Latest / In the Loop AI / NVIDIA GTC: Jensen Huang's 5 Biggest Announcements

Transcript

NVIDIA GTC: Jensen Huang's 5 Biggest Announcements · In the Loop AI

0:00
Josh: NVIDIA just held its GTC conference in San Jose, where Jensen Huang walked on
0:04
Josh: stage in front of 30,000 people and opened with a number that's probably going
0:08
Josh: to echo across Wall Street for weeks, a trillion dollars in expected orders through 2027.
0:12
Josh: That's double what he predicted just six months ago from that very same stage.
0:16
Josh: And then he spent the next two hours, this was a very long presentation,
0:21
Josh: unveiling why even a trillion dollars is conservative.
0:24
Josh: But a lot of people throughout this presentation seem to have missed the actual
0:29
Josh: reveal. I think they're focused on a few specific highlights when the reality is
0:32
Josh: The things that he presented that are going to yield this trillion dollars are
0:36
Josh: probably much different than I think the average person expects.
0:39
Josh: Ijaz, I know we were chatting as we were watching this two-hour movie marathon.
0:43
Josh: What were your thoughts? Did you make it through? Was it too boring? Was it exciting?
0:47
Josh: What were the first impressions of this presentation?
0:49
Ejaaz: The thing that excited me the most was the announcement of the DLSS 5,
0:54
Ejaaz: which seems to be the most controversial announcement.
0:57
Ejaaz: It's this new 3D rendering AI model that basically refactors old games or gaming
1:03
Ejaaz: graphics into newer, higher performance graphics.
1:07
Ejaaz: So if you're looking on the screen right now, you're seeing a version of a video
1:09
Ejaaz: game and then suddenly it's enhanced.
1:11
Ejaaz: It's kind of like a Snapchat filter, which I think a lot of the gaming community
1:15
Ejaaz: had backlash about. They thought it was just AI slop. They didn't really vibe with it.
1:19
Ejaaz: But in my opinion, it's actually quite a good product and would make me more
1:22
Ejaaz: engaged to play the actual game.
1:24
Josh: So it was surprising to see DLSS 5 get the attention that it
1:27
Josh: did just because nvidia announced some unbelievable stuff and
1:29
Josh: seemingly this was the headline at all the news outlets and it's
1:32
Josh: basically an ai upscaler for video games it takes existing
1:35
Josh: graphics that are you know pretty decent and upscales them it makes the facial
1:38
Josh: features better it increases the dynamic range the highlights the shadows and
1:42
Josh: when i saw it i loved it i was like oh this is pretty cool but the internet
1:45
Josh: reaction to this was far from what mine was i mean if you're looking at the
1:50
Josh: video on screen it increases the facial features if you're And now looking at
1:53
Josh: the meme on screen, that was the public perception.
1:55
Josh: It was also negative for such a small feature that Jensen dropped in a two-hour presentation.
2:00
Josh: So EJs, do you have any idea what's going on with this backlash here,
2:03
Josh: particularly around DLSS 5?
2:05
Ejaaz: I think the point around the gaming community is they're just very sensitive
2:07
Ejaaz: around AI being involved in art.
2:10
Ejaaz: And I get it, right? Like some things can be kind of cringe and doesn't seem
2:15
Ejaaz: very human. But the point is, like, this just makes graphics of games way,
2:18
Ejaaz: way better. I mean, the meme you're showing on the screen isn't the accurate
2:21
Ejaaz: representation of what this thing is going to do.
2:23
Ejaaz: We had the head of Bethesda Games make a partnership with NVIDIA just for this tool.
2:28
Ejaaz: It's going to save him and his team hours and hours of work.
2:31
Ejaaz: And I saw someone make a really good point online yesterday, which says,
2:34
Ejaaz: If you're a gaming developer that's spending years designing AAA games,
2:39
Ejaaz: this not only saves you a bunch of time, but it also helps you realize your artistic vision.
