1
00:00:00,040 --> 00:00:02,320
Speaker 1: We are live here in San Jose with Lisa Sue,

2
00:00:02,520 --> 00:00:05,120
Speaker 1: the COO a MD. It's been incredibly busy day for you.

3
00:00:05,200 --> 00:00:07,760
Speaker 1: But there's a lot of emphasis on the importance of

4
00:00:07,920 --> 00:00:11,640
Speaker 1: MI I three hundred x, your latest AI accelerator. But

5
00:00:11,760 --> 00:00:14,920
Speaker 1: the technological difference visa v the H one hundred in

6
00:00:15,040 --> 00:00:17,360
Speaker 1: videos is HBM.

7
00:00:17,079 --> 00:00:18,599
Speaker 2: Three high bandwidth memory.

8
00:00:19,000 --> 00:00:21,560
Speaker 1: But what advantage does that give you in the immediate

9
00:00:21,680 --> 00:00:25,360
Speaker 1: term against what is a clear market incumbent in the space.

10
00:00:25,600 --> 00:00:27,600
Speaker 3: Yeah, well, first of all, it's great to be here.

11
00:00:27,680 --> 00:00:29,920
Speaker 3: Thank you guys for being on site and spending so

12
00:00:30,000 --> 00:00:32,600
Speaker 3: much time with us today. It's been a big, big

13
00:00:32,680 --> 00:00:36,520
Speaker 3: day for AMD. We're so excited about first the opportunity

14
00:00:36,520 --> 00:00:40,280
Speaker 3: in AI is just absolutely exploding. And then we're talking

15
00:00:40,280 --> 00:00:42,960
Speaker 3: today about the launch of our I three hundred x,

16
00:00:42,960 --> 00:00:45,440
Speaker 3: which is our you know, let's call it the leading

17
00:00:45,560 --> 00:00:48,919
Speaker 3: edge data center AI accelerator. And you know, we were

18
00:00:48,920 --> 00:00:50,800
Speaker 3: here with a lot of our partners as well. So

19
00:00:50,960 --> 00:00:53,520
Speaker 3: you know your comment about you know, what's special about

20
00:00:53,760 --> 00:00:56,720
Speaker 3: three hundred x, I mean, the truth is we've all

21
00:00:56,760 --> 00:00:59,760
Speaker 3: experienced over the last you know, twelve months, this incredible

22
00:01:00,560 --> 00:01:03,800
Speaker 3: right you know, chat GPT has has really changed the

23
00:01:03,840 --> 00:01:06,440
Speaker 3: way we think about what tech can do, and the

24
00:01:06,600 --> 00:01:10,319
Speaker 3: underlying capability of that is GPUs and you know, very

25
00:01:10,440 --> 00:01:13,399
Speaker 3: very capable of GPUs. You know, we made some very

26
00:01:13,440 --> 00:01:16,120
Speaker 3: very good decisions, you know, a few years ago about

27
00:01:16,160 --> 00:01:20,720
Speaker 3: how to put together this technology and that includes both

28
00:01:20,760 --> 00:01:24,440
Speaker 3: being very good for training, so training large models, but

29
00:01:24,480 --> 00:01:28,039
Speaker 3: also very good for answering questions or inference. So when

30
00:01:28,080 --> 00:01:30,920
Speaker 3: you ask you know, the chat rept a question, it

31
00:01:30,959 --> 00:01:32,720
Speaker 3: takes sometimes a little bit of time for it to

32
00:01:32,760 --> 00:01:33,880
Speaker 3: respond to an answer.

33
00:01:34,000 --> 00:01:35,280
Speaker 2: Some latency, there's.

34
00:01:35,080 --> 00:01:38,520
Speaker 3: Some latency there, and you know, we've found you know,

35
00:01:38,600 --> 00:01:42,000
Speaker 3: really a great technological solution by adding you know, lots

36
00:01:42,120 --> 00:01:45,280
Speaker 3: of high bandwidth memory or memory capacity.

37
00:01:45,200 --> 00:01:48,120
Speaker 1: Which video will not have into H two hundred second

38
00:01:48,200 --> 00:01:49,240
Speaker 1: quarter of next year.

