1
00:00:00,000 --> 00:00:01,800
我們要帶你們來看到這個是聯發科

2
00:00:01,800 --> 00:00:03,500
現在我們看到了這個業界

3
00:00:03,500 --> 00:00:06,100
產業界大家對它的一個前景有志一同

4
00:00:06,100 --> 00:00:08,000
到底呢 聯發科最近在做什麼

5
00:00:08,000 --> 00:00:10,400
趕快有請這個國政哥帶我們來看看囉

6
00:00:10,400 --> 00:00:11,500
聯發科告訴大家說

7
00:00:11,500 --> 00:00:12,800
這個客製化的晶片

8
00:00:12,800 --> 00:00:15,700
這個市占率目標上修到20%

9
00:00:15,700 --> 00:00:18,000
他選擇了一條跟博通不一樣的路

10
00:00:18,000 --> 00:00:18,800
到底為什麼

11
00:00:18,800 --> 00:00:21,700
對 我們來看聯發科

12
00:00:21,700 --> 00:00:23,200
我們來看聯發科這一次說的

13
00:00:23,200 --> 00:00:24,100
進來一點 進來一點

14
00:00:24,100 --> 00:00:26,300
我們來

15
00:00:26,300 --> 00:00:28,000
我們來看聯發科這次說法

16
00:00:28,000 --> 00:00:28,600
說說了什麼

17
00:00:28,600 --> 00:00:29,780
他把ESC的晶片

18
00:00:29,780 --> 00:00:32,000
本來他預估的目標是10到15%

19
00:00:32,000 --> 00:00:34,740
他突然說上修到15到20%

20
00:00:34,740 --> 00:00:36,060
當然上限就是20%

21
00:00:36,060 --> 00:00:38,600
他為什麼對ESC這麼的有信心

22
00:00:39,220 --> 00:00:40,760
而且未來占比會很高

23
00:00:40,760 --> 00:00:42,900
他其實走了一條跟博通

24
00:00:42,900 --> 00:00:44,240
完全不同的道路

25
00:00:44,240 --> 00:00:45,200
他告訴別人是

26
00:00:45,200 --> 00:00:47,160
他把這個目標修正之後

27
00:00:47,160 --> 00:00:48,860
他同時去做一件事情

28
00:00:48,860 --> 00:00:52,740
是聯發科會做系統級的設計

29
00:00:52,740 --> 00:00:54,420
他不只要做晶片

30
00:00:54,420 --> 00:00:55,640
包括他要替Google

31
00:00:55,640 --> 00:00:57,420
去做所謂的基板的設計

32
00:00:57,420 --> 00:01:00,220
等于我帮Google可能往下去做到

33
00:01:00,220 --> 00:01:01,880
所谓的窄版这块的设计

34
00:01:01,880 --> 00:01:03,060
以及基板的设计

35
00:01:03,060 --> 00:01:05,000
甚至我接下来要替Meta

36
00:01:05,000 --> 00:01:07,060
他们去做伺服器的机柜

37
00:01:07,060 --> 00:01:07,940
这很不一样

38
00:01:07,940 --> 00:01:11,200
它不是一家单纯的IC设计公司

39
00:01:11,200 --> 00:01:12,940
它已经变成是

40
00:01:12,940 --> 00:01:15,320
我要整体帮你设计到

41
00:01:15,320 --> 00:01:18,540
整个AI伺服器的生意了

42
00:01:18,540 --> 00:01:20,560
所以包括Google的运算基板L6

43
00:01:20,560 --> 00:01:23,060
未来它除了替它做TPU之外

44
00:01:23,060 --> 00:01:24,920
它也帮它做所谓的

45
00:01:24,920 --> 00:01:25,940
整个基板的设计

46
00:01:25,940 --> 00:01:27,640
再來馬斯克的伺服器的機櫃

47
00:01:27,640 --> 00:01:29,340
雖然馬斯克現在在稱讚說

48
00:01:29,340 --> 00:01:30,720
灰島的晶片很好

49
00:01:30,720 --> 00:01:33,480
但是馬斯克的伺服器機櫃

50
00:01:33,480 --> 00:01:35,760
是要給聯發科跟他一起合作的

51
00:01:35,760 --> 00:01:37,180
所以市場現在看好是

52
00:01:37,180 --> 00:01:38,940
他未來的訂單會越好

53
00:01:38,940 --> 00:01:40,420
所以他把他X個上修

54
00:01:40,420 --> 00:01:43,200
因為他提供的是系統級的服務

55
00:01:43,200 --> 00:01:44,660
可是我們回來講

56
00:01:44,660 --> 00:01:46,720
那美國現在的所謂X可龍頭

57
00:01:46,720 --> 00:01:47,620
其實是博通

58
00:01:47,620 --> 00:01:48,680
博通做的是

59
00:01:48,680 --> 00:01:50,100
反而他說什麼

60
00:01:50,100 --> 00:01:52,300
我不要提供供應所謂的

61
00:01:52,300 --> 00:01:53,720
整合系統服務

62
00:01:53,720 --> 00:01:55,560
我要回歸只做

