【The Motley Fool】AMD引爆CPU市场：现在值得买入的3只潜力股

1
00:00:00,000 --> 00:00:00,960
欢迎回到频道

2
00:00:00,960 --> 00:00:02,580
各位在过去一年里

3
00:00:02,580 --> 00:00:06,219
华尔街所有人都完全痴迷于图形处理单元

4
00:00:06,219 --> 00:00:10,279
GPU已成为标准人工智能硬件加速器

5
00:00:10,279 --> 00:00:13,820
但当整个零售市场关注数据中心的一侧时

6
00:00:13,820 --> 00:00:17,320
另一侧悄然发生了重大结构转变

7
00:00:17,320 --> 00:00:20,789
一位顶级半导体高管刚刚登上财报电话会

8
00:00:24,259 --> 00:00:25,599
但在开始之前

9
00:00:25,599 --> 00:00:29,439
如果你想了解聪明资金是如何悄然推动这场隐藏的技术繁荣

10
00:00:29,439 --> 00:00:32,799
请立即点击订阅按钮并开启通知

11
00:00:32,799 --> 00:00:33,159
是的

12
00:00:33,159 --> 00:00:33,380
瑞秋

13
00:00:33,380 --> 00:00:34,740
在AMD最近的季度财报中

14
00:00:34,740 --> 00:00:36,039
一次财报电话会议

15
00:00:36,039 --> 00:00:42,710
CEO丽莎·苏博士完全改变了市场对长期预期的判断

16
00:00:42,710 --> 00:00:47,359
空头一直认为传统CPU已过时

17
00:00:47,359 --> 00:00:50,679
在AI主导的世界中传统技术

18
00:00:50,679 --> 00:00:53,020
苏丽莎完全推翻了这一论调

19
00:00:53,020 --> 00:00:57,619
她宣布服务器CPU的可寻址市场将以三

20
00:00:57,619 --> 00:01:00,390
五个百分点年复合增长率增长至2030年

21
00:01:00,390 --> 00:01:00,670
现在

22
00:01:00,670 --> 00:01:06,040
这一增速远超管理层去年11月公布的18%

23
00:01:06,040 --> 00:01:09,180
这意味着他们已将长期市场预测翻倍

24
00:01:09,180 --> 00:01:15,079
从600亿美元一路增长至2030年超1200亿美元

25
00:01:15,079 --> 00:01:16,739
今天我们将解析

26
00:01:16,739 --> 00:01:18,319
这一转变的原因

27
00:01:18,319 --> 00:01:21,659
以及我们正在关注的三只股票

28
00:01:21,659 --> 00:01:24,079
以捕捉这一重大拐点

29
00:01:24,079 --> 00:01:24,519
何塞

30
00:01:24,519 --> 00:01:26,099
让我们从你的第一选择开始

31
00:01:26,099 --> 00:01:26,659
是的

32
00:01:26,659 --> 00:01:30,180
显然我必须从这场冲击波的源头说起

33
00:01:30,180 --> 00:01:31,319
超威半导体

34
00:01:31,319 --> 00:01:34,969
代码AMT，市场完全误解了

35
00:01:34,969 --> 00:01:37,299
为何苏丽莎在财报会上将服务器

36
00:01:37,299 --> 00:01:40,659
CPU前景上调至超1200亿美元

37
00:01:40,659 --> 00:01:43,450
所有关键点都指向一场重大转型

38
00:01:43,450 --> 00:01:47,549
从基础大语言模型转向ENAI系统

39
00:01:47,549 --> 00:01:51,769
训练模型高度依赖GPU

40
00:01:51,769 --> 00:01:55,510
但当在全球规模部署自主AI代理时

41
00:01:55,510 --> 00:01:57,709
能够生成子任务的系统

42
00:01:57,709 --> 00:01:58,569
执行代码

43
00:01:58,569 --> 00:02:00,840
管理数据管道并搜索网络

44
00:02:00,840 --> 00:02:04,060
主要瓶颈转向系统编排

45
00:02:04,060 --> 00:02:07,859
计算代理需要大量顺序计算能力

46
00:02:07,859 --> 00:02:11,008
这在最近季度由CPU驱动

47
00:02:11,008 --> 00:02:13,788
