1
00:00:00,200 --> 00:00:01,866
It's only been less than three months

2
00:00:01,866 --> 00:00:05,333
The wafer factory "Terafab" that claims to crush TSMC

3
00:00:05,333 --> 00:00:06,566
Why did it turn off?

4
00:00:06,566 --> 00:00:08,866
Musk personally poured cold water on the front foot

5
00:00:08,866 --> 00:00:10,866
Its technical backing is Intel

6
00:00:10,866 --> 00:00:13,066
I even went to TSMC to grab production capacity.

7
00:00:13,066 --> 00:00:14,200
even

8
00:00:14,200 --> 00:00:18,266
Musk’s own Tesla is ready to exit at any time

9
00:00:18,266 --> 00:00:20,033
It's obviously a strong alliance

10
00:00:20,100 --> 00:00:22,833
Why can’t we build enough AI chips?

11
00:00:22,833 --> 00:00:24,633
Friends, just recently

12
00:00:24,633 --> 00:00:26,766
Two very strange signals appeared

13
00:00:26,766 --> 00:00:28,600
One happened last week

14
00:00:28,733 --> 00:00:32,266
In SpaceX’s S-1 listing application document submitted to the SEC

15
00:00:32,366 --> 00:00:34,733
Musk confesses to investors around the world

16
00:00:34,733 --> 00:00:37,100
"I may not be able to handle the chip supply."

17
00:00:37,100 --> 00:00:38,433
The original words are as follows:

18
00:00:38,433 --> 00:00:42,700
“We hope to solve the chip supply problem by building Terafab

19
00:00:42,833 --> 00:00:44,933
But the project could still fail."

20
00:00:44,933 --> 00:00:48,633
“We cannot guarantee that Terafab will achieve its targets within the expected cycle

21
00:00:48,733 --> 00:00:51,066
There is even no guarantee that the project will be implemented.”

22
00:00:51,066 --> 00:00:52,200
What’s even more worrying is

23
00:00:52,233 --> 00:00:54,266
Another sentence was added in another document:

24
00:00:54,266 --> 00:00:57,066
“Both Intel and Tesla may exit the project”

25
00:00:57,066 --> 00:00:59,633
Once these two core partners withdraw their investment and leave the market,

26
00:00:59,800 --> 00:01:03,500
Terafab will lose both key customers and process technology support

27
00:01:03,600 --> 00:01:06,133
Then wouldn’t this project be left empty-handed?

28
00:01:06,133 --> 00:01:08,600
But it's not over yet, just a few days later

29
00:01:08,733 --> 00:01:11,833
June 1, at the COMPUTEX 2026 venue

30
00:01:11,833 --> 00:01:13,666
Another person who shouldn't have given up.

31
00:01:13,666 --> 00:01:15,266
Intel CEO Chen Liwu

32
00:01:15,266 --> 00:01:18,400
He said something that all analysts did not expect:

33
00:01:18,400 --> 00:01:20,800
“Intel’s transformation will take 5 to 10 years

34
00:01:20,800 --> 00:01:23,533
We are very happy to become TSMC’s top customers.”

35
00:01:23,533 --> 00:01:24,433
You heard it right

36
00:01:24,433 --> 00:01:27,666
The one who once described "design by yourself, manufacture by yourself, and define standards by yourself"

37
00:01:27,666 --> 00:01:29,133
Intel written in the bones

38
00:01:29,133 --> 00:01:33,900
Intel, which was still fiercely competing with TSMC with 18A not long ago

39
00:01:33,900 --> 00:01:37,766
Now it has publicly admitted that it will hand over some key production to TSMC

40
00:01:37,900 --> 00:01:41,000
He even talked about the past friendship with TSMC.

41
00:01:41,000 --> 00:01:42,633
Why would he do this?

42
00:01:42,633 --> 00:01:43,700
Crazy?

43
00:01:43,700 --> 00:01:45,200
Still awake?