63
00:01:55,560 --> 00:01:56,620
晶片业务

64
00:01:56,620 --> 00:01:58,280
那这两个其实产生了一个

65
00:01:58,280 --> 00:01:59,600
很悬殊的道路

66
00:01:59,600 --> 00:02:01,260
那当然先来讲博通

67
00:02:01,260 --> 00:02:03,600
博通就去想一件事情是

68
00:02:03,600 --> 00:02:05,100
我去做这些系统

69
00:02:05,100 --> 00:02:06,840
其实我会拉低我的毛利

70
00:02:06,840 --> 00:02:08,200
我好好的专心

71
00:02:08,200 --> 00:02:10,140
做好我的纯晶片服务的话

72
00:02:10,140 --> 00:02:12,240
我的毛利可以锁定在65%

73
00:02:12,240 --> 00:02:13,300
所以大家会觉得说

74
00:02:13,300 --> 00:02:15,180
那我这样还是很赚钱

75
00:02:15,180 --> 00:02:17,720
所以我投资价值不会被减下来

76
00:02:17,720 --> 00:02:19,040
可是对联发科来讲

77
00:02:19,040 --> 00:02:20,820
如果我去做基板的话

78
00:02:20,820 --> 00:02:23,200
那我可能相对的毛利

79
00:02:23,200 --> 00:02:24,300
或机柜设计

80
00:02:24,300 --> 00:02:26,240
我相对的毛利就会比较低

81
00:02:26,240 --> 00:02:28,080
那大家怎么去看联发科呢

82
00:02:28,080 --> 00:02:29,620
就可能会产生不一样的

83
00:02:29,620 --> 00:02:31,380
可是这两条路选择

84
00:02:31,380 --> 00:02:34,140
我们再看联发科到底在想什么

85
00:02:34,140 --> 00:02:36,920
当然我们这里还必须回应一件事情是

86
00:02:36,920 --> 00:02:39,320
这些新买家以及旧买家

87
00:02:39,320 --> 00:02:42,180
他们过去都有体验一件事情是

88
00:02:42,180 --> 00:02:43,640
他们缴了很多东西

89
00:02:43,640 --> 00:02:46,160
你知道叫做辉达税跟博通税

90
00:02:46,160 --> 00:02:48,920
也就是说我本来在跟你买晶片的时候

91
00:02:48,920 --> 00:02:50,320
像Google在买晶片的时候

92
00:02:50,320 --> 00:02:52,340
我还必须跟博通跟辉达

93
00:02:52,340 --> 00:02:55,440
绑定相关系统的产品

94
00:02:55,440 --> 00:02:56,520
所以以至于

95
00:02:56,520 --> 00:02:58,880
他付了相对比较多的钱

96
00:02:58,880 --> 00:02:59,780
这采购硬体

97
00:02:59,780 --> 00:03:01,100
那这个对于他们讲

98
00:03:01,100 --> 00:03:03,040
叫做博通税 挥达税

99
00:03:03,040 --> 00:03:04,480
在Google这些云章业者

100
00:03:04,480 --> 00:03:05,480
都体验到了

101
00:03:05,480 --> 00:03:07,380
那联发科就想到

102
00:03:07,380 --> 00:03:08,800
那我要解决这个问题

103
00:03:08,800 --> 00:03:10,540
Google他们就解决的问题就是

104
00:03:10,540 --> 00:03:14,080
那有人是不是能够帮我

105
00:03:14,080 --> 00:03:17,080
把基板的采购跟伺服器的采购

106
00:03:17,080 --> 00:03:19,280
好好的搞明白 说清楚

107
00:03:19,280 --> 00:03:20,540
所以对于来讲

108
00:03:20,540 --> 00:03:23,420
联发科就可以把所谓的

109
00:03:23,420 --> 00:03:25,920
系统效能极大化

110
00:03:25,920 --> 00:03:27,560
以及客户的需求

111
00:03:27,560 --> 00:03:29,260
给提高到极致

112
00:03:29,260 --> 00:03:30,340
所以对于Google

113
00:03:30,340 --> 00:03:31,440
这些买家来讲

114
00:03:31,440 --> 00:03:32,900
我总算有人来

115
00:03:32,900 --> 00:03:34,220
好好的帮我搞清楚

116
00:03:34,220 --> 00:03:35,940
硬体这一端的事情

117
00:03:35,940 --> 00:03:37,340
跟整个机柜的事情

118
00:03:37,340 --> 00:03:39,040
好 那对于联发科来讲

119
00:03:39,040 --> 00:03:41,820
这问题好处是什么

120
00:03:41,820 --> 00:03:43,720
那我可以学会一件事情是

121
00:03:43,720 --> 00:03:46,460
我会学会挥打的模式

122
00:03:46,460 --> 00:03:47,860
什么叫挥打模式是

123
00:03:47,860 --> 00:03:48,940
我不只挥打

124
00:03:48,940 --> 00:03:49,940
只会做晶片

125