AMD在本季度展现了绝对的行业标杆

48
00:02:13,788 --> 00:02:17,069
总收入同比激增38%至103亿美元

49
00:02:17,069 --> 00:02:22,169
实现每股收益1.37美元

50
00:02:22,169 --> 00:02:24,310
这轻松击败了街道预期

51
00:02:24,310 --> 00:02:28,229
他们的数据中心业务引领增长飙升57%

52
00:02:28,229 --> 00:02:31,439
同比增长7%至58亿美元

53
00:02:31,439 --> 00:02:35,189
主要得益于史诗级服务器CPU销售

54
00:02:35,189 --> 00:02:36,530
更重要的是

55
00:02:36,530 --> 00:02:39,009
自由现金流超过三倍创纪录

56
00:02:39,009 --> 00:02:39,669
达26亿美元

57
00:02:39,669 --> 00:02:44,409
随着第六代架构量产推进

58
00:02:44,409 --> 00:02:49,349
AMD正瞄准服务器CPU市场主导份额

59
00:02:49,349 --> 00:02:53,229
他们在加速器领域是英特尔的替代方案

60
00:02:53,229 --> 00:02:57,569
他们在计算领域是绝对技术领导者

61
00:02:57,569 --> 00:02:58,769
目前

62
00:02:59,049 --> 00:03:01,649
你提到的基因AI理论

63
00:03:01,649 --> 00:03:02,729
这是个绝妙观点

64
00:03:02,729 --> 00:03:07,080
如果数百万AI代理同时运行子任务

65
00:03:07,080 --> 00:03:11,990
数据中心通用计算层必须指数级扩展

66
00:03:11,990 --> 00:03:16,210
我想看看运营利润率和执行风险

67
00:03:16,210 --> 00:03:18,879
AMD的数据中心增长令人印象深刻

68
00:03:18,879 --> 00:03:21,240
现在确实不容否认

69
00:03:21,240 --> 00:03:25,379
一季度整体净利润环比下降

70
00:03:25,379 --> 00:03:26,219
我得说

71
00:03:26,219 --> 00:03:27,479
看看这家AMD

72
00:03:27,479 --> 00:03:31,580
投入数十亿美元转向先进封装产线

73
00:03:31,580 --> 00:03:34,479
为下一代产品锁定代工厂产能

74
00:03:34,479 --> 00:03:39,189
这些是季度环比下滑的合理解释

75
00:03:39,189 --> 00:03:40,710
我们也看到阻力因素

76
00:03:40,710 --> 00:03:42,550
以及游戏和嵌入式业务

77
00:03:42,550 --> 00:03:43,530
当然近期

78
00:03:43,530 --> 00:03:47,520
但如果现金流持续消耗

79
00:03:47,520 --> 00:03:48,819
只为维持R

80
00:03:48,819 --> 00:03:54,180
D对高端GPU和新扩展CPU市场的需求

81
00:03:54,180 --> 00:03:58,889
你认为他们能否快速提升盈利以支撑当前估值

82
00:03:58,889 --> 00:03:59,870
还是担心

83
00:03:59,870 --> 00:04:01,830
可能被拉得太薄

84
00:04:01,830 --> 00:04:06,280
在与英特尔和英伟达的双线竞争中

85
00:04:06,639 --> 00:04:07,039
是的

86
00:04:07,039 --> 00:04:07,259
没错

87
00:04:07,259 --> 00:04:09,120
这才是关键所在

88
00:04:09,120 --> 00:04:11,560
Rachel关于该业务拖累

89
00:04:11,560 --> 00:04:16,788
但正是为何CPU TAM扩展成为服务器领域重大催化剂

90
00:04:16,788 --> 00:04:22,269
CPU通常拥有更高的成熟毛利率

91
00:04:22,269 --> 00:04:24,329
相比早期阶段低良率AI加速器

92
00:04:24,329 --> 00:04:24,769
现在

93
00:04:24,769 --> 00:04:30,509
管理层已预测数据中心业务将成长为绝对核心

94
00:04:30,509 --> 00:04:36,610
企业毛利率将在未来几季持续提升至5-6%区间

95