44
00:01:45,200 --> 00:01:47,666
Hello everyone, I’m Olivia, the last two episodes

45
00:01:47,666 --> 00:01:50,400
We talked about the 150 billion that Jen-Hsun Huang spent

46
00:01:50,400 --> 00:01:52,400
And the 10 billion that Su Zifeng flew over

47
00:01:52,400 --> 00:01:56,266
They are all doing the same thing - desperately grabbing TSMC's production capacity

48
00:01:56,266 --> 00:01:57,100
And now

49
00:01:57,100 --> 00:01:58,300
Chen Liwu is also here

50
00:01:58,300 --> 00:02:01,633
Musk, who wanted to bypass TSMC, also bowed his head.

51
00:02:01,633 --> 00:02:02,333
today

52
00:02:02,333 --> 00:02:04,933
Let's put in this most important piece of the puzzle

53
00:02:04,933 --> 00:02:07,133
Clear up the whole picture of this war

54
00:02:07,133 --> 00:02:09,166
Why Terafab might be failing

55
00:02:09,166 --> 00:02:10,966
Why did Chen Liwu admit defeat?

56
00:02:10,966 --> 00:02:13,566
Why are these the most important giants in the AI era?

57
00:02:13,566 --> 00:02:15,333
Have you been running to TSMC recently?

58
00:02:15,333 --> 00:02:16,400
What's more important

59
00:02:16,433 --> 00:02:18,366
In this silent transfer of power

60
00:02:18,366 --> 00:02:20,700
As investors, what should we see?

61
00:02:20,700 --> 00:02:22,466
This is not an isolated incident

62
00:02:22,466 --> 00:02:25,666
Rather, this round of AI war is revaluing the market.

63
00:02:25,666 --> 00:02:27,700
Upstream materials and advanced processes

64
00:02:27,700 --> 00:02:31,133
Advanced packaging, optical communications, storage, power...

65
00:02:31,133 --> 00:02:32,500
All on this line

66
00:02:32,500 --> 00:02:34,666
In fact, some of the companies we have talked about before

67
00:02:34,666 --> 00:02:36,800
Everyone is free riding in this round of carnival

68
00:02:36,800 --> 00:02:39,266
Thank you all for your support and companionship all the way

69
00:02:39,266 --> 00:02:43,500
I bet Huang Renxun, Su Zifeng and Chen Liwu together this time

70
00:02:43,500 --> 00:02:47,033
Complete list of industry chains, as well as the three most valuable companies

71
00:02:47,033 --> 00:02:51,700
From core logic, to financial report focus, to market differences, key catalysts

72
00:02:51,700 --> 00:02:53,466
Organized into a complete list

73
00:02:53,466 --> 00:02:56,300
Friends who want it can type 888 in the comment area

74
00:02:56,300 --> 00:02:58,300
If there are many people leaving messages

75
00:02:58,300 --> 00:03:02,733
I will spend two days preparing it and send it to friends who really need it.

76
00:03:02,733 --> 00:03:04,466
Let’s look at Intel first

77
00:03:04,500 --> 00:03:06,766
Why don’t you suddenly insist on making your own?

78
00:03:06,766 --> 00:03:09,300
Instead, why should we go to TSMC to grab production capacity?

79
00:03:09,300 --> 00:03:11,566
Many friends feel that it is quite abnormal, right?

80
00:03:11,566 --> 00:03:14,466
In fact, it was forced into a corner by time

81
00:03:14,466 --> 00:03:16,900
It’s not that Chen Liwu doesn’t know that his own manufacturing is important

82
00:03:17,000 --> 00:03:19,500
On the contrary, he knows better than anyone else

83
00:03:19,533 --> 00:03:23,633
18A and 14A are the keys to rebuilding his chip manufacturing empire.

84
00:03:23,633 --> 00:03:25,233
But the problem is

85
00:03:25,333 --> 00:03:28,300
Advanced processes are not about shouting "I'm coming back"

86
00:03:28,300 --> 00:03:30,666
The yield rate can immediately return to the king level.

87
00:03:30,800 --> 00:03:33,833
Although Intel 18A has shown signs of yield improvement

88
00:03:33,833 --> 00:03:37,233
But it's still in the critical stages of ramp-up and customer validation

89
00:03:37,333 --> 00:03:40,300
14A is the next step for heavy investment.