00:03:49,940 --> 00:03:54,380
我还会做所谓的NB系统的设计

126
00:03:54,380 --> 00:03:55,740
那我就会懂得是

127
00:03:55,740 --> 00:03:58,380
整个系统是怎么设计出来的

128
00:03:58,380 --> 00:04:00,140
他就会树立他的差异化

129
00:04:00,140 --> 00:04:02,700
虽然他会牺牲毛利

130
00:04:02,700 --> 00:04:04,400
但是差异化就会出来了

131
00:04:04,400 --> 00:04:06,720
整个研发科他会变成什么

132
00:04:06,720 --> 00:04:08,220
他会不像博通

133
00:04:08,220 --> 00:04:10,340
而是他会越来越像挥达

134
00:04:10,340 --> 00:04:12,600
就是往挥达那条路去走

135
00:04:12,600 --> 00:04:15,100
那有可能他会牺牲毛利

136
00:04:15,100 --> 00:04:16,880
那是他可以提高市占

137
00:04:16,880 --> 00:04:17,920
那这里回来

138
00:04:17,920 --> 00:04:19,620
为什么他可以提高市占的问题

139
00:04:19,620 --> 00:04:21,420
联花科为什么敢说是

140
00:04:21,420 --> 00:04:22,680
未来市场越来越大

141
00:04:22,680 --> 00:04:24,400
除了刚才我们讲的

142
00:04:24,400 --> 00:04:25,840
所谓的Google这些人

143
00:04:25,840 --> 00:04:26,600
Meta这些人

144
00:04:26,600 --> 00:04:28,140
或者是马斯克特斯拉

145
00:04:28,140 --> 00:04:31,100
他们因为所谓的系统机柜

146
00:04:31,100 --> 00:04:32,380
付了都付很多钱

147
00:04:32,380 --> 00:04:34,160
所以需要有人来帮他设计

148
00:04:34,160 --> 00:04:36,180
晶片加整套服务之外

149
00:04:36,180 --> 00:04:38,940
其实还有很多新的买家

150
00:04:38,940 --> 00:04:40,100
正在崛起

151
00:04:40,100 --> 00:04:42,000
包括OpenAI跟Enthopi

152
00:04:42,000 --> 00:04:43,740
像Enthopi最近也宣布说

153
00:04:43,740 --> 00:04:46,180
他也要跨入所谓的

154
00:04:46,180 --> 00:04:47,780
内系设计

155
00:04:47,780 --> 00:04:49,040
他也要自己做晶片

156
00:04:49,040 --> 00:04:50,740
那这里必须要解释一件事情

157
00:04:50,740 --> 00:04:51,840
要设计晶片

158
00:04:51,840 --> 00:04:53,400
没有那么容易

159
00:04:53,400 --> 00:04:56,780
现在进入了所谓的7nm之后

160
00:04:56,780 --> 00:04:58,440
要能够设计晶片

161
00:04:58,440 --> 00:04:59,660
其实全球500大

162
00:04:59,660 --> 00:05:01,000
放眼全球500大

163
00:05:01,000 --> 00:05:02,320
每几家公司

164
00:05:02,320 --> 00:05:03,700
愿有那么大的资源

165
00:05:03,700 --> 00:05:04,880
去真的设计出晶片

166
00:05:04,880 --> 00:05:06,000
所以还是交给专业的

167
00:05:06,000 --> 00:05:08,140
当然你可以有核心团队

168
00:05:08,140 --> 00:05:09,880
但是后面的怎么导入

169
00:05:09,880 --> 00:05:11,760
所谓的台积电的制程

170
00:05:11,760 --> 00:05:13,660
Intel的制程

171
00:05:13,660 --> 00:05:14,580
三星的制程

172
00:05:14,580 --> 00:05:16,400
还是需要有人协助

173
00:05:16,400 --> 00:05:17,360
所以相对的

174
00:05:17,360 --> 00:05:18,540
就叫ASC的服务

175
00:05:18,540 --> 00:05:21,160
所以新的客户会越来越多

176
00:05:21,160 --> 00:05:22,360
甚至算力提供业者

177
00:05:22,360 --> 00:05:23,960
也想要跨入这个市场

178
00:05:23,960 --> 00:05:25,180
可是对于联发科来讲

179
00:05:25,180 --> 00:05:27,920
大家也不用说期待这块会很大

180
00:05:27,920 --> 00:05:30,280
因为这个业务是相对会比较长

181
00:05:30,280 --> 00:05:32,140
所以真正的系统的业务

182
00:05:32,140 --> 00:05:34,060
其实要到两年后

183
00:05:34,060 --> 00:05:36,380
所以短暂你要看到说

184
00:05:36,380 --> 00:05:38,560
马上的数字也不可能

185
00:05:38,560 --> 00:05:39,160
不太可能

186
00:05:39,160 --> 00:05:41,440
