a huge market demand. We're going to step in to fill this.
6:51
Josh: And to me, it feels like, I mean, they are the NVIDIA, but of the full self-driving industry.
6:55
Josh: Whereas you could kind of compare Tesla to Google, where they own the entire
6:59
Josh: stack. They own the chip architecture. They own the training. They own everything.
7:02
Josh: NVIDIA is just the pick and shovel salesman. They're like, hey,
7:05
Josh: if you want to sell a car, you're going to need to have it fully autonomously. Yeah.
7:08
Josh: You're going to want our software. And this is actually, I mean,
7:11
Josh: people will view this as a threat to Tesla, but a lot of people don't know that
7:14
Josh: Tesla open sourced all of their patents for any other car manufacturer to use from day one.
7:19
Josh: The problem is just, it's incredibly difficult to do these things.
7:23
Josh: So it's not like Tesla has not wanted the industry to keep up because the mission
7:26
Josh: is to move the whole world forward.
7:27
Josh: It's just that no one actually took them up on this and use the patents to actually
7:30
Josh: build something meaningful.
7:32
Josh: So Nvidia came along, they're building the picks and shovels and they're doing
7:35
Josh: it at scale, right? Like a significant percentage of the company has now shifted to
7:39
Ejaaz: Work 20 yeah 7 000 people so
7:42
Ejaaz: 20 of nvidia employees are now going to be working um
7:45
Ejaaz: you know monday to friday and probably on the weekends uh for
7:48
Ejaaz: the foreseeable future and that is just like a huge signal that
7:52
Ejaaz: jensen thinks that this is one of the most important industries to win he said
7:56
Ejaaz: multiple times during ces that he thinks robotics is undeniably going to be
8:00
Ejaaz: the future and the next industry to face a chat gpt like moment um and so he
8:06
Ejaaz: thinks that with this bet it's going to be autonomous cars that kind of like flag the play first,
8:12
Ejaaz: it's interesting I saw yeah I saw a response from Elon here because someone
8:16
Ejaaz: asked him like hey Elon do you see this as kind of like a threat the way that
8:20
Ejaaz: I kind of think about this is that it's not really a threat because,
8:25
Ejaaz: whilst NVIDIA is kind of like creating a software that is 95% as good,
8:30
Ejaaz: the extra 5% takes years and years to actually improve.
8:34
Ejaaz: And Elon replies, you know what, you're right, the actual time from when FSD
8:38
Ejaaz: sort of works to where it is much safer than a human is several years.
8:42
Ejaaz: So maybe this is a competitive pressure on Tesla in five or six years, but probably longer.
8:47
Ejaaz: So Elon's probably, you know, saying this for public opinion,
8:50
Ejaaz: but he doesn't see it generally as a threat.
8:52
Ejaaz: And to your point around like the Google full stack architecture,
8:55
Ejaaz: I think Tesla still very much has the moat right now.
8:58
Josh: Yeah, and they'll win on the margins. I mean, no one's going to be able to create
9:01
Josh: a lower cost per kilometer for traveling if you don't own the whole stack than Tesla will.
9:06
Josh: So it is good that, I think it's a net positive for the world.
9:09
Josh: It is slightly negative for Tesla, but not in a meaningful way.
9:13
Josh: Both of these companies are going to do excellent. And we probably should get
9:17
Josh: into the other excellent thing that was announced from NVIDIA this week,
9:20
Josh: which is probably even bigger than the full stuff. of drawing architecture,
9:23
Josh: that is their new Vera Rubin chip, which Ejaz, I know you were looking at this.
9:28
Josh: I was taking a peek. This chip is incredible.
9:31
Josh: It's like outrageous how meaningfully an impact this will have on the world
9:36
Josh: of AI training. Some fun facts just to get started.
9:40
Josh: Hopper was the first major chip that was being trained that was from NVIDIA.
9:44
Josh: They went from Hopper to Blackwell. That was a 10% or sorry,
9:48
Josh: a 10x energy efficiency improvement.
9:51
Josh: And then from Blackwell to what we have today, Vera Rubin, is another 10x improvement.
9:55
Josh: So we have two orders of magnitude energy efficiency in addition to other crazy
10:01
Josh: stats like Rubin, the GPU is now five times faster than just the previous generation
10:06
Josh: of Blackwell, but it only has 1.6 times the amount of transistors.
10:10
Josh: So they've managed to squeeze a tremendous amount of compute,
10:13
Josh: energy efficiency, and also ease of integration. If you remember,
10:18
Josh: the Hopper chips from Hopper to Blackwell was a really nasty transition period
10:23
Josh: because the whole architecture was different.
