Inside the AI Cloud Shift and the Future of Infrastructure

1

00:00:12,346 --> 00:00:13,192

Hey, everyone.

2

00:00:13,192 --> 00:00:16,192

Welcome to the AI Cloud Essentials podcast.

3

00:00:16,462 --> 00:00:18,731

I'm Lisa Martin, your host for the next couple of days.

4

00:00:18,731 --> 00:00:21,808

We are live at GTC. This is day 2.

5

00:00:22,115 --> 00:00:24,808

We are in, as you can see behind me, the buzz of the event,

6

00:00:24,808 --> 00:00:27,808

This gives you a little bit of a glimpse of the energy at GTC.

7

00:00:28,154 --> 00:00:31,500

It is electric. I’m so thrilled to be joined by Chen Goldberg.

8

00:00:31,769 --> 00:00:34,654

She's the EVP of Product & Engineering at CoreWeave.

9

00:00:34,654 --> 00:00:36,462

We're going to be talking to you a little bit about what

10

00:00:36,462 --> 00:00:38,000

Chen has been talking about with the audience,

11

00:00:38,000 --> 00:00:39,462

what she’s been hearing.

12

00:00:39,462 --> 00:00:43,231

and the overall partnership with NVIDIA and CoreWeave.

13

00:00:43,500 --> 00:00:45,038

Chen it's so great to have you on the podcast.

14

00:00:45,038 --> 00:00:46,154

Thank you so much for having me.

15

00:00:46,154 --> 00:00:47,308

Thank you so much for having me.

16

00:00:47,308 --> 00:00:48,308

So this is Day 2. Yes.

17

00:00:48,308 --> 00:00:51,731

Ane the energy yesterday was off the charts.

18

00:00:52,654 --> 00:00:56,231

People were describing this is the AI Super Bowl,

19

00:00:56,385 --> 00:00:59,038

the heartbeat of AI. Tell me a little bit...

20

00:00:59,038 --> 00:01:01,692

You were on stage with Corey yesterday.

21

00:01:01,692 --> 00:01:05,000

What were some of the things that you were sharing, and what stuck out to you

22

00:01:05,000 --> 00:01:08,231

in terms of what the audience is really absorbing.

23

00:01:08,577 --> 00:01:10,692

So this is, for me, the second time as part of CoreWeave

24

00:01:10,692 --> 00:01:12,846

that we are presenting at GTC.

25

00:01:13,423 --> 00:01:15,769

And the big change, right.

26

00:01:15,769 --> 00:01:18,038

If you think about last year,

27

00:01:18,038 --> 00:01:19,692

there was a lot of uncertainty.

28

00:01:19,692 --> 00:01:21,462

Things around, like how inference would be.

29

00:01:21,500 --> 00:01:22,615

What would be the size of the models.

30

00:01:22,615 --> 00:01:26,192

Can you create applications with AI?

31

00:01:26,692 --> 00:01:28,538

And what we’ve seen over the past

32

00:01:28,538 --> 00:01:30,423

12 months has been mind blowing.

33

00:01:30,654 --> 00:01:32,808

And everybody that is here noticed that.

34

00:01:32,808 --> 00:01:35,808

And I think that's from our perspective, the most exciting thing

35

00:01:36,115 --> 00:01:41,269

is that we, instead of just talking about basic education,

36

00:01:41,500 --> 00:01:45,385

we're talking about how can you move from experimentation to production?

37

00:01:45,692 --> 00:01:46,808

What do you need to do?

38

00:01:46,808 --> 00:01:48,808

How do you bring folks onboard?

39

00:01:48,808 --> 00:01:50,615

As an engineering leader,

40

00:01:50,615 --> 00:01:54,654

there's also a lot of conversation of productivity and the tools you are using.

41

00:01:54,692 --> 00:01:55,462

There are so many things

42

00:01:55,462 --> 00:01:59,192

that have changed over these 12 months, and that's really amazing.

43

00:01:59,385 --> 00:02:03,231

You speaking... I mean, in AI we speak of like six month,

44

00:02:03,269 --> 00:02:06,462

three month, 12 month timeframes,

45

00:02:06,462 --> 00:02:08,115

because it is moving...

46

00:02:08,115 --> 00:02:09,692

I can't even I can't describe the speed. It's amazing.

47

00:02:09,692 --> 00:02:14,308

But yesterday in Jensen's keynote was NVIDIA [heart] CoreWeave.

48

00:02:14,308 --> 00:02:15,731

Yes.

49

00:02:15,731 --> 00:02:17,115

I got chills just seeing that.

50

00:02:17,115 --> 00:02:19,731

And then when Jensen came by the booth,

51

00:02:19,731 --> 00:02:21,154

I heard him say to you and the leaders,

52

00:02:21,154 --> 00:02:22,538

just very genuine things.