90
00:03:40,300 --> 00:03:41,266
In other words

91
00:03:41,300 --> 00:03:43,633
Intel's manufacturing renaissance is on the way

92
00:03:43,633 --> 00:03:45,500
But not quite there yet

93
00:03:45,500 --> 00:03:47,700
But the problem is, time waits for no one

94
00:03:47,700 --> 00:03:48,533
other side

95
00:03:48,533 --> 00:03:50,733
The CPU renaissance has arrived

96
00:03:50,733 --> 00:03:53,600
The AI PC war has also reached our doorstep.

97
00:03:53,633 --> 00:03:56,100
AI is moving from training to inference

98
00:03:56,133 --> 00:03:58,666
From chat box to Agentic AI

99
00:03:58,866 --> 00:04:00,333
Large models from the cloud

100
00:04:00,400 --> 00:04:03,900
Towards enterprise systems, edge devices and AI PCs

101
00:04:03,900 --> 00:04:06,500
Once the AI Agent starts doing real work

102
00:04:06,500 --> 00:04:08,466
It's not just a simple answer

103
00:04:08,466 --> 00:04:11,566
It’s about splitting tasks, adjusting tools, and reading files.

104
00:04:11,566 --> 00:04:15,700
Run code, access databases, manage permissions, and handle network communications

105
00:04:15,700 --> 00:04:18,566
Also coordinate multiple models and multiple applications simultaneously

106
00:04:18,566 --> 00:04:21,333
The GPU cannot do these tasks alone.

107
00:04:21,666 --> 00:04:24,233
GPU is good at large-scale matrix calculations

108
00:04:24,233 --> 00:04:25,466
It’s violent computing power

109
00:04:25,466 --> 00:04:29,266
But what the CPU is good at is system scheduling, complex logic, and I/O

110
00:04:29,266 --> 00:04:33,033
Memory management, network communications and enterprise software ecosystem

111
00:04:33,033 --> 00:04:35,333
In other words, the era of AI training

112
00:04:35,333 --> 00:04:37,100
GPU is the absolute protagonist

113
00:04:37,100 --> 00:04:40,033
But when AI Agent comes to large-scale implementation,

114
00:04:40,200 --> 00:04:42,966
The CPU has moved from the background to the front desk again

115
00:04:42,966 --> 00:04:43,700
And this

116
00:04:43,700 --> 00:04:45,266
It's Intel's domain

117
00:04:45,266 --> 00:04:47,266
This is not me trying to smooth things over for Intel.

118
00:04:47,266 --> 00:04:49,666
But the data has begun to reflect it

119
00:04:49,666 --> 00:04:52,666
Intel's first quarter 2026 financial report

120
00:04:52,666 --> 00:04:54,333
Client Computing Group

121
00:04:54,466 --> 00:04:57,100
That is the client business where the PC chip is located

122
00:04:57,100 --> 00:04:58,866
Revenue is $7.7 billion

123
00:04:58,900 --> 00:05:00,400
Data Center and AI

124
00:05:00,400 --> 00:05:02,700
That is, data center and AI business

125
00:05:02,700 --> 00:05:06,433
Revenue was US$5.1 billion, a year-on-year increase of 22%

126
00:05:06,433 --> 00:05:07,966
The two businesses combined

127
00:05:07,966 --> 00:05:11,466
It is the core source of revenue for Intel’s product business.

128
00:05:11,466 --> 00:05:15,333
Intel Q1 2026 total net revenue is $13.6 billion

129
00:05:15,500 --> 00:05:19,966
CCG and DCAI alone add up to US$12.8 billion

130
00:05:19,966 --> 00:05:20,866
That is to say

131
00:05:20,866 --> 00:05:24,466
Where Intel really makes money and has a real market presence

132
00:05:24,500 --> 00:05:28,233
It’s still the CPU and the product lines surrounding the CPU

133
00:05:28,400 --> 00:05:31,033
This is certainly good news for Intel

134
00:05:31,100 --> 00:05:33,566
But the problem is, the good news comes too quickly

135
00:05:33,566 --> 00:05:37,100
It will take time to prove that it can regain the rhythm of advanced manufacturing

136
00:05:37,100 --> 00:05:38,933
But customers want the product now

137
00:05:38,966 --> 00:05:40,733
Cloud vendors now want CPUs

138
00:05:40,833 --> 00:05:42,600
PC manufacturers need solutions now

139
00:05:42,600 --> 00:05:45,900
And AMD, an old rival that has been attacking and defending each other for so many years,

140
00:05:45,900 --> 00:05:49,866
I won't stand still and wait for Intel to slowly increase its yield rate before starting a war.