2:44
Ejaaz: Usually when you're a game developer, you make sacrifices when you are designing
2:48
Ejaaz: a particular character or an asset because you don't have enough money,
2:51
Ejaaz: compute, all the tools to be able to do this.
2:53
Ejaaz: This should just be seen as another tool to get to your actual vision.
2:57
Ejaaz: So I think it's a good reason.
2:58
Ejaaz: But there's another reason why this is super cool and everyone missed it, in my opinion.
3:02
Ejaaz: This is the exact same technology that you can
3:05
Ejaaz: use to create visual learning for robotics
3:09
Ejaaz: and for automotive sorry autonomous driving
3:12
Ejaaz: cars so this is the same technology that nvidia is using to build out their
3:17
Ejaaz: partner program with i think it was byd and a bunch of other car companies which
3:21
Ejaaz: we'll talk about in a second as well as being used in their robotics division
3:24
Ejaaz: with their group robotics models this is the same tech so i actually think it's
3:28
Ejaaz: cool that it's so pervasive and it's entering gaming, but that's not really
3:31
Ejaaz: the big story for me here.
3:32
Ejaaz: Whether you hate it or you like it, you're not gonna be able to use this thing
3:35
Ejaaz: for the mass audience until probably next year.
3:37
Ejaaz: This thing runs on like two RTX 5090s, which are very expensive.
3:42
Ejaaz: It's not very accessible to the average day-to-day person.
3:45
Ejaaz: So by the time it gets released to the mass audience, I think it's gonna be
3:49
Ejaaz: a lot better than what we see today.
3:50
Josh: So that's the headliner. If you are not paying attention to the headliner,
3:53
Josh: there is a lot of other stuff that was announced that is far more interesting
3:56
Josh: than this. And we have a lot to unpack, so buckle up.
3:59
Josh: Starting with the Vera Rubin platform, which is the big headliner.
4:03
Josh: I mean, this is the big boy. This is what was teased previously six months ago,
4:08
Josh: I think, when Jensen was announcing his like $500 billion in revenue.
4:12
Josh: Now he's up to a trillion.
4:13
Josh: He was unveiling a little bit more information about the chip.
4:15
Josh: Ejaz, what's new with Vera Rubin?
4:17
Ejaaz: Yeah, so the headline metric is it's 35 times more performant than the previous generation.
4:24
Ejaaz: For anyone who's been tracking, Typically, a new NVIDIA GPU gets you about a
4:28
Ejaaz: 2 to 5x performance upgrade on a good day.
4:31
Ejaaz: This is the largest jump overall. And the secret is there are about five to
4:37
Ejaaz: seven major components of a GPU.
4:39
Ejaaz: Typically, when you improve for the next generation of GPUs,
4:43
Ejaaz: you just refactor one of those things.
4:44
Ejaaz: Why? Because if you did all of them at once, that's really high risk.
4:48
Ejaaz: Anything could go wrong and it results in delays of improving your GPUs.
4:53
Ejaaz: Jensen said, forget, I'm just going to do it anyway. And he pulled it off.
4:57
Ejaaz: Seven new chips make up this entire new thing and it gets implemented into five
5:02
Ejaaz: new racks, creating what he calls on stage an AI supercomputer.
5:06
Ejaaz: And that's why you get this massive performance increase. It's just insane.
5:09
Josh: These chips are what's running all the AI that we use every single day.
5:13
Josh: Previously, everyone was training on Hopper.
5:15
Josh: Hopper was the chips that are running a lot of the AI models that you're actually using today.
5:19
Josh: The Frontier Labs have just started to spin up the Blackwell models.
5:22
Josh: That's what we've seen with Opus 4.6, what we've seen with GPT 5.4.