所以相对的ESC这个顺利量产

187
00:05:41,440 --> 00:05:42,720
其实才是联发科真正

188
00:05:42,720 --> 00:05:44,420
我们要关心他能够

189
00:05:44,420 --> 00:05:46,260
把这接晶片提供出来

190
00:05:46,260 --> 00:05:47,240
所以真正的成长

191
00:05:47,240 --> 00:05:48,680
还会是在ESC这一块

192
00:05:48,680 --> 00:05:50,980
那所以这个应该是一条正确的路

193
00:05:50,980 --> 00:05:52,980
对于联发科来讲

194
00:05:52,980 --> 00:05:54,240
还有台湾的供应链

195
00:05:54,240 --> 00:05:56,160
因为为什么博通不去选择这条路

196
00:05:56,160 --> 00:05:58,540
跟台湾基于两个地方不太一样

197
00:05:58,540 --> 00:05:59,780
台湾的优势是

198
00:05:59,780 --> 00:06:02,660
相关的系统业者都在台湾

199
00:06:02,660 --> 00:06:05,400
所以联发科才有办法做这件事情

200
00:06:05,400 --> 00:06:06,500
博通没那么简单

201
00:06:06,500 --> 00:06:08,280
它的优势比较明显

202
00:06:08,280 --> 00:06:09,680
优势会比较明显

203
00:06:09,680 --> 00:06:10,600
当然博通也有

204
00:06:10,600 --> 00:06:11,400
当然博通必须

205
00:06:11,400 --> 00:06:13,500
它想要的是坚守这个利润

206
00:06:13,500 --> 00:06:14,820
可是联发科相对的

207
00:06:14,820 --> 00:06:16,840
在整个供应链都在台湾的时候

208
00:06:16,840 --> 00:06:18,740
相對它的優勢就會比較明顯

209
00:06:18,740 --> 00:06:20,000
好 非常謝謝國登哥

210
00:06:20,000 --> 00:06:20,540
但我們看到

211
00:06:20,540 --> 00:06:22,020
這是聯發科做出了一條

212
00:06:22,020 --> 00:06:23,520
跟博通不一樣的選擇

213
00:06:23,520 --> 00:06:25,660
那當然我們有我們的一個利基點

214
00:06:25,660 --> 00:06:26,780
那聯發科這條路

215
00:06:26,780 --> 00:06:28,160
會不會走得很順暢呢

216
00:06:28,160 --> 00:06:29,840
就讓我們持續來觀察它

217
00:06:29,840 --> 00:06:31,540
我們要帶大家來看看了

218
00:06:31,540 --> 00:06:33,060
這個最近的一個大消息

219
00:06:33,060 --> 00:06:34,620
就是台積電最近

220
00:06:34,620 --> 00:06:35,520
有一點壓力

221
00:06:35,520 --> 00:06:36,620
是因為大家發現

222
00:06:36,620 --> 00:06:39,220
英特爾開始往前追了

223
00:06:39,220 --> 00:06:41,540
我們來看到說他威脅台積電

224
00:06:41,540 --> 00:06:43,300
當然了 打一個大問號了

225
00:06:43,300 --> 00:06:45,500
但是人家的良率在提升

226
00:06:45,500 --> 00:06:47,100
这也是不争的事实

227
00:06:47,100 --> 00:06:49,620
好我们来看先进封装当中

228
00:06:49,620 --> 00:06:51,700
英特尔的良率据说

229
00:06:51,700 --> 00:06:54,680
拉到了将近九成

230
00:06:54,680 --> 00:06:56,720
好那它拉到九成

231
00:06:56,720 --> 00:06:58,080
还有一个就是

232
00:06:58,080 --> 00:07:01,360
他们说台积电非常罕见的

233
00:07:01,360 --> 00:07:04,140
要跟着英特尔的屁股后面

234
00:07:04,140 --> 00:07:06,800
也要做他们的同样的技术

235
00:07:06,800 --> 00:07:10,060
要开发类EMIB的一个专案

236
00:07:10,060 --> 00:07:11,540
是拉着紧硕一起

237
00:07:11,540 --> 00:07:13,120
这又代表什么意思

238
00:07:13,120 --> 00:07:14,600
是英特尔做对了

239
00:07:14,600 --> 00:07:16,060
所以台積電要學嗎

240
00:07:16,060 --> 00:07:17,980
那英特爾現在良率拉升了

241
00:07:17,980 --> 00:07:19,460
真的會威脅到台積電嗎

242
00:07:19,460 --> 00:07:20,500
請教一下石聰哥

243
00:07:20,500 --> 00:07:21,900
對 我覺得這個EMIB

244
00:07:21,900 --> 00:07:23,160
這個短時間之內

245
00:07:23,160 --> 00:07:24,800
不會對台積電造成威脅

246
00:07:24,800 --> 00:07:26,020
但是長遠的來看

247
00:07:26,020 --> 00:07:26,980
的確是有可能

248