141
00:05:49,866 --> 00:05:50,666
these years

142
00:05:50,766 --> 00:05:53,700
AMD's CPU market share has surpassed Intel's

143
00:05:53,700 --> 00:05:55,500
Of course, strength cannot be underestimated

144
00:05:55,500 --> 00:05:59,800
Now AMD has announced an investment of more than 10 billion US dollars in the Taiwan ecosystem.

145
00:05:59,800 --> 00:06:04,500
Focus on expanding strategic cooperation, advanced packaging and AI infrastructure capabilities

146
00:06:04,500 --> 00:06:09,366
It also integrates the 6th generation EPYC, MI450X and Helios cabinet-level AI platforms

147
00:06:09,366 --> 00:06:11,066
Tied to Taiwan’s supply chain

148
00:06:11,266 --> 00:06:12,400
What does this mean?

149
00:06:12,400 --> 00:06:15,733
This shows that Su Zifeng doesn’t just want to catch up with Nvidia on GPUs.

150
00:06:16,000 --> 00:06:18,800
She is also ambitious to develop new technologies in CPU and AI servers.

151
00:06:18,800 --> 00:06:22,300
Advanced packaging and system platforms continue to attack Intel

152
00:06:22,366 --> 00:06:23,233
At the same time

153
00:06:23,233 --> 00:06:25,733
Huang Renxun is also watching covetously

154
00:06:25,733 --> 00:06:27,500
We talked about this yesterday

155
00:06:27,500 --> 00:06:30,633
He is planning to bring him specifically for a new wave of AI agents

156
00:06:30,633 --> 00:06:33,333
Created super chip RTX Spark

157
00:06:33,333 --> 00:06:35,533
Redefining the PC battlefield

158
00:06:35,533 --> 00:06:39,200
What’s even more worrying is that RTX Spark is still an Arm architecture

159
00:06:39,200 --> 00:06:40,933
Huang Renxun holds Microsoft with one hand

160
00:06:40,933 --> 00:06:42,500
One hand around x86

161
00:06:42,500 --> 00:06:45,733
Directly rushed into Intel’s most traditional and core territory

162
00:06:45,733 --> 00:06:48,366
So what Chen Liwu is facing now is not an enemy.

163
00:06:48,366 --> 00:06:50,466
Instead, several battlefields are being fought at the same time.

164
00:06:50,466 --> 00:06:51,200
At this time

165
00:06:51,200 --> 00:06:52,666
If Intel still insists

166
00:06:52,666 --> 00:06:55,833
"Everything must wait until my own 18A is fully mature."

167
00:06:55,833 --> 00:06:58,466
Then it may not be about creating a revival

168
00:06:58,466 --> 00:07:01,933
Instead, the window of opportunity for PC and AI servers has been taken away by others.

169
00:07:01,933 --> 00:07:04,166
So Chen Liwu must be more realistic

170
00:07:04,166 --> 00:07:06,533
This is why he wants to ask TSMC for production capacity

171
00:07:06,766 --> 00:07:09,400
Because TSMC can give him more than just wafers

172
00:07:09,500 --> 00:07:10,500
But time

173
00:07:10,800 --> 00:07:11,766
And his turn

174
00:07:11,766 --> 00:07:14,433
It is undoubtedly a public dismantling of "Terafab"

175
00:07:14,666 --> 00:07:17,433
Although it is not the heaviest and most critical punch

176
00:07:17,433 --> 00:07:20,966
But it directly opened up the most embarrassing reality of this project:

177
00:07:20,966 --> 00:07:23,933
Even Intel, which is responsible for providing advanced process imagination

178
00:07:23,933 --> 00:07:26,100
I still need support from TSMC

179
00:07:26,100 --> 00:07:29,766
Then why did Terafab copy TSMC in a short period of time?