53

00:02:22,538 --> 00:02:26,115

Talk about the expansion, the deepening of the NVIDIA CoreWeave

54

00:02:26,115 --> 00:02:28,115

relationship and what that will enable customers

55

00:02:28,115 --> 00:02:31,615

to do to get from what you said, experimentation to production.

56

00:02:32,731 --> 00:02:35,577

That was definitely a very special moment for us, as well.

57

00:02:35,577 --> 00:02:39,154

And we we thank Jensen for that.

58

00:02:39,154 --> 00:02:41,423

Because it was a recognition of the hard work that

59

00:02:41,423 --> 00:02:43,692

the entire CoreWeave team and our customers

60

00:02:43,692 --> 00:02:45,577

and the partners and and everybody that trusted in us,

61

00:02:45,577 --> 00:02:50,615

went through because, know...

62

00:02:50,615 --> 00:02:57,269

Everybody thinks of us as, like, a GPU reseller.

63

00:02:57,269 --> 00:03:00,038

And then recognition that we are leading in this space

64

00:03:00,038 --> 00:03:02,731

and being the AI cloud, was amazing.

65

00:03:03,115 --> 00:03:05,846

And really candidly,

66

00:03:05,846 --> 00:03:10,077

NVIDIA has been an amazing partner customer.

67

00:03:10,231 --> 00:03:12,885

And we are their customer, as well.

68

00:03:12,885 --> 00:03:16,192

And I would say that there are like a couple of things

69

00:03:16,192 --> 00:03:19,000

that really work well, between our two companies.

70

00:03:19,000 --> 00:03:22,692

One, we are really leaning in.

71

00:03:22,692 --> 00:03:24,846

Okay. We we believe in AI.

72

00:03:24,846 --> 00:03:26,885

We have a very similar vision.

73

00:03:26,885 --> 00:03:29,077

And that's really helping.

74

00:03:29,077 --> 00:03:31,538

We are really focused on customers together.

75

00:03:31,538 --> 00:03:33,808

The second thing that I think is really critical,

76

00:03:33,808 --> 00:03:36,808

you know, we talk about how our customers want to experiment.

77

00:03:37,077 --> 00:03:38,769

We like to experiment.

78

00:03:38,769 --> 00:03:40,269

Okay. I think all of us as people,

79

00:03:40,269 --> 00:03:42,692

we need to be humble and knowing like, hey, we're not great at

80

00:03:42,692 --> 00:03:43,692

predicting the future.

81

00:03:44,808 --> 00:03:45,846

But if we experiment and we see signals

82

00:03:45,846 --> 00:03:49,808

and we move fast, we get to great results.

83

00:03:49,808 --> 00:03:52,808

And NVIDIA also has that kind of culture.

84

00:03:53,192 --> 00:03:57,308

And we are both really wanting to build the best products.

85

00:03:57,846 --> 00:04:00,115

And that's actually what led, you know

86

00:04:00,115 --> 00:04:02,692

we had a big announcement last month.

87

00:04:02,692 --> 00:04:07,231

And you know the media definitely got the investment part,

88

00:04:07,500 --> 00:04:09,154

the billion dollar amount.

89

00:04:09,154 --> 00:04:12,269

But there were more things there that made me really excited.

90

00:04:12,269 --> 00:04:13,654

Tell me.

91

00:04:14,038 --> 00:04:16,615

You know, one thing was, of course, us using

92

00:04:16,615 --> 00:04:21,231

CPUs now from so Jensen in his keynote, he talked about expanding new platforms,

93

00:04:21,231 --> 00:04:24,192

so expanding from just GPUs to also CPU.

94

00:04:24,192 --> 00:04:27,692

And the other thing that we were really excited about is two things.

95

00:04:27,692 --> 00:04:30,038

One is the collaboration.

96

00:04:30,038 --> 00:04:33,769

It was actually telling the world how our teams have been collaborating

97

00:04:34,154 --> 00:04:37,308

on producing reference architecture and improving products.

98

00:04:38,000 --> 00:04:39,077

So that was one part.

99

00:04:39,077 --> 00:04:41,192

And the second part was really about us.

100

00:04:42,269 --> 00:04:45,346

Getting into the world and offering our services

101

00:04:45,346 --> 00:04:47,192

and our software to other people

102

00:04:47,462 --> 00:04:50,462

in the industry outside of even CoreWeave cloud.

103

00:04:50,577 --> 00:04:52,692

We started that with the acquisition we made

104

00:04:52,692 --> 00:04:55,077

with Weights and Biases, which is already a multi-cloud.