10:25
Josh: So you had to rip out the guts of the
10:26
Josh: entire data center, rewire all the electricity, rewire all the cooling.
10:30
Josh: What's nice about this is it's a hot swap. It's easy. You just pull one out,
10:34
Josh: you pop one in, and you're on your way.
10:36
Josh: And what's also interesting about this new chip is this is one of the most amazing
10:39
Josh: things to me is the cooling, how they cool this thing.
10:42
Josh: So with Blackwell, they had a lot of these tubes and wires that would run through
10:45
Josh: that would cool it with liquid.
10:47
Josh: They still have liquid cooling, but there's no more tubes or wires hanging out of this thing.
10:51
Josh: And the water required to cool it can be, I think, 115 degrees Fahrenheit.
10:56
Josh: It's too hot. So they're cooling these GPUs with hot water now because they're so efficient.
11:01
Ejaaz: Doesn't make sense.
11:01
Josh: So now, this is huge for cost per token. If you want to generate more tokens
11:05
Josh: at a lower cost. This is great for training. I mean, Ruben across the board seems incredible.
11:09
Ejaaz: And Josh, it's four times cheaper when it comes to inference costs as well. That's what it is.
11:15
Josh: Four times.
11:15
Ejaaz: That's huge. It's insane. So to put this into context for listeners and viewers
11:20
Ejaaz: here, this kind of a jump between one GPU model to the next is unprecedented. entered.
11:28
Ejaaz: Usually you'll get like a 2x better performance output. To get a 5x,
11:33
Ejaaz: you must have done something different here, right?
11:36
Ejaaz: And the answer is they did.
11:38
Ejaaz: So typically when you refactor a GPU for the next version, which is what they've
11:43
Ejaaz: done from Blackwell to Vera Rubin, you kind of have six components in your AI
11:49
Ejaaz: chip architecture, right?
11:51
Ejaaz: It's not just the GPU that you want to refactor it's the
11:54
Ejaaz: software components it's the cpu that helps orchestrate and
11:57
Ejaaz: help these gpus kind of like talk to each other it's how
12:00
Ejaaz: you stack these gpus together you'll notice in the tweet that we
12:03
Ejaaz: just had up here i'll show it right here um jensen huang
12:07
Ejaaz: talks about a rubin pod which consists of 1152 gpus 16 racks of these gpus stacked
12:14
Ejaaz: together which if you stack this together in this pod gives you the best performance
12:18
Ejaaz: output the performance output and metrics that josh and i just covered right
12:22
Ejaaz: um what nvidia did was they rebuilt.
12:26
Ejaaz: From scratch, all six of these components to build out this pod, Josh.
12:33
Ejaaz: So it wasn't just the GPU architecture.
12:35
Ejaaz: It wasn't just kind of like neatly fitting in a bunch of transistors in a much more impactful way.
12:40
Ejaaz: They refactored the central processing unit, so the CPU, the graphics processing
12:46
Ejaaz: unit, the MVLink 6 switch, which is what helps the software communicate between
12:50
Ejaaz: the CPU and the GPUs, and a bunch of other Ethernet cables.
12:54
Ejaaz: You'll notice the Spectrum 6 Ethernet switch as well.
12:56
Ejaaz: So the point I'm saying here is they did something unprecedented and rebuilt
13:00
Ejaaz: everything, all six components from scratch, which is what led to this massive, massive jump.
13:06
Ejaaz: And it's just like crazy to kind of see some of the takeaways here.
13:09
Ejaaz: So there's a summary here where Rohan over here goes, NVIDIA rolled out six
13:14
Ejaaz: new chips under the Rubin platform.
13:15
Ejaaz: And one highlight is the Vira Rubin Superchip, which pairs one Vira CPU with
13:21
Ejaaz: two Rubin GPUs on a single processor. Now, if none of that is making sense,
13:25
Ejaaz: the summary is here where it talks about the Rubin pod and I mentioned that.
13:29
Ejaaz: Now, they haven't released pricing for this thing.
13:32
Ejaaz: But let's conservatively take the cost of a Blackwell GPU, which was between $35,000 to $40,000.
13:39
Ejaaz: I'm going to take the high-end estimate of that. If we factor in the number
13:44
Ejaaz: of GPUs in this pod, which is 1,152, that comes out at $46 million for the cost of this entire pod.
13:52
Ejaaz: So if you're an AI lab or a startup, God forbid, that's trying to enter this
13:56
Ejaaz: market with the best of the best GPUs, you're going to need to fork out 46 to
14:00
Ejaaz: get some of the best metrics ever.
14:01
Ejaaz: So this will continue to be a game played by very big hyperscalers and big AI
14:07
Ejaaz: labs like Google, OpenAI and XAI, which are securing majority of the capacity so far.