00:04:36,610 --> 00:04:39,490
他们并未过度扩张

96
00:04:39,490 --> 00:04:40,410
在我看来

97
00:04:40,410 --> 00:04:42,449
他们正用高毛利

98
00:04:42,449 --> 00:04:47,410
史诗级CPU现金流来完全自筹AI硬件路线

99
00:04:47,410 --> 00:04:49,910
像微软这样的超大规模云服务商

100
00:04:49,910 --> 00:04:51,069
Azure和亚马逊

101
00:04:51,069 --> 00:04:55,089
AWS目前正在积极收购X86服务器

102
00:04:55,089 --> 00:04:59,658
以应对每日处理流量的十倍增长

103
00:04:59,658 --> 00:05:03,379
A和D具备填补供应真空的能力

104
00:05:03,379 --> 00:05:04,358
而英特尔

105
00:05:04,358 --> 00:05:05,778
不幸仍在挣扎

106
00:05:05,778 --> 00:05:10,158
因广泛报道的晶圆厂延迟和处理器短缺

107
00:05:10,158 --> 00:05:13,579
市场正将AMD视为周期性芯片股

108
00:05:13,579 --> 00:05:17,980
但其实际已是数据中心结构性垄断

109
00:05:17,980 --> 00:05:18,500
是的

110
00:05:18,500 --> 00:05:19,680
我认为没错

111
00:05:19,680 --> 00:05:19,939
我是说

112
00:05:19,939 --> 00:05:23,019
如果云数据中心的可见性也清晰可见

113
00:05:23,019 --> 00:05:24,620
我们转向你的第二选择

114
00:05:24,620 --> 00:05:24,819
不过

115
00:05:24,819 --> 00:05:27,120
因为这同样涉及架构转变

116
00:05:27,120 --> 00:05:31,649
但将其应用于本地设备而非云端

117
00:05:31,649 --> 00:05:32,170
没错

118
00:05:32,170 --> 00:05:35,560
第二只股票是高通，代码QC O M

119
00:05:35,560 --> 00:05:40,959
高通大家常将其与周期性智能手机市场关联

120
00:05:40,959 --> 00:05:45,439
但Lisa Se的宣布彻底改变计算逻辑

121
00:05:45,439 --> 00:05:46,480
如果电路

122
00:05:46,480 --> 00:05:48,819
如果服务器市场以30%

123
00:05:48,819 --> 00:05:52,629
年复合增长率突破1200亿美元

124
00:05:52,629 --> 00:05:58,199
关键问题是基础设施如何应对海量新负载

125
00:05:58,199 --> 00:06:04,319
高通悄然定位为这里的终极黑马

126
00:06:04,319 --> 00:06:07,360
但需审视其商业模式

127
00:06:07,360 --> 00:06:08,800
他们分为两个业务板块

128
00:06:08,800 --> 00:06:09,040
Q

129
00:06:09,040 --> 00:06:09,259
C

130
00:06:09,259 --> 00:06:09,670
T

131
00:06:09,670 --> 00:06:13,129
半导体产品贡献约86%营收

132
00:06:13,129 --> 00:06:14,709
然后是QTL

133
00:06:14,709 --> 00:06:19,040
技术授权业务占比14%

134
00:06:19,040 --> 00:06:24,160
其中QTL手机业务达60亿美元

135
00:06:24,160 --> 00:06:26,680
全球驱动高端安卓手机

136
00:06:26,680 --> 00:06:31,500
在三星Galaxy旗舰线占据超70%份额

137
00:06:31,500 --> 00:06:35,430
物联网覆盖工业边缘设备

138
00:06:35,430 --> 00:06:36,709
消费电子

139
00:06:36,709 --> 00:06:38,949
及增强现实与虚拟现实

140
00:06:38,949 --> 00:06:42,079
上季度营收17亿美元

141
00:06:42,079 --> 00:06:47,309
但真正爆发性增长在于

142
00:06:47,309 --> 00:06:52,000
他们利用庞大移动现金流

143
00:06:52,160 --> 00:06:56,079