105

00:04:55,077 --> 00:04:58,654

But having more and more cross-cloud cloud solutions

106

00:04:58,654 --> 00:05:01,038

is something that, we plan to invest more in.

107

00:05:01,346 --> 00:05:02,577

And that's what customers are demanding.

108

00:05:02,577 --> 00:05:03,846

You talked about, basically the customer obsession

109

00:05:03,846 --> 00:05:08,385

and that symbiosis that you share with NVIDIA.

110

00:05:08,385 --> 00:05:10,692

You said something yesterday on LinkedIn, I stalked you.

111

00:05:10,692 --> 00:05:12,154

And I wanted to get your thoughts behind it.

112

00:05:12,154 --> 00:05:13,538

You were speaking with Corey.

113

00:05:13,538 --> 00:05:15,385

And you said, GTC this year 2026

114

00:05:15,385 --> 00:05:18,385

is about the next great leap.

115

00:05:18,385 --> 00:05:21,692

Tokens powering robots,

116

00:05:21,692 --> 00:05:23,500

energy grids, scientific discovery.

117

00:05:23,500 --> 00:05:24,885

I love that.

118

00:05:24,885 --> 00:05:27,846

You said, that leap needs a factory behind it.

119

00:05:27,846 --> 00:05:29,308

Talk about how NVIDIA is that factory,

120

00:05:29,462 --> 00:05:32,462

and how CoreWeave is an enabler of that.

121

00:05:32,692 --> 00:05:36,577

Back again, this year what Jensen was talking on stage,

122

00:05:36,577 --> 00:05:39,500

and also what we hear from people around us,

123

00:05:39,500 --> 00:05:44,231

is that in most companies,

124

00:05:44,231 --> 00:05:47,654

because the tools, by the way, have gotten so great

125

00:05:47,654 --> 00:05:49,462

they start seeing value.

126

00:05:49,462 --> 00:05:52,192

Yesterday on stage we had, for example,

127

00:05:52,192 --> 00:05:53,615

a person from Mercado Libre.

128

00:05:54,077 --> 00:05:56,077

So Sebastian from Mercado Libre,

129

00:05:56,077 --> 00:05:58,038

he joined us and he was telling

130

00:05:58,038 --> 00:06:01,038

the audience how they are planning to

131

00:06:01,846 --> 00:06:05,000

re imagine their search capabilities

132

00:06:05,000 --> 00:06:06,385

in their e-commerce platform.

133

00:06:06,385 --> 00:06:10,000

But even before doing that, he was talking about how even with

134

00:06:10,000 --> 00:06:13,885

small experimentation, they've been getting amazing results already.

135

00:06:14,654 --> 00:06:16,000

And what I love about it is,

136

00:06:16,000 --> 00:06:18,885

you know, like downstairs we have like a physical AI demo.

137

00:06:19,192 --> 00:06:22,192

So you see in different industry, whether it's health

138

00:06:22,500 --> 00:06:24,885

finance, e-commerce,

139

00:06:24,885 --> 00:06:29,077

media, a lot of areas where we see

140

00:06:29,577 --> 00:06:33,346

starting for a small experimentation to bigger opportunities.

141

00:06:34,038 --> 00:06:37,192

And that's really what NVIDIA is talking about.

142

00:06:37,192 --> 00:06:40,308

And part of what I think NVIDIA again, was really highlighting, Jensen

143

00:06:40,308 --> 00:06:44,385

was highlighting yesterday, that it's not just NVIDIA on its own.

144

00:06:44,808 --> 00:06:48,538

Jensen has been really investing in building an ecosystem.

145

00:06:49,308 --> 00:06:52,154

And I really appreciate that.

146

00:06:52,154 --> 00:06:55,000

I was actually part of the cloud

147

00:06:55,000 --> 00:06:59,077

transformation in the industry and the ecosystem was key to that.

148

00:06:59,462 --> 00:07:02,462

And I think Jensen is definitely recognizing

149

00:07:02,731 --> 00:07:05,731

and we are participating across the board,

150

00:07:06,000 --> 00:07:08,808

from developer tools to researcher tools

151

00:07:08,808 --> 00:07:13,000

to applications to infrastructure and just building

152

00:07:13,000 --> 00:07:16,846

that momentum, that flywheel, it creates that innovation.

153

00:07:16,885 --> 00:07:20,654

Yes. Well, the validation, the recognition you’re talking about

154

00:07:20,654 --> 00:07:22,308

from Jensen, but also to your point,

155

00:07:22,308 --> 00:07:24,808

And I talked about this on TV all the time,

156

00:07:24,808 --> 00:07:26,846

NVIDIA is not doing this alone.

157

00:07:26,846 --> 00:07:28,731

They are synonymous with AI.