14:12
Josh: But you know why that price doesn't matter at all is because the constraint isn't dollars.
14:18
Josh: The constraint is energy. So they could probably charge double for that rack.
14:23
Josh: And there will still be no shortage of demand because, again,
14:27
Josh: that energy constraint means everything. And when you look at the performance
14:30
Josh: per watt of a Vera Rubin rack versus a black hole rack, it is that multiple,
14:36
Josh: would you say four times multiple?
14:37
Josh: That is worth so much money because you are limited in energy.
14:41
Josh: That you cannot make more of.
14:43
Josh: But now you can get four times the amount of compute for the same energy.
14:46
Josh: That is worth more than any amount of dollars you can possibly spend.
14:49
Josh: Like they will spend infinite money to get that efficiency.
14:52
Josh: And that's why this chip is going to do so good. Another really interesting
14:55
Josh: fact for those who don't really spend their life in this world,
14:59
Josh: think of the internet, how much global bandwidth the entire internet uses every single day.
15:04
Josh: One of these racks moves 240 terabits per second, which is equivalent to twice
15:09
Josh: the entire global internet bandwidth in one rack, double.
15:14
Josh: So this is a tremendous amount of compute that we're talking about,
15:18
Josh: and it is all powered by NVIDIA.
15:20
Josh: Now, you just, if you scroll down to the first reply here on this post,
15:23
Josh: it's from none other than our good pal Elon Musk, who says this will probably
15:27
Josh: take about nine months or so before the hardware is operational at scale.
15:30
Josh: And this is an important thing to note.
15:32
Josh: Because to move from Hopper to Blackwell, it took like 16, 18 months.
15:37
Josh: And we had this weird lull period where there wasn't a lot of progress on the hardware front.
15:42
Josh: The progress came from software. So this is when we started to get these chain
15:45
Josh: of thought and these reasoning breakthroughs that kind of held us through that
15:48
Josh: lull period between Hopper to Blackwell.
15:50
Josh: What we're getting now is we're just starting to feel the effects of Blackwell.
15:55
Josh: And then we have this nine-month period until Vera Rubin. So there should be
16:00
Josh: this weird lull period where we're starting to see what happens with Blackwell,
16:03
Josh: but we're not actually gonna feel the effects of this new chip until probably the end of this year.
16:07
Josh: So I would say if we're gonna get AGI, it's gonna come from these chips.
16:11
Josh: It's going to happen in Q4 once these
16:14
Josh: things are actually online and fully operational and training things.
16:17
Josh: And there's also just one random fun fact that I thought, Ijez,
16:20
Josh: you probably didn't know that you should know that I found out.
16:22
Josh: Vera Rubin is actually the name of a person. It's an American astronomer whose
16:26
Josh: work helped convince the world that dark matter was real.
16:28
Josh: So I just thought that was like a fun fact. It's like, okay, cool.
16:31
Josh: They got some cool meaning behind it. But yeah, this ship is remarkable.
16:35
Ejaaz: Insane. I'm super pumped about it. It is an impressive feat of engineering.
16:39
Ejaaz: I just wish it was here to play with already.
16:44
Ejaaz: You know, like Blackwell announced, what is it, like five to six months ago?
16:48
Ejaaz: And it's only now coming online. Probably the first data center is going to be XAI's Colossus 2.
16:53
Ejaaz: So we're going to see the real effects of that generation of GPUs.
16:57
Ejaaz: Kind of like relatively soon. So Vera Rubin is probably not going to be seen
17:01
Ejaaz: until early next year if we factor in pre-training and post-training for AI
17:06
Ejaaz: models once they've spun up these podcasters.
17:08
Ejaaz: So I'm excited to see that, but I hate that we have to wait so long.
17:13
Ejaaz: Another point to make, I guess, is NVIDIA is pumping out these new GPU versions
17:17
Ejaaz: way more frequently than they used to be.
17:20
Ejaaz: And that just is kind of a sign of the competitive times as the likes of AMD,
17:25
Ejaaz: Intel, Google's TPU kind of breeds down their neck.
17:28
Ejaaz: They kind of want to make sure that they're top dog.
17:31
Ejaaz: And they've shown this with the 5x performance increase. I expect to see probably
17:35
Ejaaz: a new NVIDIA GPU probably come out at a cadence of once every year eventually,
17:39
Ejaaz: which gets me thinking, if there are constraints around memory,
17:43
Ejaaz: we actually did an amazing episode on this earlier this week,
17:47
Ejaaz: definitely go check that out.
17:48
Ejaaz: And other types of things like energy, as you mentioned, Josh,
17:51
Ejaaz: I wonder how many of these GPUs are just going to be collecting dust.