180
00:07:29,766 --> 00:07:32,666
Speaking of "Terafab", we have also talked about it before

181
00:07:32,666 --> 00:07:35,366
Although I personally don't like it, I have to say

182
00:07:35,366 --> 00:07:38,166
Many people’s pursuit of it is not groundless.

183
00:07:38,166 --> 00:07:39,900
If we only look at book strength

184
00:07:39,900 --> 00:07:42,700
Terafab can even be said to be the best in the past ten years

185
00:07:42,700 --> 00:07:45,633
The closest thing to a crazy attempt to “forcibly build a TSMC”

186
00:07:45,633 --> 00:07:47,633
Think about it, who is standing behind it?

187
00:07:47,633 --> 00:07:48,733
demand side

188
00:07:48,733 --> 00:07:51,100
It’s directly Musk’s own AI empire

189
00:07:51,100 --> 00:07:53,666
Tesla wants FSD and Robotaxi

190
00:07:53,666 --> 00:07:55,433
Want Optimus Robot

191
00:07:55,433 --> 00:07:57,466
xAI wants to train and deploy Grok

192
00:07:57,466 --> 00:08:00,166
SpaceX is even more exaggerated. It not only wants to make Starlink

193
00:08:00,166 --> 00:08:02,533
I also want to use AI computing power and satellite network

194
00:08:02,533 --> 00:08:05,066
Even future space data centers are strung together

195
00:08:05,066 --> 00:08:05,933
That is to say

196
00:08:05,966 --> 00:08:08,500
Terafab does not build a factory first and then find customers

197
00:08:08,500 --> 00:08:10,333
And was born with a silver spoon in his mouth

198
00:08:10,366 --> 00:08:14,133
Directly bind the most expensive chip and most powerful computing power in Musk’s system

199
00:08:14,133 --> 00:08:16,333
The companies that tell the best growth stories

200
00:08:16,333 --> 00:08:17,600
What about the technical side?

201
00:08:17,600 --> 00:08:20,433
There is no shortage, and we directly joined hands with Intel.

202
00:08:20,433 --> 00:08:22,000
According to previous ideas

203
00:08:22,033 --> 00:08:24,800
Terafab is not making chips from scratch

204
00:08:24,866 --> 00:08:28,400
Instead, it plans to borrow advanced process routes such as Intel 14A

205
00:08:28,400 --> 00:08:30,966
Although Intel has been suppressed by TSMC in recent years,

206
00:08:30,966 --> 00:08:32,766
But after all, its foundation is still there

207
00:08:32,766 --> 00:08:34,633
And since the Kissinger era

208
00:08:34,633 --> 00:08:37,933
They are desperately trying to catch up with their four-year, five-node radical plan.

209
00:08:37,933 --> 00:08:40,500
From being two generations behind TSMC to catching up now

210
00:08:40,566 --> 00:08:44,066
The technology accumulation and engineering system accumulated over the years are all ready-made

211
00:08:44,133 --> 00:08:45,166
What's more important

212
00:08:45,166 --> 00:08:48,133
There is also the policy aura of advanced manufacturing in the United States.

213
00:08:48,133 --> 00:08:51,166
Didn't Terafab find a grassroots team to make chips?

214
00:08:51,166 --> 00:08:54,533
It is at least trying to tie itself to a few global

215
00:08:54,533 --> 00:08:57,300
Players who dare to attack the most advanced manufacturing processes

216
00:08:57,300 --> 00:08:58,433
And in this way

217
00:08:58,566 --> 00:09:00,466
The problem on the device side is also solved.

218
00:09:00,633 --> 00:09:05,200
Terafab can directly rely on Intel's supply chain to bind floor equipment quotas

219
00:09:05,200 --> 00:09:06,500
As for the talent side?