39
00:01:49,840 --> 00:01:52,720
Speaker 3: That is correct. We are industry leaning, so you know,

40
00:01:52,800 --> 00:01:55,240
Speaker 3: best in class in terms of inference performance.

41
00:01:55,440 --> 00:01:58,240
Speaker 1: And what is the side by side Lisa on training

42
00:01:58,280 --> 00:02:01,400
Speaker 1: and performance? Am I three hundred versus H one hundred.

43
00:02:01,480 --> 00:02:03,520
Speaker 3: Yeah, So if you look at we've showed some of

44
00:02:03,560 --> 00:02:06,560
Speaker 3: the benchmarks earlier today, If you look at training performance,

45
00:02:06,880 --> 00:02:08,639
Speaker 3: we're very very competitive, let's call it.

46
00:02:08,680 --> 00:02:10,000
Speaker 2: You know, it's a toss up.

47
00:02:10,320 --> 00:02:13,360
Speaker 3: When you look at inference performance, we're one point four

48
00:02:13,400 --> 00:02:16,160
Speaker 3: to one point six times better. And you know what

49
00:02:16,160 --> 00:02:18,680
Speaker 3: that means is, you know, if you're running these models,

50
00:02:18,680 --> 00:02:20,960
Speaker 3: you can actually run more models, or you can run

51
00:02:21,040 --> 00:02:24,239
Speaker 3: larger models. You know, with MI I three hundred and

52
00:02:24,560 --> 00:02:28,440
Speaker 3: right now, you know the key to AI is GPU compute.

53
00:02:28,480 --> 00:02:31,040
Speaker 3: I mean, that is absolutely what everybody says, and so

54
00:02:31,080 --> 00:02:33,320
Speaker 3: we're here to provide lots of GPU compute.

55
00:02:33,520 --> 00:02:38,720
Speaker 1: You've had the confidence to dramatically alter your your forecast

56
00:02:38,760 --> 00:02:42,160
Speaker 1: for this market for AI accelerators. You're saying a total

57
00:02:42,280 --> 00:02:46,160
Speaker 1: addressable market of four hundred billion US dollars in twenty

58
00:02:46,200 --> 00:02:49,160
Speaker 1: twenty seven. In August, just in August you said it

59
00:02:49,200 --> 00:02:50,760
Speaker 1: was one hundred and fifty billion.

60
00:02:51,160 --> 00:02:52,280
Speaker 2: What has changed?

61
00:02:52,600 --> 00:02:55,240
Speaker 3: Yeah, And you know, really the way we look at

62
00:02:55,240 --> 00:02:57,080
Speaker 3: these things is we usually look at these things on

63
00:02:57,120 --> 00:02:59,720
Speaker 3: an annual basis, and so you know, when we were

64
00:03:00,200 --> 00:03:03,760
Speaker 3: doing our plan for twenty twenty three and beyond last year,

65
00:03:04,480 --> 00:03:06,920
Speaker 3: we thought that you know, this year there would be

66
00:03:07,080 --> 00:03:09,400
Speaker 3: about a thirty billion dollar market and it would grow

67
00:03:09,560 --> 00:03:13,000
Speaker 3: you know, fifty percent compound annual growth rates, so be

68
00:03:13,000 --> 00:03:15,960
Speaker 3: about one hundred and fifty billion in twenty twenty seven. Well,

69
00:03:16,000 --> 00:03:19,400
Speaker 3: frankly was very very large. But what's changed is we

70
00:03:19,680 --> 00:03:22,640
Speaker 3: can all see what's changed, right. People need more compute.

71
00:03:22,680 --> 00:03:26,280
Speaker 3: They're installing more. You know, the numbers for this year

72
00:03:26,320 --> 00:03:29,799
Speaker 3: are probably closer to forty five billion. And when we

73
00:03:29,840 --> 00:03:32,440
Speaker 3: talk to customers, when I spend time with our partners,

74
00:03:32,480 --> 00:03:36,640
Speaker 3: and you know, what they tell us is the technology

75
00:03:36,680 --> 00:03:40,360
Speaker 3: requires more compute. And so we now believe the total

76
00:03:40,400 --> 00:03:43,120
Speaker 3: market for this it's upwards of four hundred billion and

77
00:03:43,160 --> 00:03:44,000
Speaker 3: twenty twenty seven.