17:54
Ejaaz: Do you remember we spoke about, there was that clip of Satya Nadella saying,
17:59
Ejaaz: hey, I have like $300 million worth of H100s just collecting dust because I
18:03
Ejaaz: don't have the energy to kind of like fit this out.
18:05
Ejaaz: I wonder if we'll start running into those kinds of problems going forwards
18:07
Ejaaz: but such a such a cool announcement.
18:09
Josh: Yeah it's exciting and then in the meantime we're just going to get these
18:12
Josh: software breakthroughs so the um like you
18:15
Josh: were saying the breakthroughs from blackwell have not totally
18:18
Josh: come we have this like nine month lull period in between where we're going to
18:21
Josh: make some software innovations and it's it's just this like double parallel
18:25
Josh: exponential progress they're both building on each other they're both compounding
18:29
Josh: the line is going vertical progress as fast as you think it's moving it is moving
18:32
Josh: so much faster but i want to round out the CES segment was something a little more fun,
18:37
Josh: Ijaz, because it wasn't just the NVIDIA show, there was other stuff.
18:40
Josh: I'm curious if you saw anything personally that got you excited,
18:43
Josh: because for people who don't know, this is the consumer electronics show.
18:46
Josh: This is where they announce all the cool new technology that,
18:49
Josh: for the most part, you can actually buy sometime this year to put into your
18:53
Josh: life. So is there anything that caught your eye in particular?
18:56
Ejaaz: Yeah, this is peak AI. For any of you who are just listening,
19:00
Ejaaz: I suggest you turn the screens on.
19:02
Ejaaz: The world's first robot vacuum, Josh, that has legs. With legs.
19:07
Ejaaz: So, hey, hey, hey, listen. The number one issue I had with Roomba,
19:12
Ejaaz: which, by the way, went bankrupt last week,
19:16
Ejaaz: was that, okay, it can move around my floor and vacuum and kind of get into
19:20
Ejaaz: crevices, but it can't climb stairs.
19:23
Ejaaz: So how can it, can I, I just want to leave it alone and let it clean my entire house.
19:27
Ejaaz: This Roborock Saros rover has the ability to now climb stairs.
19:31
Ejaaz: You have a video of this where it basically helps to put it up there.
19:35
Ejaaz: It has a rectangular surface versus a circular surface.
19:38
Ejaaz: It has AI-powered software and sensors, which allows it to map out your entire
19:43
Ejaaz: house in 3D so that once it's mapped the entire house out, it can just kind
19:47
Ejaaz: of run the same routine over and over again.
19:49
Ejaaz: Here is a separate angle for those of you who are really keen to see like,
19:53
Ejaaz: hey, I don't know if it's cleaning the entire step.
19:54
Ejaaz: It is look at this look at this like flexibility and mobility on this i i just
19:59
Ejaaz: freaking love love it um so it's going to be coming out uh relatively soon uh
20:03
Ejaaz: they haven't kind of like gone into production yet but i like the fact that
20:06
Ejaaz: it's a prototype that actually works josh versus something that is a kind of
20:09
Ejaaz: like graphics video demo very cool that's pretty sick.
20:12
Josh: Yeah i love i love the robots look it could even jump it was hopping a little bit earlier
20:16
Ejaaz: Exactly all right what about what about you what you got what were you for me excited about.
20:21
Josh: Yes this year this was the year of displays, man. This display technology in
20:25
Josh: general was so exciting for me.
20:27
Josh: First, I want to reference the Samsung folding display because what you're seeing
20:31
Josh: here isn't just a Samsung folding display.
20:32
Josh: This is the display in the new iPhone that's coming later this year.
20:35
Josh: And it's cool that we get to see the core technology displayed before it gets
20:38
Josh: placed into products that we'll all use later in the year.
20:41
Josh: And the cool thing about this new Samsung folding display is for the first time
20:44
Josh: ever, your display can fold in half and when it unfolds, will not leave a seam.
20:49
Josh: So if you see on the left here, there's a display that has a seam right down
20:52
Josh: the middle. It doesn't look that good. You could tell that it's folded,
20:55
Josh: The display on the right looks no different than an iPad. And this is what we're
20:59
Josh: going to be getting in the iPhone Fold later this year.
21:01
Josh: So it's cool to see Samsung, who is a supplier of Apple, showcasing their new
21:05
Josh: technology that will eventually be in their hands and pockets later this year.
21:09
Ejaaz: Android users are punching the air right now, Josh. Because they've had this tech for so long.
21:15
Josh: Yeah, but yours had a seam. Ours doesn't have a seam,
21:18
Ejaaz: Bro. Hey, don't. Okay, yeah. Sorry, I thought you were personally attacking
21:21
Ejaaz: me. I'm an Apple user. Come on.