220
00:09:06,500 --> 00:09:10,200
Just wave money to poach people from Hsinchu Science Park in Taiwan

221
00:09:10,200 --> 00:09:11,566
For Musk

222
00:09:11,566 --> 00:09:15,433
TeraFab aims to mass produce 2nm chips by 2027

223
00:09:15,666 --> 00:09:18,933
He doesn't have ten years to slowly develop his own team of engineers

224
00:09:18,966 --> 00:09:20,733
He needs to do it in the shortest possible time

225
00:09:20,733 --> 00:09:23,400
Bringing together a group of people who know "how"

226
00:09:23,400 --> 00:09:25,366
And TSMC’s senior engineers

227
00:09:25,366 --> 00:09:28,600
Of course, they are the group of people in the world who know best "how to do it"

228
00:09:28,600 --> 00:09:29,333
for this

229
00:09:29,333 --> 00:09:32,933
He even took out dual shares in Tesla and SpaceX

230
00:09:32,933 --> 00:09:33,666
Where's the money?

231
00:09:33,666 --> 00:09:34,600
Not to mention

232
00:09:34,800 --> 00:09:36,600
If Terafab is really implemented

233
00:09:36,666 --> 00:09:38,800
Behind the scenes is SpaceX’s listing and financing

234
00:09:38,866 --> 00:09:42,200
It is Tesla’s cash flow and xAI’s financing ability

235
00:09:42,200 --> 00:09:44,700
It is the policy narrative of the reshoring of manufacturing in the United States.

236
00:09:44,700 --> 00:09:47,033
Other companies are most afraid of running out of money when building factories.

237
00:09:47,300 --> 00:09:48,500
What Musk lacks most is

238
00:09:48,500 --> 00:09:51,566
Let the capital market believe in his ability to change the world

239
00:09:51,566 --> 00:09:54,433
So you see, it stands to reason that this game of chess is very luxurious.

240
00:09:54,433 --> 00:09:57,866
We have customers, we have technology, we have equipment

241
00:09:57,933 --> 00:10:00,200
The talent is there and the money is there

242
00:10:00,200 --> 00:10:02,100
But why doesn't it work?

243
00:10:02,100 --> 00:10:04,600
Officially released to the public from March 21

244
00:10:04,600 --> 00:10:06,966
It’s less than three months till today

245
00:10:06,966 --> 00:10:09,666
Why did both parties become pessimistic?

246
00:10:09,666 --> 00:10:11,266
What's the difference?

247
00:10:11,266 --> 00:10:12,500
In fact, in the final analysis

248
00:10:12,500 --> 00:10:15,466
Or is it the difficulty of wafer manufacturing that Musk underestimates?

249
00:10:15,466 --> 00:10:17,933
Musk likes speed and subversion

250
00:10:17,933 --> 00:10:19,300
Like "move fast"

251
00:10:19,300 --> 00:10:22,200
But wafer manufacturing is the slowest and heaviest in the world.

252
00:10:22,200 --> 00:10:24,066
One of the most orderly industries

253
00:10:24,066 --> 00:10:27,000
It's not like you buy the best parts and put them together

254
00:10:27,000 --> 00:10:28,866
can become a TSMC

255
00:10:28,866 --> 00:10:30,000
first start

256
00:10:30,000 --> 00:10:32,300
This cooperation itself is not really locked

257
00:10:32,300 --> 00:10:36,200
Terafab appears to be the iron triangle of SpaceX, Tesla, and Intel.

258
00:10:36,200 --> 00:10:39,800
But the document is not a final, binding joint venture agreement.

259
00:10:39,800 --> 00:10:42,100
What if Intel no longer wants to cooperate?

260
00:10:42,100 --> 00:10:46,366
Then Terafab will directly lose the Intel 14A process and engineering team

261
00:10:46,366 --> 00:10:48,366
Even ASML lithography machine quota

262
00:10:48,366 --> 00:10:49,400
That is to say

263
00:10:49,400 --> 00:10:52,500
The two important advantages of Terafab we just mentioned

264
00:10:52,500 --> 00:10:53,766
technology and equipment

265
00:10:53,766 --> 00:10:55,633
It’s basically a mirror image

266
00:10:55,633 --> 00:10:57,000
once they are gone

267
00:10:57,000 --> 00:10:59,033
What about this so-called super fab?