00:07:26,980 --> 00:07:27,960
所以為什麼台積電

249
00:07:27,960 --> 00:07:31,160
它也要發展類似這樣的一個產品

250
00:07:31,160 --> 00:07:32,280
那我先跟它講

251
00:07:32,280 --> 00:07:34,000
主要原因是因為最近傳出來說

252
00:07:34,000 --> 00:07:36,220
英特爾的EMIB-T的良率

253
00:07:36,220 --> 00:07:37,900
已經來到90%

254
00:07:37,900 --> 00:07:38,920
那90%來說

255
00:07:38,920 --> 00:07:41,280
其實跟台積電的CoOS的封裝

256
00:07:41,280 --> 00:07:42,240
是差不多的

257
00:07:42,240 --> 00:07:44,520
所以如果真的假設是這樣子的話

258
00:07:44,520 --> 00:07:47,180
我先說是假設是這個樣子的話

259
00:07:47,180 --> 00:07:49,880
那的確會對台積電造成很大的壓力

260
00:07:49,880 --> 00:07:50,760
為什麼這樣講

261
00:07:50,760 --> 00:07:52,200
我先跟大家稍微解釋一下

262
00:07:52,200 --> 00:07:53,100
什麼叫做CoA

263
00:07:53,100 --> 00:07:54,360
什麼叫EMIB

264
00:07:54,360 --> 00:07:56,300
那我們知道事實上現在

265
00:07:56,300 --> 00:07:58,360
我們不是在一個晶片上面有

266
00:07:58,360 --> 00:07:59,800
包括說像CPU GPU

267
00:08:00,000 --> 00:08:00,840
HBM等等嗎

268
00:08:00,840 --> 00:08:03,600
然後把它完全的組裝在一起嗎

269
00:08:03,600 --> 00:08:05,400
你要封裝到下面的

270
00:08:05,400 --> 00:08:06,800
這個PCB上面的時候

271
00:08:06,800 --> 00:08:08,800
你要用一個中介層

272
00:08:08,800 --> 00:08:10,620
中介層下面是基板

273
00:08:10,620 --> 00:08:11,760
就是ABF基板

274
00:08:11,760 --> 00:08:14,700
但是你ABF跟這個晶片中間

275
00:08:14,700 --> 00:08:16,240
還有一個叫做這個COWAS

276
00:08:16,240 --> 00:08:17,700
這個系中介層

277
00:08:17,700 --> 00:08:19,700
COWAS就是所謂的系中介層

278
00:08:19,700 --> 00:08:22,400
當成是這個連接的這個過程

279
00:08:22,400 --> 00:08:24,500
那因為它也是要用系的來

280
00:08:24,500 --> 00:08:25,240
這個製程

281
00:08:25,240 --> 00:08:27,340
所以我們知道系是很貴的東西

282
00:08:27,340 --> 00:08:28,840
所以它就是成本比較貴

283
00:08:28,840 --> 00:08:30,340
那系橋是怎麼樣

284
00:08:30,340 --> 00:08:31,160
它把它變成是說

285
00:08:31,160 --> 00:08:33,760
它在這個ABF裡面去挖這個橋

286
00:08:33,760 --> 00:08:35,260
挖那種這個橋

287
00:08:35,260 --> 00:08:36,760
這個我等於是晶片跟晶片

288
00:08:36,760 --> 00:08:38,500
之後我就用這個類似一個系橋

289
00:08:38,500 --> 00:08:40,900
用橋的方式把它接起來的一個狀況

290
00:08:40,900 --> 00:08:41,860
我就不要一整片

291
00:08:41,860 --> 00:08:42,760
對 我不用一整片

292
00:08:42,760 --> 00:08:44,360
我就省一些錢起來

293
00:08:44,360 --> 00:08:45,600
對 所以理論上

294
00:08:45,600 --> 00:08:47,700
它本來就會比這個太極電的

295
00:08:47,700 --> 00:08:48,720
因為它也要用矽

296
00:08:48,720 --> 00:08:51,300
我們知道這個矽條是誰在做

297
00:08:51,300 --> 00:08:51,900
聯電

298
00:08:51,900 --> 00:08:54,000
它也是要用晶圓代工的矽製程

299
00:08:54,000 --> 00:08:55,720
但是它的面積就小很多

300
00:08:55,720 --> 00:08:56,980
所以它直觀上

301
00:08:56,980 --> 00:08:58,140
它絕對會是比較

302
00:08:58,140 --> 00:08:59,380
成本比較低的

303
00:08:59,380 --> 00:09:00,980
但它會有一個問題是這樣子

304
00:09:00,980 --> 00:09:03,680
你在這個很多晶片要組合的時候

305
00:09:03,680 --> 00:09:05,480
你就要挖很多矽條

306
00:09:05,480 --> 00:09:07,060
這個難度會很高

307
00:09:07,060 --> 00:09:08,620
所以現在在大晶片

308
00:09:08,620 --> 00:09:10,760
或是說你要這個傳輸速度快的時候