21:23
Josh: No, I'm attacking the Android people. but like, yeah, okay, you guys have been
21:25
Josh: folding for years, but like, yeah, have fun with your seam.
21:28
Josh: Anyways, the second cool technology is micro LED displays. And what's cool,
21:33
Josh: what we're seeing on screen is they're transparent.
21:34
Josh: You can actually create these beautiful transparent displays that use micro LEDs.
21:38
Josh: And the difference between a standard LED and a micro LED is a micro LED actually
21:42
Josh: has all three color diodes.
21:44
Josh: It has RGB built into each one of the tiny little pixels.
21:47
Josh: So it's really high fidelity, really cool, interesting displays.
21:50
Josh: And then the third and final thing in the display world is for people who use Macs.
21:54
Josh: There is no 5k 120 hertz
21:57
Josh: display for mac and that has
22:00
Josh: driven me absolutely crazy because 120 hertz is
22:03
Josh: what is it feels very smooth it's what the new iphones use and 60 hertz just
22:08
Josh: doesn't cut it but they haven't been able to do this because they've been throttled
22:11
Josh: by thunderbolt and long story short if you are a macbook user and you want the
22:14
Josh: native display resolution this is a big year for you you will be buying a new
22:17
Josh: monitor it is very exciting but that wraps up our CES news this week.
22:22
Josh: Ejaz, we had some other big news, not CES related, but instead from Big Dog
22:26
Josh: OpenAI. This is their first big announcement of the year.
22:29
Ejaaz: I am pumped about this.
22:29
Josh: With, yeah, ChatGPT Health. You were very excited about this.
22:32
Josh: So please explain to myself and the audience what caught your eye with ChatGPT Health.
22:38
Ejaaz: So one of the biggest ways that I use ChatGPT is through all the research and
22:43
Ejaaz: reading that we do, including prep for the episodes that we talk about.
22:47
Ejaaz: But the other, probably second most used one is around personal health and fitness.
22:53
Ejaaz: I would use it to kind of design gym routines for me, diet plans for myself.
22:58
Ejaaz: And in some ways, Josh, I will give it my medical labs,
23:02
Ejaaz: my reports, my records to kind of like gauge whether I'm going to be facing
23:07
Ejaaz: some sort of issue in the future,
23:08
Ejaaz: or whether the diagnosis or rather symptom that I'm feeling or getting today
23:14
Ejaaz: is actually accurate. And I've run into a few obstacles.
23:17
Ejaaz: Typically, ChatGPT would give me a weird sort of answer that isn't really accurate,
23:21
Ejaaz: or it'll just politely decline and say, sorry, I'm not a doctor,
23:25
Ejaaz: I can't do this. Until this week.
23:28
Ejaaz: OpenAI released something called ChatGPT Health, which is a dedicated space
23:32
Ejaaz: for health conversations in ChatGPT. Okay, so what does that mean?
23:37
Ejaaz: Basically, it's a separate space that you can connect your medical records and health fitness apps.
23:42
Ejaaz: So now Apple Health, the likes of MyFitnessPal, and also Peloton and an array of other apps.
23:49
Ejaaz: I'm hoping Aidsleep, Aidsleep, if you're listening to this, please connect to this.
23:53
Ejaaz: And you can feed this data into
23:55
Ejaaz: your ChatGPT, and it can start to identify certain patterns of health,
24:00
Ejaaz: whether you're feeling something, you can recognize whether you've consumed
24:03
Ejaaz: too much alcohol the night before, and it can give you personalized information
24:07
Ejaaz: about how you can start to improve your life.
24:10
Ejaaz: Now, it's important to stress that it's not trying to replace the role of a
24:14
Ejaaz: doctor, and they've been very explicit in actually saying that,
24:17
Ejaaz: but it's meant to be an aid.
24:18
Ejaaz: It's meant to be an assistant, and it's that jump into a personal AI assistant
24:22
Ejaaz: that extends beyond just becoming your essay writer or kind of writing up a
24:27
Ejaaz: product plan for you for your kind of nine-to-five knowledge worker job.
24:30
Ejaaz: This is something that actually applies to
24:32
Ejaaz: pretty much anyone that wants to live, which is every human on earth.
24:37
Ejaaz: And so I'm excited to see something like this scale. This, this to me is something
24:40
Ejaaz: useful that I would like pay an extra 10 to 20 tech, even 50 bucks to use.
24:46
Josh: I'm finding it funny the roles that companies are finding themselves in where
24:51
Josh: we have in one corner, Anthropic, they are the world's best at coding.