78
00:03:44,040 --> 00:03:44,720
Speaker 2: It's huge.

79
00:03:45,200 --> 00:03:46,920
Speaker 3: There's no one size fits all. They're going to be

80
00:03:47,000 --> 00:03:51,160
Speaker 3: multiple solutions. There are lots of good solutions out there today,

81
00:03:51,280 --> 00:03:55,800
Speaker 3: but we believe the AMD capability is very significant, and

82
00:03:56,120 --> 00:03:57,200
Speaker 3: so we're excited about it.

83
00:03:57,200 --> 00:04:00,400
Speaker 1: It was interesting to see on stage how mi I

84
00:04:00,440 --> 00:04:02,640
Speaker 1: three hundred x manifests itself in the real world, but

85
00:04:02,680 --> 00:04:05,600
Speaker 1: you'd already guided us that it will likely be the

86
00:04:05,920 --> 00:04:09,600
Speaker 1: quickest AMD product to one billion dollars. There were sections

87
00:04:09,600 --> 00:04:12,080
Speaker 1: of the market in the street that said your forecast

88
00:04:12,200 --> 00:04:14,360
Speaker 1: of two billion dollars of sales for mi I three

89
00:04:14,400 --> 00:04:18,560
Speaker 1: hundred x in twenty four was conservative. If you're saying

90
00:04:18,600 --> 00:04:21,640
Speaker 1: that the total addressable market by twenty twenty seven is

91
00:04:21,760 --> 00:04:26,120
Speaker 1: now four hundred billion, then is that two billion forecast

92
00:04:26,160 --> 00:04:28,720
Speaker 1: for next year, specifically for MI I three hundred x

93
00:04:28,839 --> 00:04:30,640
Speaker 1: conservative as the market thinks it.

94
00:04:30,600 --> 00:04:32,919
Speaker 3: Is, well, I think you have to take a step

95
00:04:32,960 --> 00:04:35,839
Speaker 3: back and just look at how this technology is evolving. So,

96
00:04:36,400 --> 00:04:39,480
Speaker 3: you know, we did update in our last conference call

97
00:04:39,720 --> 00:04:42,880
Speaker 3: to an expectation about two billion in twenty twenty four

98
00:04:43,520 --> 00:04:47,720
Speaker 3: for our data CENTERGYPUS. It's a very early estimate. I

99
00:04:47,760 --> 00:04:49,560
Speaker 3: would say, you know, we have clearer line of sight

100
00:04:49,640 --> 00:04:51,960
Speaker 3: to that. But you know, what people ask me is

101
00:04:52,320 --> 00:04:55,400
Speaker 3: like there is much more customer demand, definitely, and there's

102
00:04:55,440 --> 00:04:59,359
Speaker 3: also significantly more supply because we've had to prepare the

103
00:04:59,360 --> 00:05:02,320
Speaker 3: supply chain so that we're ready to ramp. So we'll

104
00:05:02,400 --> 00:05:05,120
Speaker 3: update as we go along. You know, we are, you know,

105
00:05:05,240 --> 00:05:08,800
Speaker 3: definitely on this path to ramp I three hundred the

106
00:05:08,880 --> 00:05:12,599
Speaker 3: fastest is anything's ever ramped at AMD And you know,

107
00:05:12,640 --> 00:05:16,120
Speaker 3: I view this as a multi year opportunity for us.

108
00:05:16,560 --> 00:05:19,880
Speaker 1: A reminders to our Bloomberg television and radio audience worldwide.

109
00:05:19,880 --> 00:05:22,159
Speaker 1: We're live with Lisa Sue, the a M D CEO

110
00:05:22,480 --> 00:05:25,320
Speaker 1: here in San Jose. I mean, supply is a key

111
00:05:25,400 --> 00:05:28,200
Speaker 1: question because when you say about two billion dollars, about

112
00:05:28,240 --> 00:05:30,960
Speaker 1: could mean less or more than two billion dollars but

113
00:05:31,600 --> 00:05:32,000
Speaker 1: what is.

114
00:05:31,960 --> 00:05:33,480
Speaker 2: The state of supply right now?