进军汽车和数据中心市场

144
00:06:56,079 --> 00:06:58,879
这种多元化策略纸面看起来很棒

145
00:06:58,879 --> 00:07:02,480
显然他们进入企业数据中心领域

146
00:07:02,480 --> 00:07:07,240
这意味着要与 entrenched巨头多线作战

147
00:07:07,240 --> 00:07:10,680
若讨论定制硅基高速基础设施

148
00:07:10,680 --> 00:07:11,160
博通和Marvell绝对主导该领域

149
00:07:11,160 --> 00:07:11,399
现在

150
00:07:11,399 --> 00:07:12,819
高通是一家原生移动企业

151
00:07:12,819 --> 00:07:15,750
试图进军企业云网络

152
00:07:15,750 --> 00:07:18,689
而超大规模数据中心需要数年的可信验证

153
00:07:18,689 --> 00:07:22,949
高通在服务器架构领域缺乏传统积累

154
00:07:22,949 --> 00:07:25,170
我很好奇你对这个的看法

155
00:07:25,170 --> 00:07:25,610
是的

156
00:07:25,610 --> 00:07:28,790
我认为这种缺乏历史记录

157
00:07:28,790 --> 00:07:33,939
正是现在给予高通巨大安全边际的关键

158
00:07:33,939 --> 00:07:37,189
前瞻业绩市盈率约23倍

159
00:07:37,189 --> 00:07:37,829
嗯

160
00:07:37,829 --> 00:07:40,639
所以不算明显高估的股票

161
00:07:40,639 --> 00:07:44,920
高通通过整合Orion架构

162
00:07:50,459 --> 00:07:50,660
我

163
00:07:50,660 --> 00:07:50,980
P

164
00:07:50,980 --> 00:07:54,480
他们通过收购AlphaWave实现突破

165
00:07:54,480 --> 00:07:58,259
这并非对2030年的投机押注

166
00:07:58,259 --> 00:07:58,540
没错

167
00:07:58,540 --> 00:08:03,449
CFO在2026财年财报中确认

168
00:08:03,449 --> 00:08:08,810
与头部超大规模客户签订首批定制芯片订单

169
00:08:08,810 --> 00:08:13,000
将于2026年12月季度开始交付

170
00:08:13,000 --> 00:08:19,660
CFO证实该定制芯片合作预计达运营利润率

171
00:08:19,660 --> 00:08:21,500
从第一天起

172
00:08:26,139 --> 00:08:29,189
新加入的高通CPU

173
00:08:29,189 --> 00:08:29,730
嗯

174
00:08:29,730 --> 00:08:32,429
进入同一市场构成真实威胁

175
00:08:32,429 --> 00:08:35,970
可能被市场低估的风险

176
00:08:35,970 --> 00:08:36,590
所以如果

177
00:08:36,590 --> 00:08:39,789
即使他们在定制芯片和数据中心执行成功

178
00:08:39,789 --> 00:08:41,460
那其他方面呢

179
00:08:41,460 --> 00:08:41,940
例如

180
00:08:41,940 --> 00:08:45,720
手机业务高度周期性且需拓展新工业领域

181
00:08:45,720 --> 00:08:45,879
是的

182
00:08:45,879 --> 00:08:48,519
需要大量资本投入

183
00:08:48,960 --> 00:08:49,480
嗯

184
00:08:49,480 --> 00:08:49,879
嗯

185
00:08:49,879 --> 00:08:53,600
完全同意Rachel，这才是终极验证点

186
00:08:53,600 --> 00:08:58,309
汽车业务首次实现年化五亿美元营收

187
00:08:58,309 --> 00:08:59,649
五年前

188
00:08:59,649 --> 00:09:03,240
汽车业务在财务模型中几乎无足轻重

189
00:09:03,240 --> 00:09:05,879
如今成为增长最快的QCC产品线

190
00:09:05,879 --> 00:09:10,710
上季度营收13亿美元，同比增长38%

191
00:09:10,710 --> 00:09:15,600