158

00:07:28,731 --> 00:07:31,154

Even people locally around here that are Uber drivers

159

00:07:31,154 --> 00:07:33,000

drivers are asking me, they know AI. They see Jensen and the know AI.

160

00:07:33,000 --> 00:07:34,500

They see Jensen, AI.

161

00:07:34,500 --> 00:07:36,385

But it's the ecosystem

162

00:07:36,385 --> 00:07:39,462

and NVIDIA seems to really respect that and acknowledge it,

163

00:07:39,462 --> 00:07:42,769

obviously, with the NVIDIA [heart] CoreWeave.

164

00:07:42,769 --> 00:07:46,692

But in terms of like differentiation, there's a lot of a lot of companies here.

165

00:07:47,269 --> 00:07:51,192

The energy at this conference, I've been to a lot of them as you, is next level.

166

00:07:51,192 --> 00:07:53,769

but there's a lot of, I don't want to say me, too.

167

00:07:53,769 --> 00:07:56,423

But a lot of people saying, we’re the AI cloud.

168

00:07:56,423 --> 00:07:59,192

What makes CoreWeave really stand out

169

00:07:59,192 --> 00:08:03,154

as the essential cloud for AI in 2026 and beyond?

170

00:08:04,577 --> 00:08:05,808

I I think that a year ago

171

00:08:05,808 --> 00:08:08,769

people were not true, that we need an AI cloud.

172

00:08:09,115 --> 00:08:12,500

But now folks, that once they are trying to experiment more,

173

00:08:12,538 --> 00:08:14,077

they see there is a need.

174

00:08:14,077 --> 00:08:17,423

And one of the things, you know, and maybe I can tell you why I joined call wave.

175

00:08:17,654 --> 00:08:21,500

So before Call wave, I was part of Google Cloud and working,

176

00:08:21,500 --> 00:08:24,577

on building that cloud native ecosystem.

177

00:08:25,385 --> 00:08:28,385

And back at the day

178

00:08:28,462 --> 00:08:31,808

when we talked about the value of the cloud,

179

00:08:32,308 --> 00:08:35,077

it wasn't just enough to move to the cloud, right?

180

00:08:35,077 --> 00:08:36,269

The lift and shift.

181

00:08:36,269 --> 00:08:38,000

There was a change of ecosystem.

182

00:08:38,000 --> 00:08:39,269

You had to change the way you do tooling,

183

00:08:39,269 --> 00:08:41,538

how you develop, how you deploy, how you manage.

184

00:08:42,538 --> 00:08:44,077

And when I started

185

00:08:44,077 --> 00:08:47,500

experimenting and experiencing more of these new type of workloads,

186

00:08:48,500 --> 00:08:51,500

it felt similar, but actually a much bigger,

187

00:08:51,769 --> 00:08:54,769

opportunity to reimagine how cloud should look like.

188

00:08:55,154 --> 00:08:58,115

And I think that's the most unique thing about CoreWeave.

189

00:08:58,462 --> 00:09:01,000

And that from the get go, right,

190

00:09:01,000 --> 00:09:04,077

if you talk with the founders, that's what they were trying to do.

191

00:09:04,231 --> 00:09:09,731

It was not like me to, Peter, our CTO, they took a step back and said,

192

00:09:09,731 --> 00:09:12,692

like, okay, what are the hard problems we need to solve?

193

00:09:12,846 --> 00:09:15,192

They didn't just look at the other reference architecture.

194

00:09:15,192 --> 00:09:17,038

They actually build from scratch.

195

00:09:17,038 --> 00:09:20,077

And there are some things that we are doing very differently

196

00:09:20,308 --> 00:09:23,308

than others, and that's what really delivered the results.

197

00:09:23,808 --> 00:09:27,346

I think the way that you I, I think it's easy to understand

198

00:09:27,346 --> 00:09:30,346

there are like probably three categories that we are doing differently.

199

00:09:30,500 --> 00:09:33,423

The first one is that we are,

200

00:09:34,885 --> 00:09:37,885

started with a vertically integrated stack.

201

00:09:38,000 --> 00:09:38,346

Okay.

202

00:09:38,346 --> 00:09:38,731

You now

203

00:09:38,731 --> 00:09:42,231

see some of the folks announcing it, but it's like they're trying to retrofit

204

00:09:42,615 --> 00:09:45,885

that vertical integration was a huge difference.

205

00:09:47,308 --> 00:09:50,308

What we are doing, we understand the complexity of the stack

206

00:09:51,385 --> 00:09:52,385

is huge.

207

00:09:52,385 --> 00:09:55,885

And in order to manage that, you need to allow flexibility

208

00:09:55,885 --> 00:10:00,115

and quick decisions across the stack and knowing how to react to events

209

00:10:00,654 --> 00:10:01,423

lower in the stack.