115
00:05:33,880 --> 00:05:36,520
Speaker 1: Has it improved such that actually you could exceed your

116
00:05:36,520 --> 00:05:40,480
Speaker 1: expectations because you have visibility on a greater volume of

117
00:05:40,880 --> 00:05:42,640
Speaker 1: GPUs to hand over to customers.

118
00:05:42,839 --> 00:05:46,359
Speaker 3: Yeah, for sure. When we plan, we plan for success,

119
00:05:46,440 --> 00:05:50,200
Speaker 3: and so our planning has the capability to be significantly

120
00:05:50,279 --> 00:05:53,640
Speaker 3: higher than two billion. We have, you know, customer demand,

121
00:05:53,960 --> 00:05:56,440
Speaker 3: you know, serve lots and lots of interest for I

122
00:05:56,480 --> 00:05:58,599
Speaker 3: three hundred and I think the key for us is,

123
00:05:58,640 --> 00:06:00,200
Speaker 3: you know, one step at a time. Right to day

124
00:06:00,279 --> 00:06:02,880
Speaker 3: was a huge day in terms of the launch. We're

125
00:06:02,920 --> 00:06:05,480
Speaker 3: actively in deployment with a number of the customers and

126
00:06:05,520 --> 00:06:10,480
Speaker 3: partners you know, Microsoft on stage, Oracle, Meta, our OEM partners,

127
00:06:10,560 --> 00:06:12,240
Speaker 3: del Lenovo, super Micro.

128
00:06:13,000 --> 00:06:15,680
Speaker 2: Everyone is really doing.

129
00:06:15,480 --> 00:06:18,440
Speaker 3: Just phenomenal HPE on the three hundred A side, So

130
00:06:19,200 --> 00:06:22,400
Speaker 3: a great, great set of partners and great partnerships for

131
00:06:22,520 --> 00:06:24,320
Speaker 3: us to ramp as fast as possible.

132
00:06:24,600 --> 00:06:28,760
Speaker 1: What's happening right now is you have AMD coming to

133
00:06:28,760 --> 00:06:31,920
Speaker 1: the cutting edge with three hundred x by adding HBM three.

134
00:06:32,839 --> 00:06:35,520
Speaker 1: Nvidia has the H one hundred h two hundreds coming,

135
00:06:35,520 --> 00:06:38,279
Speaker 1: they have Grayhoppers, super Chit and at the same time,

136
00:06:38,720 --> 00:06:43,240
Speaker 1: the hyperscalers are really aggressively investing in their own silicon.

137
00:06:43,680 --> 00:06:45,400
Speaker 2: How does that work in practice?

138
00:06:45,440 --> 00:06:47,520
Speaker 1: If you're trying to say I've got the cutting edge

139
00:06:47,520 --> 00:06:50,919
Speaker 1: and AI accelerators and the hyperscaler saying right, I also

140
00:06:50,960 --> 00:06:53,040
Speaker 1: have the cutting edge in AI accelerators.

141
00:06:53,320 --> 00:06:56,000
Speaker 2: Are you competitors? Are you collaborators? Which is it?

142
00:06:56,400 --> 00:06:59,159
Speaker 3: I think we are first and foremost collaborators. I mean,

143
00:06:59,520 --> 00:07:03,480
Speaker 3: you know what we see that's really happening is everybody

144
00:07:03,560 --> 00:07:07,480
Speaker 3: realizes the foundation is the silicon compute. So of course

145
00:07:07,480 --> 00:07:11,200
Speaker 3: people are going to invest in silicon. Now, from my standpoint,

146
00:07:11,560 --> 00:07:14,680
Speaker 3: compute is hard, and it's especially hard if you're trying

147
00:07:14,680 --> 00:07:18,040
Speaker 3: to address the bleeding edge. So you know, our expectation

148
00:07:18,160 --> 00:07:21,320
Speaker 3: is there will be solutions. There will be some proprietary solutions.