公司预计2026财年将达六亿美元年化规模

192
00:09:15,600 --> 00:09:17,980
在下一财季

193
00:09:17,980 --> 00:09:19,500
2026年汽车业务

194
00:09:19,500 --> 00:09:24,509
预计年增率将进一步加速至50%

195
00:09:24,509 --> 00:09:28,490
得益于第四代骁龙数字座舱

196
00:09:28,490 --> 00:09:30,470
集成系统处理

197
00:09:30,470 --> 00:09:31,740
数字座舱功能

198
00:09:31,740 --> 00:09:32,539
连接性

199
00:09:32,539 --> 00:09:34,639
高级驾驶辅助与自动驾驶

200
00:09:34,639 --> 00:09:36,120
到年底时

201
00:09:36,120 --> 00:09:39,490
其第五代骁龙芯片预计发货

202
00:09:39,490 --> 00:09:40,250
实现

203
00:09:40,250 --> 00:09:47,490
CPU吞吐量提升三倍，NPU性能达当前的十二倍

204
00:09:47,490 --> 00:09:50,009
每辆车的配置将大幅跃升

205
00:09:50,009 --> 00:09:51,289
就在此刻

206
00:09:51,289 --> 00:09:54,600
ADAS应用正在拐点转折

207
00:09:54,600 --> 00:10:00,639
但为何车辆NPU和CPU性能对数据中心爆发至关重要

208
00:10:00,639 --> 00:10:03,019
我认为这可能是我最大的疑问

209
00:10:03,299 --> 00:10:08,080
因为生成式AI不仅需要海量数据传感器

210
00:10:08,080 --> 00:10:08,600
没错瑞秋

211
00:10:08,600 --> 00:10:10,799
当AI代理运行在车内

212
00:10:10,799 --> 00:10:13,399
任务调度的编排层

213
00:10:13,399 --> 00:10:20,019
融合传感器数据并管理多步骤工具使用全程在CPU本地运行

214
00:10:20,019 --> 00:10:20,740
高通

215
00:10:20,740 --> 00:10:21,399
已投入多年

216
00:10:21,399 --> 00:10:26,740
打造移动与汽车平台性能最强的CPU

217
00:10:26,740 --> 00:10:32,980
生成式负载的到来将这笔投资转化为直接竞争优势

218
00:10:32,980 --> 00:10:33,539
我的意思是

219
00:10:33,539 --> 00:10:36,940
首席执行官克里斯蒂亚诺在2026财年第二季度财报中提到

220
00:10:36,940 --> 00:10:42,700
指出生成式AI工作负载以OpenCL为例

221
00:10:42,700 --> 00:10:47,940
正在根本性改变用户体验并重塑各平台路线图

222
00:10:47,940 --> 00:10:48,889
他们开发

223
00:10:48,889 --> 00:10:51,809
现有安装基础并未为此类功能设计

224
00:10:51,809 --> 00:10:57,659
代表巨大升级机会和可扩展市场增长

225
00:10:57,659 --> 00:10:59,039
而在构建过程中

226
00:10:59,039 --> 00:11:01,019
他们拥有QCL授权业务

227
00:11:01,019 --> 00:11:05,379
持有3G/4G/5G无线技术基础专利

228
00:11:05,379 --> 00:11:08,860
从全球每部智能手机中收取授权费

229
00:11:08,860 --> 00:11:12,700
上季度实现72%的EBIT利润率

230
00:11:12,700 --> 00:11:17,659
这是半导体行业最持久的高毛利收入流之一

231
00:11:17,659 --> 00:11:21,970
为其研发引擎提供结构性安全垫

232
00:11:21,970 --> 00:11:24,269
高通正利用移动现金流布局

233
00:11:24,269 --> 00:11:29,460
捕捉终端设备爆发的前端与服务器爆炸的后端

234
00:11:29,460 --> 00:11:30,720
说到这瑞秋

235
00:11:30,720 --> 00:11:36,559
我知道你带来了第三只股票从完全不同的角度切入这一基建扩张

236
00:11:36,559 --> 00:11:38,460
我准备好了，你有什么？

237