24:54
Josh: They're sounding the alarm that we are building AGI very quickly. We need to be careful.
24:58
Josh: We need to make sure this is aligned then we have grok which
25:01
Josh: is dead set on building the truth-seeking ai they're building the biggest data
25:05
Josh: center in the world as fast as possible and then chat gbt is like we partner
25:09
Josh: with disney to give you cute little things that you could pair with your videos
25:12
Josh: and now we have this like fun little health feature where we can improve your
25:16
Josh: health by ingesting your records and it's it's nice it's important it's just
25:20
Josh: the stark contrast between
25:22
Josh: the other companies and open ai kind of made it clear this year that they're
25:26
Josh: going to be leaning more into the commercial side of the business versus the
25:30
Josh: consumer side of the business.
25:32
Josh: So this is interesting to me. I think the personal take was that we're reaching
25:36
Josh: a crossroads now where you kind of have to make the personal decision of how
25:39
Josh: deep do you want to go with these LLMs, with these companies?
25:43
Josh: Are you ready to go all in? Because if you do, then you open yourself up to
25:47
Josh: all the benefits that will come from it. You give them your health data.
25:50
Josh: Well, you'll get all the health advisory stuff. And then as the hardware comes,
25:53
Josh: it'll track more of your health.
25:55
Josh: It'll It'll gain this full stack profile on you and know you better than anything else in the world.
25:59
Josh: And if you're okay with the people at OpenAI having that data to use at their
26:03
Josh: will, then this is a great thing.
26:04
Josh: But I think it's a personal decision whether or not you want to decide to go
26:08
Josh: all in on a company like this and give them all your data versus kind of reserving
26:13
Josh: it and maybe giving it to someone else who you trust more or just keeping it for yourself.
26:16
Josh: It's this interesting world dilemma, but this is like really awesome and really impressive.
26:21
Josh: And I frequently think about how
26:23
Josh: Steve Jobs would think about AI and how someone like Kim would sell AI.
26:29
Josh: And it's not the way that OpenAI is doing it, but I would imagine it would follow
26:34
Josh: similar principles where, listen, now you have a doctor in your pocket that
26:38
Josh: knows everything about you. It can help save your life.
26:40
Josh: It can save your loved one's life. It's like this very powerful technology.
26:43
Josh: And OpenAI is leaning into it with this whole thing. So I don't know.
26:46
Josh: I'm feeling excited about it. I'm not there yet.
26:48
Josh: I signed up for the wait list. I'm not sure I'm going to be user number one.
26:52
Josh: But if they could prove that it really is valuable enough, it seems like it makes sense to do it.
26:56
Ejaaz: Well, what I appreciate with OpenAI from the start is that they've hyper-focused
27:00
Ejaaz: on the retail consumer, like the average Joe and how they can benefit from AI.
27:05
Ejaaz: Yeah, they're focused on enterprise.
27:07
Ejaaz: I'd say Anthropoc actually is focused more on enterprise, but OpenAI has always
27:10
Ejaaz: tried to figure out, okay, what's the best consumer product loop that I can provide them?
27:14
Ejaaz: And I think health is a really good one to focus on.
27:18
Ejaaz: We've just obviously ended 2025. And one of the most painful tasks that I need
27:23
Ejaaz: to do at the end of every year is figure out my health insurance for the next
27:27
Ejaaz: year, living in America, right?
27:29
Ejaaz: And being in New York on the East Coast, it is the most arduous.
27:35
Ejaaz: I absolutely hate it. I have to like sign up to like 10 different accounts.
27:39
Ejaaz: I need to evaluate a million different plans.
27:41
Ejaaz: And then even when I'm on the plan, it is super hard to kind of verify whether
27:45
Ejaaz: the information about a particular doctor and their address is even correct.
27:49
Ejaaz: So it's very, very antiquated.
27:51
Ejaaz: And I've been waiting for this industry to be disrupted.
27:54
Ejaaz: And then came along an app like Oscar Health.
27:57
Ejaaz: For those of you who haven't heard about this, this kind of like changed healthcare
28:01
Ejaaz: in many different ways in America in particular, because you could just pick
28:04
Ejaaz: up your phone and FaceTime your doctor and get that live healthcare, right?
28:09
Ejaaz: ChatGPT Health, to me, is that next step evolution where eventually,
28:13
Ejaaz: you mentioned, Josh, it's funny, that you'll have a doctor in your pocket
28:16
Ejaaz: you'll have the best doctor in your pocket. In fact, you won't just have the
28:19
Ejaaz: best specialist in your pocket, you'll have the equivalent of 10 specialists in one doctor.
28:24
Ejaaz: How much would you be willing to pay to see that doctor every single day, 24-7?