149
00:07:21,360 --> 00:07:23,720
Speaker 3: There will be a lot of GPUs. You know, in

150
00:07:23,760 --> 00:07:26,400
Speaker 3: my four hundred billion dollar TAM, I would say it's

151
00:07:26,720 --> 00:07:30,720
Speaker 3: predominantly GPUs, and we work in collaborations, so there will

152
00:07:30,760 --> 00:07:33,880
Speaker 3: be multiple solutions. But for the largest language models, for

153
00:07:33,920 --> 00:07:38,160
Speaker 3: the most complex workloads, we believe that we're extremely well positioned.

154
00:07:38,360 --> 00:07:41,400
Speaker 1: Actually, a question from our Bloomberg Technology audience globally when

155
00:07:41,400 --> 00:07:43,960
Speaker 1: I said that you're coming on the show, is take

156
00:07:44,040 --> 00:07:47,040
Speaker 1: that tam for twenty twenty seven four hundred billion, But

157
00:07:47,160 --> 00:07:49,880
Speaker 1: tell us how much of it is driven by inference

158
00:07:49,880 --> 00:07:52,560
Speaker 1: and how much is driven by training, because there's a

159
00:07:52,640 --> 00:07:54,560
Speaker 1: chance that a lot of the training is complete by then.

160
00:07:54,880 --> 00:07:56,080
Speaker 2: Yeah.

161
00:07:56,160 --> 00:07:57,720
Speaker 3: By the way, I don't think the training will be

162
00:07:57,720 --> 00:08:00,640
Speaker 3: complete by them, because I think there will be a

163
00:08:00,680 --> 00:08:03,520
Speaker 3: desire to continue to get better to you know, if

164
00:08:03,560 --> 00:08:06,600
Speaker 3: you think about you know, what we're really looking for is,

165
00:08:06,720 --> 00:08:10,520
Speaker 3: you know, how does AI really become, you know, as sophisticated,

166
00:08:10,640 --> 00:08:13,440
Speaker 3: as capable as humans. There's still a lot that we

167
00:08:13,480 --> 00:08:16,680
Speaker 3: can do. But that being the case, we do view

168
00:08:16,720 --> 00:08:21,160
Speaker 3: that the inference market will even grow faster, that will

169
00:08:21,200 --> 00:08:23,520
Speaker 3: be even more queries. And so you know, if I

170
00:08:23,520 --> 00:08:25,520
Speaker 3: look at twenty twenty seven, I think more than half

171
00:08:25,520 --> 00:08:26,160
Speaker 3: the market.

172
00:08:26,040 --> 00:08:28,680
Speaker 2: Will be inference, more than half inference. Where is it

173
00:08:28,760 --> 00:08:29,200
Speaker 2: right now?

174
00:08:29,560 --> 00:08:32,559
Speaker 3: It's predominantly training right now, as people.

175
00:08:32,400 --> 00:08:34,840
Speaker 1: Are and I mean in the context specifically of demand

176
00:08:34,880 --> 00:08:36,080
Speaker 1: for I three hundred x.

177
00:08:36,200 --> 00:08:39,160
Speaker 3: Yeah, So if I think about the market today, there's

178
00:08:39,160 --> 00:08:41,120
Speaker 3: a lot of training. I think if you think about

179
00:08:41,400 --> 00:08:43,920
Speaker 3: three hundred X and what we see in twenty twenty four,

180
00:08:44,280 --> 00:08:47,440
Speaker 3: it's a good balance between training and inference. But certainly

181
00:08:47,480 --> 00:08:50,000
Speaker 3: on inference. We just have killer performance.

182
00:08:50,160 --> 00:08:52,040
Speaker 1: So a lot of the chatter that here in the

183
00:08:52,120 --> 00:08:55,520
Speaker 1: valley is no matter how good the GPUs are, in

184
00:08:55,559 --> 00:08:58,400
Speaker 1: some places, the software that comes with it and manages

185
00:08:58,440 --> 00:09:01,000
Speaker 1: it is not that good. And one of the questions

186
00:09:01,040 --> 00:09:03,199
Speaker 1: put to me for you is how much are you

187
00:09:03,280 --> 00:09:05,760
Speaker 1: going to invest in software and how good do you

188
00:09:05,800 --> 00:09:06,920
Speaker 1: think you are at software?