309
00:09:10,760 --> 00:09:13,440
可能還是要用這個所謂COVAS

310
00:09:13,440 --> 00:09:16,460
但是EMIB如果这个技术越来越成熟

311
00:09:16,460 --> 00:09:18,440
比如说挖的越来越成熟的时候

312
00:09:18,440 --> 00:09:21,900
这可能的确会造成相关的影响

313
00:09:21,900 --> 00:09:24,000
尤其是如果我只要某些晶片

314
00:09:24,000 --> 00:09:25,720
跟某些晶片连接在一起的时候

315
00:09:25,720 --> 00:09:27,220
它的灵活度会更好

316
00:09:27,220 --> 00:09:28,920
那我可以省更多成分

317
00:09:28,920 --> 00:09:30,680
所以现在包括说像Google

318
00:09:30,680 --> 00:09:32,480
或者说像这个联发科

319
00:09:32,480 --> 00:09:33,880
甚至像这个所谓Meta

320
00:09:33,880 --> 00:09:34,780
为什么他们要采用

321
00:09:34,780 --> 00:09:36,760
因为我不需要用到像这个

322
00:09:36,760 --> 00:09:39,460
挥打GPU这么高的这个算力的时候

323
00:09:39,460 --> 00:09:41,600
或许用这个EMIBT

324
00:09:41,600 --> 00:09:43,300
会是比较好的这个技术

325
00:09:43,300 --> 00:09:44,780
那当然因为这样来说

326
00:09:44,780 --> 00:09:45,760
台积电当然会紧张

327
00:09:45,760 --> 00:09:46,400
也就是说

328
00:09:46,400 --> 00:09:47,880
如果我要保有灵活度

329
00:09:47,880 --> 00:09:49,060
跟我的成本的时候

330
00:09:49,060 --> 00:09:50,100
我也要发展

331
00:09:50,100 --> 00:09:51,220
类似这样一个产品

332
00:09:51,220 --> 00:09:52,260
所以现在传言说

333
00:09:52,260 --> 00:09:53,060
他跟景硕

334
00:09:53,060 --> 00:09:54,520
为什么会跟景硕合作

335
00:09:54,520 --> 00:09:55,260
因为我刚才讲到

336
00:09:55,260 --> 00:09:58,360
他是要在ABF基板里面挖桥

337
00:09:58,360 --> 00:10:01,080
所以他一定要跟ABF合作

338
00:10:01,080 --> 00:10:02,200
原因就在这个地方

339
00:10:02,200 --> 00:10:02,900
那你看到说

340
00:10:02,900 --> 00:10:04,180
这个Intel跟谁合作

341
00:10:04,180 --> 00:10:06,440
Intel当然就是跟这个星星

342
00:10:06,440 --> 00:10:07,760
或是跟这个Ebiton

343
00:10:07,760 --> 00:10:08,780
来互相的合作

344
00:10:08,780 --> 00:10:10,220
所以对台积电来讲

345
00:10:10,220 --> 00:10:11,280
为什么他要发展

346
00:10:11,280 --> 00:10:13,680
这个类似EMIB的产品

347
00:10:13,680 --> 00:10:15,800
因为它目前为止的产能蛮宅

348
00:10:15,800 --> 00:10:16,980
它很怕说

349
00:10:16,980 --> 00:10:19,320
很多人会因为这个技术的便宜

350
00:10:19,320 --> 00:10:20,280
它就转到那边

351
00:10:20,280 --> 00:10:22,580
那我也有寄出相关的产品的话

352
00:10:22,580 --> 00:10:23,560
你未来就可能

353
00:10:23,560 --> 00:10:25,880
这个你要跳槽的可能性就比较低

354
00:10:25,880 --> 00:10:28,800
不过它现在EMIB遇到一个问题

355
00:10:28,800 --> 00:10:31,820
这是我们国内的这个栽班厂商

356
00:10:31,820 --> 00:10:33,880
或是日本厂商这个讲的

357
00:10:33,880 --> 00:10:34,740
就是说没有错

358
00:10:34,740 --> 00:10:36,620
传言说它的良率90%

359
00:10:36,620 --> 00:10:39,400
但是这个栽班厂商说还早

360
00:10:39,400 --> 00:10:41,040
我们国内的厂商就说

361
00:10:41,040 --> 00:10:42,780
這個良率目前要挖下去

362
00:10:42,780 --> 00:10:44,220
目前良率只有50%

363
00:10:44,220 --> 00:10:45,680
所以如果假設

364
00:10:45,680 --> 00:10:46,920
它良率只有50%的話

365
00:10:46,920 --> 00:10:48,340
你真的要大規模商產

366
00:10:48,340 --> 00:10:49,620
你說你90%

367
00:10:49,620 --> 00:10:51,440
但是基本上說是50%

368
00:10:51,440 --> 00:10:52,620
所以看起來我才說

369
00:10:52,620 --> 00:10:53,960