5:26
Josh: That's the Blackwell chip. It takes a long time from these chips to
5:28
Josh: be invented to actually roll down to data centers and then train the models what
5:32
Josh: we're seeing next and we're not going to actually feel the effects
5:34
Josh: of this until probably early next year is vera rubin and vera rubin i mean it's
5:41
Josh: a 10 times performance improvement versus blackwell just in terms of performance
5:44
Josh: per watt so for every gigawatt of energy that these data centers have this new
5:50
Josh: chip is equivalent to 10 gigawatts worth of compute today,
5:54
Josh: So for every gigawatt, you get a 10x improvement on intelligence. And that is huge.
5:58
Josh: It is an absolutely massive growth because we're planning to scale the gigawatts
6:02
Josh: of these data factories pretty significantly by the end of the year.
6:05
Josh: Six to seven gigawatts for some of these. That's going to be equivalent to 60
6:09
Josh: to 70 gigawatts of intelligence as of today.
6:11
Josh: And I think that's pretty important to note is that there is a strong delay
6:15
Josh: when it comes to these chips actually being released, actually being implemented
6:17
Josh: on the racks, trained and deployed.
6:19
Josh: It's hard to imagine we don't get AGI from this. Yeah.
6:22
Ejaaz: The other major improvement that they made was a few, I think it was about a
6:26
Ejaaz: month and a half ago, NVIDIA acquired, and I do this like this because apparently
6:31
Ejaaz: it wasn't a formal acquisition, a company called Grok, spelled G-R-O-Q.
6:37
Ejaaz: And the reason why they acquired them is they get the rights to a very special
6:41
Ejaaz: type of AI chip called an LPU, which uses something called SRAM,
6:46
Ejaaz: static random access memory.
6:48
Ejaaz: Now, if you've been keeping up to tabs with the memory walls that are happening
6:52
Ejaaz: right now, memory prices have skyrocketed.
6:55
Ejaaz: In fact, it's probably going to affect a bunch of major companies releasing
6:58
Ejaaz: their own technology devices because the cost of memory is so high,
7:02
Ejaaz: so they can't even give it to their customers because otherwise they'll need
7:04
Ejaaz: to charge extortion of prices.
7:06
Ejaaz: Jensen made a really smart move by acquiring this company and integrating their
7:10
Ejaaz: technology into Vera Rubin.
7:12
Ejaaz: And so what you're seeing on the screen now is basically the same architecture
7:17
Ejaaz: of Vera Rubin, but integrated with this SRAM technology.
7:20
Ejaaz: And the resulting effects is you can inference AI models at a much larger scale.
7:28
Ejaaz: So that 10x that you just mentioned, Josh,
7:30
Ejaaz: partially, a bunch of that is unlocked by these new LPUs.
7:35
Ejaaz: So we're now starting to see Jensen take two things more seriously.
7:39
Ejaaz: One, a different type of chip architecture. Usually, NVIDIA is known for generalized
7:43
Ejaaz: GPUs, and that's where their bread and butter is.
7:45
Ejaaz: Now we see him branching off into these hyperspecific inference chips because
7:49
Ejaaz: he looks over his shoulder and he sees, not close, but kind of far back,
7:54
Ejaaz: Google's TPUs looming, AMD's chips, and Intel's CPUs and chips coming up behind them as well.
7:59
Ejaaz: And they're all specializing in inference-specific chips. And the argument or
8:03
Ejaaz: the reason behind that is a lot of the world isn't going to be focused on training
8:07
Ejaaz: AI models. It's going to be prompting and querying AI models.
8:10
Ejaaz: And that's going to grow exponentially more.
8:12
Ejaaz: So this is NVIDIA and Jensen basically saying, we're going to make a mark here.
8:17
Ejaaz: This is our stand. This is why we acquired Grok.
8:19
Ejaaz: And here's the chip that we're doing. And VeroRubin is going to be that chip
8:21
Ejaaz: for anything and everything, general purpose and inference.