268
00:10:59,100 --> 00:11:01,766
In an instant, you will recover from the "AI chip self-rescue plan"

269
00:11:01,766 --> 00:11:04,700
It has become an empty land with no foundation laid yet.

270
00:11:04,700 --> 00:11:06,800
And even if Intel sticks to it

271
00:11:06,800 --> 00:11:08,633
Equipment issues still exist

272
00:11:08,766 --> 00:11:11,566
Rely on Intel supply chain to bind part of device quota

273
00:11:11,566 --> 00:11:14,333
It doesn’t mean “you can get enough equipment at your own pace”

274
00:11:14,333 --> 00:11:16,433
It doesn’t mean “you can start mass production once you get the equipment.”

275
00:11:16,433 --> 00:11:19,333
High-NA EUV capacity itself is very scarce

276
00:11:19,333 --> 00:11:22,433
Moreover, TSMC, Samsung, and Intel are already lined up in front of them.

277
00:11:22,433 --> 00:11:24,600
Do you think I, Musk, can add more money?

278
00:11:24,633 --> 00:11:26,900
Okay, maybe I can squeeze it forward a little bit.

279
00:11:26,900 --> 00:11:28,033
But the problem is

280
00:11:28,033 --> 00:11:31,433
The entire supply chain has already been organized on a quarterly and yearly basis.

281
00:11:31,433 --> 00:11:33,666
Arranged according to customer verification rhythm

282
00:11:33,666 --> 00:11:36,433
It’s not like you shouted “I want to change the world” today

283
00:11:36,566 --> 00:11:39,933
ASML will be able to airdrop machines to Texas tomorrow

284
00:11:39,933 --> 00:11:43,200
What's more, advanced manufacturing processes are not limited to photolithography machines.

285
00:11:43,533 --> 00:11:45,766
Lithography is just the most dazzling pearl

286
00:11:45,766 --> 00:11:50,933
Later there is deposition, etching, ion implantation, CMP, and cleaning.

287
00:11:50,933 --> 00:11:54,533
Inspection, measurement, defect analysis, automatic handling

288
00:11:54,533 --> 00:11:56,833
Cleaning system, chemical supply

289
00:11:56,833 --> 00:11:59,966
If any link fails, the entire production line will have to wait.

290
00:11:59,966 --> 00:12:02,666
Therefore, the issue of equipment is not “can you afford it?”

291
00:12:02,700 --> 00:12:05,700
It’s about “whether it can be delivered on time, in complete sets, and stably”

292
00:12:05,700 --> 00:12:07,733
Prioritize access to your production line”

293
00:12:07,733 --> 00:12:09,666
The fight behind this is not cash.

294
00:12:09,700 --> 00:12:12,233
But industry credit and supply chain status

295
00:12:12,233 --> 00:12:15,233
Similarly, talents are not something you can just dig out if you want.

296
00:12:15,233 --> 00:12:18,533
Musk can indeed poach people with high salaries, but the problem is

297
00:12:18,533 --> 00:12:22,033
An advanced wafer fab cannot be run by just poaching a few experts.

298
00:12:22,033 --> 00:12:26,233
It requires thousands of engineers, equipment technicians, and process integration teams

299
00:12:26,233 --> 00:12:30,366
The yield team, packaging team and supply chain team have been working together for a long time

300
00:12:30,366 --> 00:12:31,933
Hire an engineer

301
00:12:31,933 --> 00:12:33,900
Can take away his personal experience

302
00:12:33,900 --> 00:12:36,166
But we can’t take away TSMC’s “platform”

303
00:12:36,166 --> 00:12:38,433
That set twisted thousands of people into one rope

304
00:12:38,433 --> 00:12:43,100
The organizational ability to consolidate decades of experience into standard operating procedures

305
00:12:43,100 --> 00:12:45,733
What’s even more frightening is the hell of yield

306
00:12:45,733 --> 00:12:47,933
Intel 14A sounds advanced

307
00:12:47,933 --> 00:12:49,700
But advanced does not mean mature

308
00:12:49,700 --> 00:12:51,833
Maturity does not mean stable mass production

309
00:12:51,833 --> 00:12:53,966
The most ruthless part of chip manufacturing is that

310
00:12:53,966 --> 00:12:57,033
Make one chip and stably mass produce a batch of chips

311
00:12:57,033 --> 00:12:58,533
Completely two abilities

312
00:12:58,533 --> 00:13:01,666
If you make a sample in the laboratory, it's called a technological breakthrough.