28:29
Ejaaz: These doctors don't sleep, right? And then I think about the next step evolution, which is,
28:33
Ejaaz: if you have this doctor in your pocket, it can start feeding into pharmacies,
28:37
Ejaaz: which can start delivering personalized pill packs, or maybe even personalized
28:41
Ejaaz: treatments, peptides, stuff like that going on to the future.
28:43
Ejaaz: So this is the first step of many.
28:45
Ejaaz: And Sam Altman has said for a while now, I'd say like for the last five months,
28:49
Ejaaz: that he intends for health to be a big component for OpenAI.
28:52
Ejaaz: They have RetroSciences as well, which are kind of like their lab to kind of
28:55
Ejaaz: build out future cures for diseases that ChatGPT helps discover.
28:59
Ejaaz: So I think this is part of a much larger puzzle that will help kind of like
29:04
Ejaaz: build this health AGI, which they're aiming for by the end of this year.
29:07
Josh: Yeah. And the key piece to the health puzzle is data. And the problem is that
29:12
Josh: a lot of, for a lot of people, it's just difficult to collect the data.
29:14
Josh: You're not really going for MRIs and CT scans and scanning a lot of parts. It's very topical.
29:19
Josh: It collects data from something like an Apple Watch or maybe some blood results you got from a doctor.
29:23
Josh: One interesting thing I've seen people doing that I would advise you to possibly
29:27
Josh: try if you've ever done a DNA test or you've ever done Ancestry or 23andMe or
29:32
Josh: whatever companies give you your DNA profile.
29:36
Josh: It's normally a couple hundred gigabytes of raw data, but you can actually upload
29:40
Josh: this into an LLM and get customized feedback on what types of things could possibly
29:46
Josh: affect you based on your genomic makeup.
29:48
Josh: So you could see if you are resistant to insulin or if you are prone to these types of diseases.
29:53
Josh: And these LLMs do a remarkably good job of analyzing this tremendous amount
29:58
Josh: of data and kind of giving you the places that you need to focus on and be most
30:02
Josh: cautious of as you move forward.
30:04
Josh: So this type of health practice, as we collect more data and figure out how to parse through it.
30:08
Josh: It's really exciting. And it's made me want to go get a DNA test and try to load it up into an LLMC.
30:14
Josh: And I think it's super exciting stuff. Health is going to be a very large category
30:17
Josh: this year and OpenAI is making sure of it.
30:19
Ejaaz: Going to be huge. Okay, in the final story on the docket, we have some capital
30:25
Ejaaz: raised news from two of our favorite AI labs that are out there.
30:29
Ejaaz: We've got Anthropic of Claude Code.
30:30
Ejaaz: They've confirmed that they're raising $10 billion at a $350 billion valuation.
30:36
Ejaaz: Now, this nearly doubles its previous valuation of $183 billion.
30:42
Ejaaz: Josh, you gave me the timeline of if you were an investor in August of last year what's the return.
30:49
Josh: Yeah i was looking at like uh i think it was foundry is the name of the company
30:52
Josh: there's like uh basically spvs you could buy into them employees will sell some
30:57
Josh: of their shares in the public market or private market you could buy into them
30:59
Josh: it showed the chart over time and i think august 1st of 2025 it was trading
31:03
Josh: at 69 a share and now it's something like 260
31:07
Josh: so 4x in a couple of months i mean if you own any of these things if you are
31:11
Josh: an employee at any of these things, congratulations.
31:13
Josh: You are getting generationally wealthy and the train is not stopping anytime soon.
31:18
Josh: This is not the only big raise this week. We had a second one,
31:22
Josh: XAI in the news, a $20 billion Series E.
31:28
Josh: And they were targeting $15 billion. It was oversubscribed.
31:31
Ejaaz: $5 billion over target. Oversupped.
31:32
Josh: One of the interesting stats here is that top line there, these 600 million
31:36
Josh: monthly active users, that is encroaching on OpenAI.
31:39
Josh: And they're getting closer and closer to that mythical 800 million
31:43
Josh: weekly active user number granted monthly to weekly so
31:45
Josh: it's a little bit different but uh they're on a roll and i
31:49
Josh: think this capital raise will be enough to this is a crazy take this could be
31:54
Josh: enough to get them to the able could be could be 20 dollars applied to blackwell
32:00
Josh: and vera rubin over the course of the next 24 months leads to some pretty unbelievable progress
32:06
Josh: And $20 billion may be enough to get there.
32:09
Ejaaz: You might have noticed for regular listeners on the show, we've been kind of quiet about Grok.
32:15
Ejaaz: And the simple reason is since Grok 4, we haven't really got their unofficial
32:19
Ejaaz: release of Grok 4.20 or any rumors around Grok 5.