8:24
Josh: When I think about these chips and just project it out to the future it's
8:27
Josh: so exciting because there's such a
8:30
Josh: clear path to going to where i think every ai lab
8:33
Josh: wants to go yes getting to that agi level and beyond and
8:36
Josh: this chart that we're showing on screen here is a beautiful example of this
8:39
Josh: because in addition to blackwell in addition to rubin they also teased fineman
8:43
Josh: already even though rubin is months to years away from actually being deployed
8:48
Josh: at scale so nvidia is essentially 18 months give or take a few ahead of what
8:53
Josh: the current reality looks like. And I think this is really important to note.
8:57
Josh: Is currently with the bleeding edge of AI, we're running Blackwell right now.
9:02
Josh: And we just started running Blackwell.
9:04
Josh: And Blackwell has about 12 months of improvements to be made before we start
9:08
Josh: to feel the effects of Rubin.
9:09
Josh: By the time we feel the effects of Rubin, which is that 10x performance per
9:13
Josh: watt improvement, they already have Feynman ready to go and to be deployed into these data centers.
9:18
Josh: And already we have two incremental steps, two exponential steps ahead of where we currently sit.
9:23
Josh: And it's hard to imagine that with the build-out that's happening,
9:26
Josh: with the performance per watt increase that we're seeing from all these chipsets,
9:29
Josh: that we're not just going to have
9:30
Josh: this completely vertical and exponential growth of AI across the board.
9:33
Josh: And I think that's probably at the core of Jensen's thesis of a trillion dollars
9:37
Josh: is like the spending isn't going to stop because he's already created the future.
9:41
Josh: It's just a matter of actually deploying it and plugging it into the grid so
9:45
Josh: you could power these chips and get the intelligence that everyone wants. And it's unbelievable.
9:48
Josh: So Feynman is coming. They didn't announce a bunch of things about Feynman,
9:51
Josh: but that's the name of the next chip architecture.
9:54
Josh: Named after your favorite mathematicians favorite mathematician richard feinman
9:57
Josh: everyone's a big fan of him very cool very excited.
9:59
Ejaaz: His book or something i
10:01
Josh: Did yep i'm surely joking mr feinman he has a few books that are all awesome
10:05
Josh: so if you're into physics or math or just really admire great teachers richard
10:09
Josh: feinman is amazing and is now the naming architecture for the future of nvidia's
10:14
Josh: ai chips so pretty cool stuff bold.
10:17
Ejaaz: Name big ambitions um nvidia currently sits at, what, $4.5 trillion?
10:23
Ejaaz: Biggest, most valuable company in the world. Odds that it's the same by the end of the year?
10:28
Josh: Is this going to be prolonged? Well, we can ask our friends at Polymarket to
10:32
Josh: answer this for us. And it looks like there has been a strong trend signaling yes.
10:37
Josh: And this was not always the case. I mean, it looks like Alphabet,
10:40
Josh: Google, was at one point during the year, February, just a month ago,
10:43
Josh: was projected to flip them.
10:45
Josh: People thought Google was going to be the world leader. it is
10:48
Josh: clear now that is absolutely not the case in fact
10:51
Josh: apple who we frequently talk about looks like they have a better chance
10:53
Josh: of doing it than google now and now nvidia is up to 70 so it
10:57
Josh: seems highly probable that people saw this presentation people have been seeing
11:00
Josh: progress and they are very much bullish on nvidia so the market is pricing in
11:05
Josh: a pretty steep increase to the stock price before the end of the month i mean
11:08
Josh: it's currently trading at 182 and it looks like there's what 25 30 it's about
11:13
Josh: a 30 chance that it trades over 200 this month so it looks like,
11:17
Josh: Things are looking good for NVIDIA for being the most valuable company in the
11:20
Josh: world and also continuing to trade up on this news. It was an incredible presentation.
11:24
Josh: Thank you to Polymarket for sponsoring this segment of the episode.
11:27
Josh: And now we could probably get into the next most interesting thing for me,
11:31
Josh: at least, which was the full self-driving moment.