189
00:09:07,160 --> 00:09:11,600
Speaker 3: Yeah. Look, we've spent a significant amount of resources, both

190
00:09:11,720 --> 00:09:15,240
Speaker 3: organically and inorganically. We just acquired a couple of companies

191
00:09:15,840 --> 00:09:18,319
Speaker 3: to augment our software resource standpoint.

192
00:09:18,679 --> 00:09:20,280
Speaker 2: We think we're very well positioned.

193
00:09:20,520 --> 00:09:24,280
Speaker 3: Today we announced our next generation rock them six, which

194
00:09:24,320 --> 00:09:26,320
Speaker 3: is really designed for Gen AI workloads. I know it's

195
00:09:26,360 --> 00:09:29,280
Speaker 3: a little bit of a detail. What customers are telling

196
00:09:29,360 --> 00:09:32,200
Speaker 3: us is m I three hundred is actually really easy

197
00:09:32,240 --> 00:09:35,080
Speaker 3: to use. You know, we've gotten sort of the heavy

198
00:09:35,120 --> 00:09:39,319
Speaker 3: lifting done. We've really focused on these higher level frameworks.

199
00:09:39,360 --> 00:09:42,960
Speaker 3: So you know, people really liked actually building models and

200
00:09:43,000 --> 00:09:47,280
Speaker 3: building their applications in PyTorch. And you know, PyTorch is

201
00:09:47,800 --> 00:09:51,000
Speaker 3: an open ecosystem. It works very very well with AMD

202
00:09:51,240 --> 00:09:53,040
Speaker 3: and so these are you know, some of many steps

203
00:09:53,080 --> 00:09:56,440
Speaker 3: we announced this morning that open Ai Triton is also

204
00:09:57,360 --> 00:10:00,520
Speaker 3: you know, optimizing with AMD on their next revision. We're

205
00:10:00,600 --> 00:10:03,839
Speaker 3: making a lot of progress and for sure, I think

206
00:10:03,840 --> 00:10:06,320
Speaker 3: on the software side, we're absolutely ready, Lisa.

207
00:10:06,559 --> 00:10:08,480
Speaker 1: Even in the short time I've been in Silicon Valley

208
00:10:08,520 --> 00:10:10,559
Speaker 1: six years, people have said, A m D won't do it.

209
00:10:10,840 --> 00:10:13,480
Speaker 1: They won't they won't beat they won't enter the market,

210
00:10:13,679 --> 00:10:17,400
Speaker 1: Intel will beat them on PC. In the context of AI,

211
00:10:17,559 --> 00:10:19,960
Speaker 1: will you beat in video or will you be competitive?

212
00:10:20,520 --> 00:10:23,880
Speaker 3: You know what I'd like to say is we are

213
00:10:24,240 --> 00:10:27,120
Speaker 3: very very focused on our roadmap. Ed I have to say,

214
00:10:27,720 --> 00:10:30,840
Speaker 3: this is about what do we believe is important for

215
00:10:30,880 --> 00:10:33,920
Speaker 3: the market and how are we shooting for you know,

216
00:10:33,920 --> 00:10:35,679
Speaker 3: where the market is going. So yeah, I think we're

217
00:10:35,720 --> 00:10:37,480
Speaker 3: going to do great in AI. I mean I think

218
00:10:37,520 --> 00:10:40,200
Speaker 3: AI is our number one priority. Hopefully that was clear today.

219
00:10:40,800 --> 00:10:43,319
Speaker 3: You know, we've pivoted the company to really focus on AI.

220
00:10:43,720 --> 00:10:45,680
Speaker 3: I think they're going to be multiple winners in AI.

221
00:10:45,880 --> 00:10:49,120
Speaker 3: And as you know, kind of important as the cloud is,

222
00:10:49,160 --> 00:10:52,160
Speaker 3: we think enterprise is really important. We think HPC is

223
00:10:52,240 --> 00:10:55,640
Speaker 3: very important, we think PC's are very important, and you know,

224
00:10:55,679 --> 00:10:57,920
Speaker 3: this is kind of the next big wave in tech.

225
00:10:58,400 --> 00:11:02,239
Speaker 2: A m dco Lisa South, thank you. This is Bloomberg

226
00:11:03,640 --> 00:11:03,679
Speaker 2: m