32:24
Ejaaz: And the reason behind this is because I think Elon has gone kind of full on
32:28
Ejaaz: founder mode and he wants to assemble the largest armory of top-end GPUs in
32:36
Ejaaz: Colossus 1 and Colossus 2 data centers.
32:38
Ejaaz: Once he's acquired these, once he's scaled these data centers quicker than any
32:42
Ejaaz: other competing AI lab, he'll be able to train the mother of all AIs.
32:47
Ejaaz: And rumor has it that he's going to amass 900,000 GPUs in Colossus 2 data center
32:54
Ejaaz: alone. And these aren't just any GPUs.
32:57
Ejaaz: He's the largest acquirer of Blackwell GPUs, fun fact, the ones that are going
33:02
Ejaaz: to go live very, very soon in this quarter.
33:05
Ejaaz: And he's also one of the largest purchasers of Vera Rubin GPUs.
33:09
Ejaaz: Jensen absolutely loves this guy and says he's the guy that can scale GPUs quicker than anyone.
33:14
Ejaaz: So he's aiming for 900,000 GPUs by the end, sorry, by mid Q2.
33:20
Ejaaz: That is like astoundingly quick and will beat OpenAI to the punch,
33:24
Ejaaz: even though they started well before.
33:25
Ejaaz: So if he pulls this off, Josh, i think that prediction is not too far-fetched
33:29
Ejaaz: he will end up building the best ai model this year that's that's my bet.
33:32
Josh: At least yeah no and i say this with very high levels of conviction do
33:35
Josh: not mistake their silence for weakness xai is building faster than anybody else
33:41
Josh: and i think they were my winner for the the like winner of 2026 in terms of
33:45
Josh: who's going to create the best ai models they are building at a rate faster
33:48
Josh: than anybody else in the world and in these races where the resources are the constraint.
33:54
Josh: The person to deploy them the fastest and the most efficiently is the one that wins the race.
34:00
Josh: And XAI right now is winning that race, if you're looking at it from a more
34:04
Josh: long-term perspective.
34:05
Josh: But with that, we conclude our weekly roundup. All the topics that we wanted to get into about
34:11
Josh: I think the prompt for this week maybe is CES announcements.
34:16
Josh: What's the coolest CES announcement? There's so many fun gadgets and gizmos
34:19
Josh: that I saw come out this week.
34:20
Josh: And a lot of them are available to purchase right now. Like even Anchor,
34:23
Josh: the charger company that I use a lot, they put like these cute little LEDs in
34:26
Josh: the charging bricks now. It's like a lot of cool stuff.
34:29
Josh: So I think the prompt for today could maybe be, what's the coolest thing you
34:33
Josh: saw from CES? What are you buying from CES?
34:34
Josh: Or what are you most excited about from CES? Did you got anything for them?
34:38
Ejaaz: I have a slightly different prompt. I know a lot of you listeners are Teslables
34:43
Ejaaz: and even in some cases Tesla haters.
34:45
Ejaaz: I want to hear from both of you guys whether you think NVIDIA's competing Alpameo
34:51
Ejaaz: vehicle AI is actually a legitimate threat or not.
34:55
Ejaaz: Kind of let us know in the comments. I want to hear all about this.
34:57
Ejaaz: And yeah, that is all we have for the docket today.
35:00
Ejaaz: We released two episodes earlier this week, one on the semiconductor memory
35:04
Ejaaz: squeeze, as well as OpenAI's new secret device.
35:07
Ejaaz: Definitely go check those videos out. we also dropped a sick kind of thesis
35:12
Ejaaz: and essay on the memory squeeze as well for those of you who don't want to be
35:16
Ejaaz: consuming video content all day
35:18
Ejaaz: we've got a nice little article that we've written out for you um and all.
35:21
Josh: This link in the description don't forget to click that down below
35:24
Ejaaz: All linked in the description and also as you're watching this video we dropped
35:28
Ejaaz: the five biggest highlights of this week um so definitely go check that out
35:32
Ejaaz: um if you're wondering where the hell that is that's on our newsletter subscribe
35:35
Ejaaz: um like and subscribe on our video channels,
35:38
Ejaaz: wherever you do Spotify, Apple Music, RSS feeds, wherever you are.
35:42
Ejaaz: And we will see you, hopefully, on the next one. Number two, 2026. Buckle up.
35:47
Josh: We are in it now. This is going to be a huge year and we will be here to cover
35:52
Josh: the entirety of it as we go through. So yeah, thanks for watching.
35:55
Josh: We'll see you guys in the next one.
35:56
Ejaaz: Cheers, guys.