短時間之內可能

370
00:10:53,960 --> 00:10:55,920
很可能是實驗室90%

371
00:10:55,920 --> 00:10:57,900
對 或者說小批量OK

372
00:10:57,900 --> 00:10:59,440
但是你要真正大量量產

373
00:10:59,440 --> 00:11:00,580
可能有很大難度

374
00:11:00,580 --> 00:11:01,520
不過我們比如說

375
00:11:01,520 --> 00:11:03,700
EMIB來說對台灣有沒有好處

376
00:11:03,700 --> 00:11:04,240
當然有

377
00:11:04,240 --> 00:11:06,940
事實上除了這個所謂

378
00:11:06,940 --> 00:11:08,020
這個Intel之外

379
00:11:08,020 --> 00:11:09,900
因為它要找一些相關的供應鏈

380
00:11:09,900 --> 00:11:11,760
目前為止最重要功力有兩個

381
00:11:11,760 --> 00:11:13,180
一個就是我們一直講到

382
00:11:13,180 --> 00:11:14,180
你要挖細橋

383
00:11:14,180 --> 00:11:15,400
細橋就是由台積

384
00:11:15,400 --> 00:11:16,600
由聯電來做

385
00:11:16,600 --> 00:11:18,080
細橋原件這個部分

386
00:11:18,080 --> 00:11:19,580
另外一個部分是什麼

387
00:11:19,580 --> 00:11:20,540
細電容

388
00:11:20,540 --> 00:11:21,880
那細電容目前為止來說

389
00:11:21,880 --> 00:11:23,400
跟立基電互相的合作

390
00:11:23,400 --> 00:11:24,220
所以目前為止

391
00:11:24,220 --> 00:11:25,760
兩個可能是EMIB的

392
00:11:25,760 --> 00:11:27,620
還有聯發科要切入這個平台

393
00:11:27,620 --> 00:11:28,700
那另外一個就是

394
00:11:28,700 --> 00:11:30,520
我覺得這個會讓怎麼樣

395
00:11:30,520 --> 00:11:32,180
因為他們會排擠掉一些

396
00:11:32,180 --> 00:11:34,060
成熟製程的這個產能

397
00:11:34,060 --> 00:11:35,540
所以原本成熟製程的

398
00:11:35,540 --> 00:11:36,940
已經是很缺了

399
00:11:36,940 --> 00:11:39,540
結果他們會受惠到這個封裝

400
00:11:39,540 --> 00:11:40,700
他们也会产能更缺

401
00:11:40,700 --> 00:11:41,740
所以你看前一阵子

402
00:11:41,740 --> 00:11:43,420
包括说像世界先进

403
00:11:43,420 --> 00:11:46,100
联电 利基电都说什么

404
00:11:46,100 --> 00:11:47,960
我们现在的产能利用率

405
00:11:47,960 --> 00:11:49,060
已经百分之90了

406
00:11:49,060 --> 00:11:50,960
那明年的产能会增加

407
00:11:50,960 --> 00:11:52,660
甚至明年还要再涨价

408
00:11:52,660 --> 00:11:54,640
所以你看毛利率以前

409
00:11:54,640 --> 00:11:56,580
如果以这个世界先进

410
00:11:56,580 --> 00:11:57,300
或联电来说

411
00:11:57,300 --> 00:11:59,360
30就很偷笑了

412
00:11:59,360 --> 00:12:00,960
现在毛利率可以挑战到

413
00:12:00,960 --> 00:12:02,280
45到50%

414
00:12:02,280 --> 00:12:03,880
所以完全是不一样了

415
00:12:03,880 --> 00:12:05,400
那尤其是最近刚才

416
00:12:05,400 --> 00:12:06,740
大家都会讨论到利基电

417
00:12:06,740 --> 00:12:08,300
因为黄冲任董事长

418
00:12:08,300 --> 00:12:09,400
不幸的往生了

419
00:12:09,400 --> 00:12:13,120
但是現在按照新團隊的說法是說

420
00:12:13,120 --> 00:12:15,300
明年我們一定會配息

421
00:12:15,300 --> 00:12:17,400
既然是黃聰能董事長的遺願

422
00:12:17,400 --> 00:12:18,400
我們一定會這樣做

423
00:12:18,400 --> 00:12:20,100
經營團隊不變

424
00:12:20,100 --> 00:12:24,600
賣給美光的可以500億的現金

425
00:12:24,600 --> 00:12:26,180
然後包括說跟美光合作

426
00:12:26,180 --> 00:12:26,960
HBM等等

427
00:12:26,960 --> 00:12:28,700
所以目前為止外資

428
00:12:28,700 --> 00:12:31,000
也把目標價上修到100塊

429
00:12:31,000 --> 00:12:32,520
所以我覺得其實這個部分

430
00:12:32,520 --> 00:12:33,980
就EMIB跟成熟製程

431
00:12:33,980 --> 00:12:35,440