210

00:10:01,423 --> 00:10:06,346

So we like saying from metal to model, from metal to job.

211

00:10:07,077 --> 00:10:07,500

And right.

212

00:10:07,500 --> 00:10:12,038

Bringing all of that information and really creating a new way to operate

213

00:10:12,654 --> 00:10:15,192

a cloud stack in the AI era,

214

00:10:15,192 --> 00:10:18,577

we differentiate we think capabilities like mission control

215

00:10:18,846 --> 00:10:22,615

that are both reactive and proactive in solving customer problems.

216

00:10:22,615 --> 00:10:23,885

So that would be one.

217

00:10:23,885 --> 00:10:27,538

The second part that even though we were thinking about an integrated stack,

218

00:10:28,115 --> 00:10:32,577

we are still looking for opportunities to optimize specific problems

219

00:10:32,577 --> 00:10:35,231

or bottlenecks. And again, we don't try to solve for everything.

220

00:10:35,231 --> 00:10:37,192

We are really, really focused.

221

00:10:37,192 --> 00:10:41,154

So if you think about AI workloads, and one of the challenges is

222

00:10:41,462 --> 00:10:45,346

you want to make sure the GPUs are, with high utilization.

223

00:10:45,346 --> 00:10:49,115

They are very expensive, very important resource.

224

00:10:49,423 --> 00:10:53,308

So we are building, for example, specific solutions that will bring more data

225

00:10:53,808 --> 00:10:57,192

into the GPU with our own distributed caching mechanism, or

226

00:10:57,731 --> 00:11:01,423

building a new solution for orchestration of workloads,

227

00:11:01,654 --> 00:11:05,231

for example, again, for the technical audience folks here that are familiar

228

00:11:05,231 --> 00:11:09,423

with Slurm, Slurm is a known

229

00:11:10,885 --> 00:11:13,038

industry standard for,

230

00:11:13,038 --> 00:11:16,038

HP workloads, high performing workloads.

231

00:11:16,423 --> 00:11:18,808

But what we've done, we created a solution that bring

232

00:11:18,808 --> 00:11:21,808

those workloads into the cloud native era.

233

00:11:22,269 --> 00:11:26,654

And the last point, which maybe that's really brings us back to where we started.

234

00:11:27,231 --> 00:11:30,885

We know when we talk with our customers that speed of experimentation

235

00:11:31,192 --> 00:11:33,115

is the most important thing.

236

00:11:33,115 --> 00:11:36,115

And I'm a believer that we cannot predict the future.

237

00:11:36,692 --> 00:11:40,654

So how can we increase the signals that people see?

238

00:11:41,000 --> 00:11:44,385

Okay, how quickly can I experiment, iterate and see what I need to go?

239

00:11:44,769 --> 00:11:48,538

And so we have a lot of investment on that, both from an infrastructure

240

00:11:48,538 --> 00:11:52,192

perspective but also on tunings, like we said, W and B models.

241

00:11:52,192 --> 00:11:56,462

And we've just giving those signals that allow our customers

242

00:11:56,462 --> 00:12:00,308

to make decisions faster and with confidence.

243

00:12:00,654 --> 00:12:03,000

That's the key word is confidence.

244

00:12:03,000 --> 00:12:04,808

Yeah. Because the speed is just

245

00:12:05,846 --> 00:12:07,192

not going to slow down.

246

00:12:07,192 --> 00:12:07,885

What do you think you said

247

00:12:07,885 --> 00:12:11,308

when we were preparing for this conversation was it's not just meetings.

248

00:12:12,000 --> 00:12:12,577

Yeah.

249

00:12:12,577 --> 00:12:16,423

So walk me through the steps of for each stack started to allude to it

250

00:12:17,038 --> 00:12:21,346

in a part for the technical side and explain where the differentiation lies

251

00:12:21,538 --> 00:12:24,500

and where are you really enabling customers to build

252

00:12:24,500 --> 00:12:28,269

the AI infrastructure from scratch to power?

253

00:12:28,308 --> 00:12:31,115

Awesome, and thank you so much for asking it.

254

00:12:31,115 --> 00:12:33,654

I think that we are using a lot of jargon all the time,

255

00:12:33,654 --> 00:12:35,423

and people don't always understand.

256

00:12:35,423 --> 00:12:37,538

What does it actually mean to build a cloud?

257

00:12:37,538 --> 00:12:39,269

How does it look like?

258

00:12:39,269 --> 00:12:42,769

And so first of all, there is of course the physical infrastructure.

259

00:12:43,231 --> 00:12:43,654

Okay.