11:33
Josh: In fact, Jensen Huang said, this is the chat GPT moment for self-driving cars. It has arrived.
11:39
Josh: This is a bold take because the full self-driving industry is pretty,
11:43
Josh: pretty vicious. Hasn't it been.
11:45
Ejaaz: Solved by tesla already at this
11:47
Josh: Point well it depends who you ask it sounds like internally
11:50
Josh: they feel confident in the fact they've solved it but they're currently solving
11:53
Josh: this march of nines where they have efficacy up to 99.x percent
11:57
Josh: and they need to get it to 9999 now waymo clearly has the most deployed version
12:02
Josh: of this you could actually go and you could get into a waymo you can get into
12:05
Josh: a cyber cab in some places in austin but they still have the kind of guiding
12:09
Josh: drivers they haven't figured out the legislation to let them be fully autonomous
12:12
Josh: but jensen is saying hey if you're not Waymo,
12:15
Josh: if you're not Tesla, we have a solution for you.
12:17
Josh: We are actually going to build the full self-driving stack and integrate it
12:20
Josh: directly into your cars for you from the sensors all the way to the software stack.
12:24
Josh: And they just recently partnered with BYD, Nissan, Hyundai, and Gili.
12:28
Josh: And for those who aren't aware, BYD is actually the largest electric car manufacturer
12:32
Josh: in the world, more so than Tesla.
12:34
Josh: They're based in China and it's showing that NVIDIA is not really country.
12:39
Josh: They're kind of country agnostic, right? Like they're just going to,
12:41
Josh: if you want a self-driving car, come to us. We got you.
12:44
Ejaaz: NVIDIA or Jensen just doesn't care if he aligns with China or not.
12:48
Ejaaz: He's just like out there to expand NVIDIA into anyone and everyone's hands.
12:52
Ejaaz: As you said, BYD is the biggest EV maker. They sell more cars than Tesla every single year.
12:57
Ejaaz: And so that distribution, like think about that. Like imagine you put your self-driving
13:01
Ejaaz: model into as many cars as possible.
13:03
Ejaaz: It's probably going to get smarter way, way quicker because it's just inside
13:07
Ejaaz: more So that's real competition against Tesla from a competitive mode.
13:12
Ejaaz: The other thing is he's also integrating into Uber as well, right?
13:15
Ejaaz: So it's going to be launching in 28 cities by 2028.
13:20
Ejaaz: So through the end of next year, which seems like a long time,
13:23
Ejaaz: but that's a lot of cities.
13:24
Ejaaz: And Uber has a lot of reach when it comes to just a driving network in general.
13:28
Ejaaz: So this is a really cool announcement. I don't quite know if it's apples to
13:33
Ejaaz: apples with Tesla full self-driving.
13:35
Ejaaz: They own the end-to-end stack there. NVIDIA doesn't really have that.
13:39
Ejaaz: This is more of a thing that you can kind of attach onto cars.
13:42
Ejaaz: And if I had to guess, this is not just me being an Elon fanboy.
13:45
Ejaaz: There's a lot more friction that NVIDIA will run into. So I don't think this
13:48
Ejaaz: is a direct one-to-one competitor.
13:51
Josh: This is a key difference. If I'm a Tesla shareholder, I'm not really nervous
13:54
Josh: about this because like you said, Tesla owns the full manufacturing stack and
13:58
Josh: they have millions of cars on the road that are full self-driving capable today.