我們可以一併合併來看

432
00:12:35,440 --> 00:12:36,620
裡面來講話

433
00:12:36,620 --> 00:12:38,180
當然利基電

434
00:12:38,180 --> 00:12:39,480
那你今天來說的話

435
00:12:39,480 --> 00:12:40,880
當然我覺得新經營團隊

436
00:12:40,880 --> 00:12:44,080
回到所謂專業經理人的部分

437
00:12:44,080 --> 00:12:46,340
那我覺得當然後續我們很期待說

438
00:12:46,340 --> 00:12:48,100
他能夠慢慢的走出

439
00:12:48,100 --> 00:12:50,200
黃松東董事長過市的壓力

440
00:12:50,200 --> 00:12:51,540
目前為止來說

441
00:12:51,540 --> 00:12:52,860
股價還沒有回到季線

442
00:12:52,860 --> 00:12:53,800
紅色是季線

443
00:12:53,800 --> 00:12:55,680
我覺得應該有機會回到季線之後

444
00:12:55,680 --> 00:12:57,900
站穩應該還有機會再往上挑戰

445
00:12:57,900 --> 00:12:59,140
那聯電來說的話

446
00:12:59,140 --> 00:13:00,900
剛才杜達士已經講了

447
00:13:00,900 --> 00:13:03,140
我覺得其實聯電在往下跌

448
00:13:03,140 --> 00:13:06,080
他今年EPS只好7塊到8塊

449
00:13:06,080 --> 00:13:07,600
你說百元左右

450
00:13:07,600 --> 00:13:08,980
我覺得其實是相對

451
00:13:08,980 --> 00:13:11,120
進入一個投資的價值

452
00:13:11,120 --> 00:13:12,380
另外一個就是

453
00:13:12,380 --> 00:13:13,920
剛才杜納斯有講到的

454
00:13:13,920 --> 00:13:15,580
ABF載板這兩塊

455
00:13:15,580 --> 00:13:16,880
這兩塊來講的話

456
00:13:16,880 --> 00:13:18,220
你如果去看他們的

457
00:13:18,220 --> 00:13:20,680
第二季的獲利跟毛利

458
00:13:20,680 --> 00:13:22,440
其實升得最快的

459
00:13:22,440 --> 00:13:24,180
就是景碩跟新興兩檔

460
00:13:24,180 --> 00:13:26,480
尤其是他們接下來來說

461
00:13:26,480 --> 00:13:29,140
景碩獲得台積電的合作

462
00:13:29,140 --> 00:13:31,140
IMIT的EMIB要跟他合作

463
00:13:31,140 --> 00:13:33,280
新興本來就有跟台積電

464
00:13:33,280 --> 00:13:35,620
有在合作Coast的產品

465
00:13:35,620 --> 00:13:37,960
那又有在跟Intel在合作

466
00:13:37,960 --> 00:13:40,080
所以这两家公司都是可以留意

467
00:13:40,080 --> 00:13:41,980
只是说他们的问题是什么

468
00:13:41,980 --> 00:13:44,220
你看他们短期涨幅都很大

469
00:13:44,220 --> 00:13:46,180
他们都已经谈过了季限

470
00:13:46,180 --> 00:13:47,960
而且还谈了有一段幅度

471
00:13:47,960 --> 00:13:50,380
甚至紧缩距离历史新高都不远

472
00:13:50,380 --> 00:13:51,700
那这种股票我们只能够

473
00:13:51,700 --> 00:13:53,420
拉回你再来考虑

474
00:13:53,420 --> 00:13:54,560
你千万不要去追高

475
00:13:54,560 --> 00:13:55,700
就是他等谈

476
00:13:55,700 --> 00:13:57,500
他这阵子谈得非常的高

477
00:13:57,500 --> 00:13:58,740
谈得非常高的时候表示什么

478
00:13:58,740 --> 00:14:00,140
他有机会再创新高

479
00:14:00,140 --> 00:14:01,720
但是你千万不要去追高

480
00:14:01,720 --> 00:14:03,060
等到拉回的时候

481
00:14:03,060 --> 00:14:04,480
再來逢低布局

482
00:14:04,480 --> 00:14:06,700
這個景碩跟信心這兩家公司

483
00:14:06,700 --> 00:14:07,880
好 所以我們看到了

484
00:14:07,880 --> 00:14:09,840
這個是先進封裝的一個

485
00:14:09,840 --> 00:14:11,260
這個技術的一個

486
00:14:11,260 --> 00:14:12,940
這個拉扯競爭

487
00:14:12,940 --> 00:14:14,600
當然這個台積電

488
00:14:14,600 --> 00:14:15,440
這個無庸置疑

489
00:14:15,440 --> 00:14:16,680
現在跑得最前面了

490
00:14:16,680 --> 00:14:18,360
那現在英特爾在往後追

491
00:14:18,360 --> 00:14:20,160
那帶出來的一些相關技術

492
00:14:20,160 --> 00:14:21,220
給大家做參考