260

00:12:43,654 --> 00:12:48,692

And there is already just on that layer, there's a lot of new challenges

261

00:12:48,885 --> 00:12:52,692

that appear in the AI era from a lot of power

262

00:12:52,692 --> 00:12:55,692

consumption, cooling needs.

263

00:12:56,115 --> 00:12:59,115

And there's a lot of work that our team is doing,

264

00:12:59,500 --> 00:13:02,423

around, power efficiency,

265

00:13:02,423 --> 00:13:07,538

space, liquid cooling, and just making sure that we build and,

266

00:13:07,538 --> 00:13:12,385

of course, security, across the stack, that we innovate in that space.

267

00:13:12,692 --> 00:13:16,808

But then on top of that, there is actually a different layers of the stack.

268

00:13:17,077 --> 00:13:20,692

You know, when we say infrastructure, we should all think about,

269

00:13:21,654 --> 00:13:23,654

starting with what we call infrastructure as a service.

270

00:13:23,654 --> 00:13:26,846

Very simple compute, storage and network.

271

00:13:26,846 --> 00:13:27,538

Right.

272

00:13:27,538 --> 00:13:31,000

What's been very interesting, when you move from the cloud 1.0

273

00:13:31,192 --> 00:13:34,192

to cloud 2.0, is that

274

00:13:35,077 --> 00:13:39,269

in the cloud 1.0 ERA, the goal was to make

275

00:13:39,538 --> 00:13:43,115

infrastructure, specifically compute, storage and network boring.

276

00:13:43,808 --> 00:13:44,808

Okay, okay, okay.

277

00:13:44,808 --> 00:13:48,808

I'm as a developer, as a business owner, I should know I think about it. Yes.

278

00:13:48,808 --> 00:13:50,308

Should not care about that. And. Right.

279

00:13:50,308 --> 00:13:52,385

And that's been my journey.

280

00:13:52,385 --> 00:13:54,615

In the industry making that

281

00:13:54,615 --> 00:13:57,615

a reality with technologies like Kubernetes.

282

00:13:57,769 --> 00:14:01,115

Make it make it invisible, make the infrastructure boring.

283

00:14:01,115 --> 00:14:02,731

You can you can like Google that.

284

00:14:02,731 --> 00:14:05,615

And you will see like those kind of quotes.

285

00:14:05,615 --> 00:14:08,385

And I think we've actually done a great job,

286

00:14:08,385 --> 00:14:11,115

in the industry of creating technologies that have done that.

287

00:14:11,115 --> 00:14:11,808

However,

288

00:14:13,192 --> 00:14:15,654

what's happening right now, and that's exactly what Jensen

289

00:14:15,654 --> 00:14:20,115

was talking about yesterday, is that the workloads, the applications,

290

00:14:21,538 --> 00:14:24,538

they care about, the compute storage network,

291

00:14:25,654 --> 00:14:27,038

okay, because

292

00:14:27,038 --> 00:14:30,385

the amount of data that has to go on the network is impacting the latency.

293

00:14:30,423 --> 00:14:33,423

And now let's say I have an agent,

294

00:14:33,654 --> 00:14:36,500

this is no longer it can be a mission

295

00:14:36,500 --> 00:14:40,000

critical workload that we need someone to immediately respond.

296

00:14:40,000 --> 00:14:43,000

Or if we have video generation, right, it cannot be lagging.

297

00:14:43,308 --> 00:14:46,500

And you think about storage, how quickly can I scale?

298

00:14:46,769 --> 00:14:48,846

How do I get the data, how quickly can I load things?

299

00:14:48,846 --> 00:14:52,808

And from a compute perspective, not only

300

00:14:52,808 --> 00:14:56,654

I want to get the best utilization, we actually have lots of workloads.

301

00:14:56,654 --> 00:14:59,154

I have models that are now running

302

00:14:59,154 --> 00:14:59,577

not on a

303

00:14:59,577 --> 00:15:01,000

one node or ten nodes,

304

00:15:01,000 --> 00:15:04,000

but sometimes ten thousands of nodes.

305

00:15:04,154 --> 00:15:07,308

And that's creates a lot of, complexity in that stack.

306

00:15:07,308 --> 00:15:10,154

And we are been innovating in that area area.

307

00:15:10,154 --> 00:15:11,269

And then on top of this

308

00:15:12,385 --> 00:15:13,269

we have

309

00:15:13,269 --> 00:15:16,192

a new inference services and training services.

310

00:15:16,192 --> 00:15:16,346

Right.

311

00:15:16,346 --> 00:15:21,231

Because as a developer, I have new tasks that I need to perform.

312

00:15:21,423 --> 00:15:24,231

So I need new tools and new ecosystem.