14:02
Josh: They're just one software update away from cracking that. When that
14:05
Josh: final software update comes when the legislation passes that is
14:08
Josh: to be determined but they're there they're ready waymo and
14:10
Josh: i guess uber now are kind of on the other side of this where they they've kind
14:13
Josh: of perhaps figured out the software stack they're close at least but
14:16
Josh: they they have nowhere near figured out the manufacturing stack
14:19
Josh: for this at scale and manufacturing as we know designing hard
14:23
Josh: things in the physical world is hard and that's going
14:26
Josh: to slow these companies down a lot so i think for uber this is probably
14:28
Josh: the best case scenario they finally have a saving grace someone
14:31
Josh: who wants to actually work with them to help deploy the full self-driving vision
14:34
Josh: um but they got a long way to go so it's nice that they're trying this is kind
14:38
Josh: of like apple car play but for full self-driving where they're not going to
14:42
Josh: make the cars they're going to sell you the software to put in the cars and
14:44
Josh: hopefully one day make them full self-driving so we'll see how that goes that
14:48
Josh: was the first of the robotics section of this episode let me.
14:52
Ejaaz: Let me introduce you to the unhinged uh version of this josh
14:56
Ejaaz: So Olaf, made popular in the fictional movie Frozen, came to life on stage.
15:02
Ejaaz: What you're looking at is an autonomous,
15:04
Ejaaz: self-directed robot that runs on NVIDIA. I'm not making this up.
15:09
Ejaaz: That runs on NVIDIA's Newton robotics engine.
15:12
Ejaaz: It also runs on their Jetson chip as well. So what you're looking at is a homegrown
15:17
Ejaaz: NVIDIA robot and product that is autonomously interacting with Jensen.
15:22
Ejaaz: I can't help but think that some of this must be scripted. there's no way that
15:26
Ejaaz: the robot is this interactive.
15:27
Ejaaz: And obviously, it's like been outfitted with the look of this Frozen character,
15:32
Ejaaz: but pretty cool all around.
15:33
Ejaaz: I don't know if this is going to be in everyone's home. I like I don't know
15:36
Ejaaz: what the point of this was. Like, maybe they're going to sell rights to Disney or something.
15:40
Ejaaz: But yeah, like, I don't really have a strong take on this.
15:42
Josh: Yeah, well, it's just I mean, it's more of the direction that they're heading
15:46
Josh: towards, which is real world physical AI, right? It's like we're in self driving
15:49
Josh: cars. Now we're going to get robots.
15:50
Josh: They're creating these small packaged computers to
15:53
Josh: put into these things are creating the entire stack nvidia is
15:56
Josh: becoming the tesla for the general purpose company
16:00
Josh: it's like if you can't build it all yourself nvidia has done
16:03
Josh: it and they will sell you all their hardware they'll sell you all their software
16:06
Josh: they're moving a lot into open source and i guess that's probably the transition
16:09
Josh: to the next announcement which is their nemo claw announcement the open claw
16:13
Josh: competitor which isn't an open claw competitor at all actually it's just a basically
16:17
Josh: enterprise solution for companies that want to use OpenClaw.
16:22
Josh: So the founder of OpenClaw, he was there.
16:24
Josh: Jensen gave him a nice shout-out. And basically, NemoClaw is...
16:28
Josh: A way for companies to deploy OpenClaw in a more secure way and to run on any
16:33
Josh: coding agent and deploy from anywhere.
16:35
Josh: And I think a lot of people, I mean, ourselves included, thought this could
16:38
Josh: be competition. The reality is it's complementary.
16:41
Josh: NVIDIA wants open source AI because they want to build the hardware that you
16:44
Josh: use to run the open source AI.
16:46
Josh: And it seems like this was kind of like a win for everyone, including the open
16:49
Josh: source community. It's pretty cool.
16:50
Ejaaz: Yeah. I mean, Pete Stuy, as you mentioned, the founder of OpenClaw,
16:54
Ejaaz: actually worked with Jensen and the NVIDIA team for months to build this out.
16:58
Ejaaz: Their target market are enterprise customers specifically because when OpenCore
17:03
Ejaaz: went viral, it went viral because everyone could spin up their own personal agent.
17:07
Ejaaz: There was one glaring issue, loads of security issues. So people could lose
17:11
Ejaaz: money, expose their credit card details or lose all their personal data or have
17:16
Ejaaz: people hack their computers.
17:17
Ejaaz: Not good if you are an enterprise company, but companies still want to get access to this thing.