00:11:38,460 --> 00:11:40,820
你提到了AMD和高通

238
00:11:40,820 --> 00:11:44,039
所以我必须引入台积电

239
00:11:44,039 --> 00:11:46,129
台积电(TMC)看

240
00:11:46,129 --> 00:11:50,090
我们可以争论AMD是否会从英特尔抢夺服务器份额

241
00:11:50,090 --> 00:11:52,980
或高通能否超越传统客户端架构

242
00:11:52,980 --> 00:11:55,039
但有一个不可否认的现实

243
00:11:55,039 --> 00:11:57,480
无论芯片设计战谁胜谁负

244
00:11:57,480 --> 00:12:01,460
所有人都必须依赖台积电进行晶圆制造

245
00:12:01,460 --> 00:12:05,460
他们是整个先进计算领域的绝对先锋

246
00:12:05,460 --> 00:12:07,779
当丽莎·苏在财报电话会上表示

247
00:12:07,779 --> 00:12:11,440
服务器CPU市场将翻倍至1200亿美元以上

248
00:12:11,440 --> 00:12:14,980
因为处理需求的增长速度远超预期

249
00:12:14,980 --> 00:12:18,679
这实际上直接转化为多年的市场需求管道

250
00:12:18,679 --> 00:12:20,960
针对台积电的先进晶圆节点

251
00:12:20,960 --> 00:12:24,240
AMD的第五代和第六代EPYC处理器

252
00:12:24,240 --> 00:12:25,720
高通骁龙平台

253
00:12:25,720 --> 00:12:27,159
英伟达的加速器

254
00:12:27,159 --> 00:12:30,929
这些均采用台积电5纳米制程

255
00:12:30,929 --> 00:12:33,350
3纳米及即将推出的2纳米生产线

256
00:12:33,350 --> 00:12:37,080
因此台积电拥有绝对且无可匹敌

257
00:12:37,080 --> 00:12:38,460
持续的定价权

258
00:12:38,460 --> 00:12:41,000
这确实是因为他们是地球上唯一

259
00:12:41,000 --> 00:12:43,919
能够制造这些超密集架构的晶圆厂

260
00:12:43,919 --> 00:12:45,068
实现商业良率

261
00:12:45,068 --> 00:12:45,749
所以

262
00:12:45,749 --> 00:12:50,948
他们能将原材料和先进封装成本直接转嫁给客户

263
00:12:50,948 --> 00:12:51,308
我意思是

264
00:12:51,308 --> 00:12:56,339
这个行业历史上毛利率长期保持在50%以上

265
00:12:56,339 --> 00:12:56,859
是的

266
00:12:56,859 --> 00:13:00,278
瑞秋，没人能否认台积电是一家

267
00:13:00,278 --> 00:13:02,178
管理卓越的优质企业

268
00:13:02,178 --> 00:13:06,240
他们是现代文明的基石

269
00:13:06,240 --> 00:13:08,240
但当我们谈论台积电

270
00:13:08,240 --> 00:13:12,230
必须正视其内部的巨大集中化风险

271
00:13:12,230 --> 00:13:15,450
他们在台湾存在极端的地理集中

272
00:13:15,450 --> 00:13:18,759
使其面临巨大的地缘政治波动风险

273
00:13:18,759 --> 00:13:22,860
作为全球唯一的先进晶圆代工厂

274
00:13:22,860 --> 00:13:27,179
每年必须投入巨额资本支出

275
00:13:27,179 --> 00:13:31,110
仅用于建设洁净室和购买昂贵的工厂设备

276
00:13:31,110 --> 00:13:32,549
如果我们看到十年后半段

277
00:13:32,549 --> 00:13:35,440
芯片出现周期性过剩

278
00:13:35,440 --> 00:13:38,639
或亚利桑那和欧洲的新工厂

279
00:13:38,639 --> 00:13:42,578
遭遇长期运营延迟和超预期人力成本

280
00:13:42,578 --> 00:13:45,558
这些巨额固定折旧费用

281
00:13:45,558 --> 00:13:48,960
绝对会摧毁其运营利润率

282
00:13:49,240 --> 00:13:50,039
所以你知道的

283
00:13:50,039 --> 00:13:51,200