313

00:15:24,231 --> 00:15:25,192

And maybe on top of that,

314

00:15:25,192 --> 00:15:28,885

I think the things that really excites me the most is our entire what we call our,

315

00:15:29,808 --> 00:15:32,231

you know, when we think about serverless, it's not about making

316

00:15:32,231 --> 00:15:35,231

the infrastructure boring.

317

00:15:35,462 --> 00:15:37,885

But helping our customers

318

00:15:37,885 --> 00:15:40,846

rely on us in making some of those decisions.

319

00:15:41,115 --> 00:15:44,115

So one of the things that I'm really excited about is,

320

00:15:44,462 --> 00:15:48,000

the tools that we are providing, it's like what we call that serverless layer,

321

00:15:48,000 --> 00:15:51,231

where we are not trying to really make the infrastructure disappear,

322

00:15:51,769 --> 00:15:55,654

but we're trying to help customers, to rely on us.

323

00:15:56,231 --> 00:15:56,500

Okay.

324

00:15:56,500 --> 00:16:00,385

And lets us do the heavy lifting in making those tough decisions.

325

00:16:01,038 --> 00:16:04,308

And we've been announcing this week and you,

326

00:16:04,308 --> 00:16:07,846

tool in our serverless RL, which is a reinforcement learning.

327

00:16:08,269 --> 00:16:11,731

And in that space, if before a lot of this experimentation

328

00:16:12,038 --> 00:16:16,269

required simulation data, now we are allowing customers to use production

329

00:16:16,269 --> 00:16:21,000

tracing in order to train the agents to get better automatically.

330

00:16:21,346 --> 00:16:24,038

So that would be one example of a token production.

331

00:16:24,038 --> 00:16:25,731

Yes. In production.

332

00:16:25,731 --> 00:16:28,231

Oh, for example, we have a new customer, Klein,

333

00:16:28,231 --> 00:16:29,577

that they are using our inference service.

334

00:16:29,577 --> 00:16:32,577

So they are a customer, a partner as well.

335

00:16:32,846 --> 00:16:33,654

And they are

336

00:16:34,808 --> 00:16:37,192

actually building a coding solution

337

00:16:37,192 --> 00:16:40,192

that allows them to leverage our platform.

338

00:16:40,500 --> 00:16:42,115

Maybe, you know, you asked about differentiation.

339

00:16:42,115 --> 00:16:43,769

I should have led with that.

340

00:16:43,769 --> 00:16:47,269

And we offer all this new stuff

341

00:16:48,000 --> 00:16:52,462

without compromising security and resiliency and production ready.

342

00:16:52,846 --> 00:16:55,269

And we take a lot of pride in that.

343

00:16:55,269 --> 00:16:58,346

And that's like, actually would be probably the number one reason

344

00:16:58,346 --> 00:16:59,731

why customers come to us.

345

00:16:59,731 --> 00:17:03,269

And second last question, what you talk about some of the challenges

346

00:17:03,269 --> 00:17:08,423

customers have, technical challenges, power, tooling, space, capacity.

347

00:17:09,077 --> 00:17:13,115

What are some of the business challenges that you're proud to for me, to help

348

00:17:13,154 --> 00:17:16,154

such customers eliminate,

349

00:17:16,577 --> 00:17:17,885

don't. There are many challenges.

350

00:17:17,885 --> 00:17:22,231

And, you know, like, we were just say I had a customer meeting, today

351

00:17:22,538 --> 00:17:25,308

and they said, like, you know, every time we saw the hard problem,

352

00:17:25,308 --> 00:17:27,346

a new hard problem pops up.

353

00:17:27,346 --> 00:17:30,577

So, it feels like that's the business that we are in.

354

00:17:31,808 --> 00:17:35,500

But the thing that I feel like we are most helping

355

00:17:35,846 --> 00:17:38,846

customers, one is speed of innovation.

356

00:17:39,346 --> 00:17:42,269

And the idea that our customers are telling us

357

00:17:42,269 --> 00:17:45,269

that on day one, they can be productive,

358

00:17:45,692 --> 00:17:48,692

that they have the signals that they can make the right decisions,

359

00:17:49,692 --> 00:17:52,000

that we are not wasting their time,

360

00:17:52,000 --> 00:17:54,885

okay, that their team is productive

361

00:17:54,885 --> 00:17:57,077

is really, really important.

362

00:17:57,077 --> 00:17:58,654

There's no competitive advantage for them.

363

00:17:58,654 --> 00:18:01,038

Yes, yes, for sure.

364

00:18:01,038 --> 00:18:04,038

The second thing is that,

365

00:18:06,038 --> 00:18:08,731

We do see our customers as partners.