17:22
Ejaaz: So Jensen kind of dreamt up this platform that sits on top of open call so it
17:27
Ejaaz: works very synonymously with it and now you can kind of use open call without
17:30
Ejaaz: any worry you can spin up an agent that does a particular enterprise workflow
17:34
Ejaaz: or you can use it for accounting back office stuff whatever you can dream of
17:37
Ejaaz: it's now safe to use and it's open source which is great yeah
17:40
Josh: Okay so two more things we have two more quick announcements one ai and space
17:44
Josh: space gpus we're doing the damn thing it's happening,
17:48
Josh: So Jensen got on stage. He said, we are going to build Verirubin for space.
17:53
Josh: You're going to have Verirubin orbiting the Earth. It's going to be in these
17:55
Josh: data centers. It's going to be fantastic.
17:57
Josh: And then he says, well, we're not quite sure how we're going to do it, but we're going to do it.
18:01
Josh: They still have a lot of issues that they need to solve, one of which is the
18:03
Josh: cooling, one of which is solving the radiation.
18:06
Josh: There are a series of issues that are going to need to be solved,
18:10
Josh: but there is the intention to do this.
18:12
Josh: And I suspect he didn't announce it here, but I suspect they're working with
18:14
Josh: SpaceX to design these chips hand to hand because that's really the only company
18:18
Josh: that's going to be getting these things up in space.
18:20
Josh: And I think it's really exciting when we think about AI data centers in space
18:23
Josh: and the quality of the Vera Rubin chip architecture, bringing those two things
18:27
Josh: together and getting them in orbit by 2027, 2028, maybe the latest,
18:31
Josh: that's going to be pretty cool. That's going to change the game.
18:34
Josh: Elon is incredibly bullish on this. He thinks that SpaceX is now going to flip
18:37
Josh: every company in the world when it comes to AI development.
18:40
Josh: And he might not be wrong because if he can get these chips at scale from Jensen,
18:45
Josh: send them up into orbit, lower the cost per watt to be, I mean,
18:49
Josh: a marginal of a small fraction of what it is today. It's a huge upgrade.
18:52
Ejaaz: I feel like this was just a custom announcement for Elon Musk for one individual.
18:57
Ejaaz: He's the only guy that's really trying to launch GPUs into space at scale.
19:01
Ejaaz: Like in this demo, he's demoing it using one of NVIDIA's investment portfolio
19:07
Ejaaz: companies, StarCloud, which are kind of the initial startup that made GPUs in space a trend, a thing.
19:12
Ejaaz: But then elon jumped on the wave and like completely took it over and he's the
19:15
Ejaaz: guy that's actually going to be economically able to launch these at scale um
19:19
Ejaaz: so it's it's it's a good day to be a tesla or spacex uh share owner equity owner
19:24
Josh: And the final announcement that we're going to talk about is the dgx spark they
19:28
Josh: released the new spark and it's now looking like it's going to be priced around
19:33
Josh: forty seven hundred dollars which seems high but if you are someone who runs
19:38
Josh: local inference at your home and you're considering buying a Mac Studio or something
19:42
Josh: to run these tokens on your own.
19:43
Josh: Perhaps you have an OpenClaw instance you want to run local AI.
19:46
Josh: This is a pretty compelling option. They're basically taking a GB300,
19:51
Josh: which is the Grace Blackwell chip, and they're turning it into a tiny little
19:54
Josh: thing that fits on your desk.
19:56
Josh: That's 750 gigabytes of coherent memory and 20 petaflops of AI compute,
20:02
Josh: which allows you to run models up to a trillion parameters right from your desk.
20:06
Josh: So it's an unbelievably dense machine.
20:10
Josh: In fact, if this was released probably even five years ago, this probably would
20:14
Josh: have been the most powerful supercomputer in the world.
20:16
Josh: And now it's compressed down to something that fits on your desk.