地缘政治焦虑是

284
00:13:51,200 --> 00:13:51,519
我认为

285
00:13:51,519 --> 00:13:54,980
我们看到台积电被交易

286
00:13:54,980 --> 00:13:56,240
所谓折扣价

287
00:13:56,240 --> 00:13:58,980
相对于其长期增长潜力和实力

288
00:13:58,980 --> 00:14:00,899
作为长期投资者

289
00:14:00,899 --> 00:14:04,320
这可能是一个极具吸引力的入场窗口

290
00:14:04,320 --> 00:14:07,120
但让我们审视这笔资本开支风险的实际运营情况

291
00:14:07,120 --> 00:14:11,788
台积电不会随意建设数十亿美元的工厂

292
00:14:11,788 --> 00:14:13,749
或者说他们基于

293
00:14:13,749 --> 00:14:17,179
来自超大规模客户的长期资本承诺

294
00:14:17,179 --> 00:14:18,240
这些客户包括苹果

295
00:14:18,240 --> 00:14:19,480
你知道的英伟达、AMD

296
00:14:19,480 --> 00:14:19,840
当然

297
00:14:19,840 --> 00:14:21,039
今天我们讨论过的企业

298
00:14:21,039 --> 00:14:24,919
因此他们的客户正在提前为这些产能路线图提供资金

299
00:14:24,919 --> 00:14:26,950
以确保他们的分配名额

300
00:14:26,950 --> 00:14:32,960
现在台积电正在积极执行大规模多元化战略

301
00:14:32,960 --> 00:14:36,860
他们在亚利桑那州正在加速先进封装和生产节点的建设

302
00:14:36,860 --> 00:14:38,720
日本德国

303
00:14:38,720 --> 00:14:39,159
我是说

304
00:14:39,159 --> 00:14:43,179
即使这些国际晶圆厂的初期建设成本更高

305
00:14:43,179 --> 00:14:45,259
回到台积电的定价能力

306
00:14:45,259 --> 00:14:50,590
他们确实可以为地理分散的本土硅芯片收取溢价

307
00:14:50,590 --> 00:14:52,830
作为投资者我来说

308
00:14:52,830 --> 00:14:54,750
我认为最大的收获是

309
00:14:54,750 --> 00:14:58,909
你无法规模化或构建价值1200亿美元的服务器

310
00:14:58,909 --> 00:15:03,720
CPU市场或运行本地边缘设备而不支付台积电的税费

311
00:15:03,960 --> 00:15:04,320
呃

312
00:15:04,320 --> 00:15:04,759
我同意

313
00:15:04,759 --> 00:15:05,120
瑞秋

314
00:15:05,120 --> 00:15:07,159
这就是本期节目的结尾

315
00:15:07,159 --> 00:15:07,440
我是说

316
00:15:07,440 --> 00:15:09,419
人工智能革命正在扩展至

317
00:15:09,419 --> 00:15:14,220
远超主流媒体过去两年报道的初始硬件层面

318
00:15:14,220 --> 00:15:17,179
随着生成式AI和本地化推理规模扩大

319
00:15:17,179 --> 00:15:22,609
基础设施需求正推动高性能通用计算的剧烈转变

320
00:15:22,609 --> 00:15:27,600
我们想知道莉萨·苏是否正确预判了服务器CPU的复兴

321
00:15:27,600 --> 00:15:32,589
或是市场高估了自主AI代理的计算需求

322
00:15:32,589 --> 00:15:34,349
在评论区留下你的观点

323
00:15:34,349 --> 00:15:36,179
重新阅读所有评论

324
00:15:36,179 --> 00:15:38,419
如果你觉得这次半导体分析有帮助

325
00:15:38,419 --> 00:15:39,580
点击点赞按钮

326
00:15:39,580 --> 00:15:41,580
订阅频道获取更多研究内容

327
00:15:41,580 --> 00:15:42,100
深度解析

328
00:15:42,100 --> 00:15:44,950
我们将在下一视频与大家相见

329
00:15:44,950 --> 00:15:45,919
保重

评论采集失败

采集完成，该视频无弹幕