366

00:18:08,731 --> 00:18:12,615

I think it's a privilege for us to partner

367

00:18:12,615 --> 00:18:16,077

with them and enable this next, gen innovation.

368

00:18:16,885 --> 00:18:18,808

And it looks differently. Okay.

369

00:18:18,808 --> 00:18:23,231

It looks in a way that our customers, we are not just talking about

370

00:18:24,385 --> 00:18:27,385

the platform or what services we are talking to them about.

371

00:18:27,462 --> 00:18:30,462

What changes do they have, how can we help?

372

00:18:30,577 --> 00:18:33,462

And if they need us, we'll be there, okay.

373

00:18:33,462 --> 00:18:36,462

We'll be there to help them with whatever they need.

374

00:18:36,577 --> 00:18:40,308

And as part of that, state of mind,

375

00:18:40,615 --> 00:18:43,500

it's also impact of how we support our customers.

376

00:18:43,500 --> 00:18:48,654

We are actually not a really like a traditional, support process.

377

00:18:49,154 --> 00:18:52,769

And the way we are working, we believe in engineering

378

00:18:52,769 --> 00:18:57,000

to engineering relationships because our customers are sophisticated.

379

00:18:57,385 --> 00:18:59,000

They have the hardest problem.

380

00:18:59,000 --> 00:19:02,423

They want to know that they can quickly find solutions.

381

00:19:02,885 --> 00:19:05,808

And so we really automated and manage that process.

382

00:19:05,808 --> 00:19:07,231

We call it direct to expert.

383

00:19:08,462 --> 00:19:11,154

And we have engineers working with our customers day

384

00:19:11,154 --> 00:19:14,269

in day out on the hardest problems, same language like

385

00:19:15,269 --> 00:19:18,731

difference from a differentiation perspective and really getting in there

386

00:19:18,731 --> 00:19:22,885

and allowing those experts to deliver to their customers what they expect.

387

00:19:22,885 --> 00:19:23,846

For sure.

388

00:19:23,846 --> 00:19:26,500

So you're not a mind reader, can't predict a future.

389

00:19:26,500 --> 00:19:29,500

But a year from now, GTC 2027,

390

00:19:30,115 --> 00:19:33,115

we think we might see.

391

00:19:33,577 --> 00:19:34,615

That's it.

392

00:19:34,615 --> 00:19:36,577

I keep thinking, I keep thinking about it.

393

00:19:36,577 --> 00:19:40,346

If someone would have told me and that like in

394

00:19:40,346 --> 00:19:43,346

2025 would look the way it is, I would not have believed it.

395

00:19:43,462 --> 00:19:43,692

Okay?

396

00:19:43,692 --> 00:19:47,231

There's been so many things that happened that surprised all of us

397

00:19:47,500 --> 00:19:49,346

from an advancement perspective.

398

00:19:49,346 --> 00:19:53,231

However, I do expect that we will see,

399

00:19:54,654 --> 00:19:56,308

an explosion of use cases.

400

00:19:57,885 --> 00:20:03,769

And. More and more people will speak the AI language.

401

00:20:04,308 --> 00:20:06,038

It's so powerful.

402

00:20:06,038 --> 00:20:09,038

And the tools are becoming so accessible

403

00:20:09,462 --> 00:20:11,346

that I, I think that

404

00:20:11,346 --> 00:20:15,308

we will see less people be worried about it,

405

00:20:15,308 --> 00:20:18,308

because what we are seeing today, that the people that lean in

406

00:20:19,038 --> 00:20:21,500

find that is their superpower.

407

00:20:21,500 --> 00:20:23,346

Yeah. Yes.

408

00:20:23,346 --> 00:20:25,000

Yeah. I love that leaning in.

409

00:20:25,000 --> 00:20:28,000

And thank you so much for an outstanding conversation.

410

00:20:28,115 --> 00:20:31,038

I appreciate you taking some time to be on the podcast and share

411

00:20:31,038 --> 00:20:33,769

what you're talking about, what you're hearing from customers,

412

00:20:33,769 --> 00:20:37,077

the strength of the NVIDIA partnership, and also other partnerships

413

00:20:37,308 --> 00:20:41,846

in that ecosystem that really make AI accessible for customers.

414

00:20:41,846 --> 00:20:43,115

I appreciate your time. Thank you.

415

00:20:43,115 --> 00:20:44,885

Thank you so much for having me.

416

00:20:44,885 --> 00:20:47,231

For Chen Goldberg, I'm Lisa Martin. You're watching

417

00:20:47,231 --> 00:20:50,385

AI Cloud Essentials podcast live from GTC.

418

00:20:50,654 --> 00:20:52,654

Thanks for watching guys. We'll see you on the next pod.

‍