20:18
Josh: So it's just, it's a testament to how much efficiency improvements have been
20:24
Josh: made every single year and how powerful the NVIDIA brand is,
20:27
Josh: man. There's no one else building stuff like this.
20:30
Josh: No one's even close. Looking at this holistically, this is a home run for NVIDIA,
20:34
Josh: for shareholders, for investors, for the AI industry.
20:36
Josh: Everyone wins because NVIDIA is just running full tilts.
20:40
Ejaaz: It's funny, you said that back in the day, you would, this would be so much
20:43
Ejaaz: more expensive. So you would also need like a dedicated like server room to fit this entire thing.
20:48
Ejaaz: And now you can just sit it on your desk next to your laptop and have,
20:51
Ejaaz: in Jensen's words, an AI supercomputer in your house.
20:55
Ejaaz: Super cool. It comes shipped with NemoClaw as well.
20:58
Ejaaz: So you get two products, two NVIDIA GTC 2026 announcements for the price of one.
21:03
Ejaaz: And you said it was $4,700, Josh?
21:05
Josh: That's super cheap. That's what it's looking like on their website right now.
21:08
Ejaaz: I think that's for the Spark. I think that's for the Spark.
21:10
Josh: That is for the Spark.
21:11
Ejaaz: That's for the Spark, yeah. And then they had a separate announcement i think
21:14
Ejaaz: on the dgx station which is like more the more powerful supercomputer which
21:19
Ejaaz: also consequently sits on your desk as well so just two different price points
21:23
Ejaaz: but two very powerful things um yeah just a home run for nvidia yeah
21:27
Josh: What a great presentation that is everything those the highlights i would love
21:31
Josh: for you to share which part you are most excited about is it DLSS 5 how many
21:35
Josh: gamers are here that actually care.
21:37
Ejaaz: About do you hate it
21:38
Josh: Tell me. Because I don't. I think it's cool. I mean, I could understand why
21:42
Josh: the artists maybe don't like their art being, you know, digitally enhanced.
21:45
Josh: But I have good news. You could just turn it off.
21:48
Josh: Like you could just play the vanilla game too. That's also cool.
21:51
Josh: So I'd love to know what people are most excited about here.
21:54
Josh: I think for me, space data centers,
21:56
Josh: man, that's my favorite thing in the world. I want to see AI in space.
21:58
Ejaaz: It's got to be the DLSS 5, but used for robotics.
22:03
Ejaaz: Like, okay, I'm like nerding out over robotics right now, because I think they're
22:06
Ejaaz: going to have their chat GBT from 2022 moment at any point this year.
22:11
Ejaaz: They're getting good enough to move around, run, lift heavy items.
22:15
Ejaaz: We just need a good model. And I think having something like DLSS,
22:19
Ejaaz: that's this code name is, so DLSS 5, kind of like expand robotics models is really exciting for me.
22:26
Josh: So yeah. And one thing's for sure, the naming of all of these is going to continue
22:28
Josh: to be absolutely horrific.
22:30
Ejaaz: Yeah, please come up with easier names.
22:30
Josh: Oh my god. But yeah, that's a wrap. Thank you so much for watching this recap
22:35
Josh: on NVIDIA's GTC. I hope you enjoyed it.
22:37
Josh: We pursued through two hours of pretty boring presentation to bring this to
22:41
Josh: you. So hopefully it was a little more interesting, a little more exciting.
22:44
Josh: Very technical. Jensen's a technical guy. I hope you enjoyed Ejaz's leather
22:48
Josh: jacket that he's rocking today in honor of Jensen and NVIDIA,
22:52
Josh: who donned a leather jacket on stage.
22:54
Ejaaz: I know you are. I hope you appreciate it.
22:56
Josh: But yeah, as always, please don't forget to share with your friends.
22:59
Josh: Like the video subscribe leave a comment rate us five stars all the great things
23:04
Josh: thank you so much for watching and yeah we'll see you guys in the next episode.