313
00:13:01,833 --> 00:13:04,566
You steadily ship hundreds of thousands of wafers every month

314
00:13:04,600 --> 00:13:06,266
Customers use it to make products

315
00:13:06,433 --> 00:13:09,333
Stable yield, controllable costs, and reliable delivery

316
00:13:09,400 --> 00:13:10,833
This is called commercial manufacturing process

317
00:13:10,833 --> 00:13:12,233
But it's not easy

318
00:13:12,233 --> 00:13:13,433
a dust particle

319
00:13:13,433 --> 00:13:14,700
a temperature fluctuation

320
00:13:14,700 --> 00:13:17,733
The etching is deeper and the film is thicker.

321
00:13:17,766 --> 00:13:19,400
a metal interconnect defect

322
00:13:19,400 --> 00:13:21,133
A package stress variation

323
00:13:21,133 --> 00:13:22,966
It is possible that a batch of chips will be scrapped.

324
00:13:23,100 --> 00:13:27,866
The yield rate climbed from 20% to 50%, and from 50% to 70%

325
00:13:28,133 --> 00:13:30,266
Climb from 70% to 85%

326
00:13:30,266 --> 00:13:34,066
Every percentage point is filled with countless engineers and equipment vendors

327
00:13:34,066 --> 00:13:37,133
The material supplier and customer design team worked together to achieve this

328
00:13:37,133 --> 00:13:39,733
The scary thing about TSMC is not just the technology

329
00:13:39,733 --> 00:13:42,366
But its almost paranoid manufacturing culture:

330
00:13:42,366 --> 00:13:44,966
Customer requirements cannot be ignored, and the yield rate is adjusted every day.

331
00:13:44,966 --> 00:13:48,366
The problem must be traced to the root cause, and engineers and suppliers must work together.

332
00:13:48,366 --> 00:13:50,433
Musk is good at iterating quickly

333
00:13:50,500 --> 00:13:54,166
But wafer manufacturing cannot be learned while exploding like a rocket.

334
00:13:54,166 --> 00:13:56,866
What chip customers want is not "we learned a lot"

335
00:13:56,933 --> 00:13:58,466
But "you deliver on time"

336
00:13:58,466 --> 00:14:00,100
What's more important

337
00:14:00,100 --> 00:14:02,700
But behind it stands an industrial chain that has cooperated for many years.

338
00:14:02,700 --> 00:14:04,966
There is also an entire Taiwanese semiconductor ecosystem

339
00:14:04,966 --> 00:14:07,700
An AI chip from design to delivery

340
00:14:07,700 --> 00:14:09,700
TSMC did not do it alone.

341
00:14:09,700 --> 00:14:12,700
There are EDA and IP design rules in front.

342
00:14:12,700 --> 00:14:17,600
In the middle there are equipment, materials, photoresist, silicon wafers, gases, chemicals

343
00:14:17,600 --> 00:14:22,166
Behind it are advanced packaging, testing, substrate, and HBM integration.

344
00:14:22,266 --> 00:14:25,333
Then there are Quanta, Wistron, and Inventec

345
00:14:25,333 --> 00:14:27,500
Hon Hai these servers ODM

346
00:14:27,500 --> 00:14:30,966
Turn chips into complete machines, cabinets, and AI factories

347
00:14:30,966 --> 00:14:33,300
This is the hardest thing to copy about Musk

348
00:14:33,300 --> 00:14:34,966
He can recruit a few engineers

349
00:14:34,966 --> 00:14:38,533
But we cannot take away the density of the supply chain formed by the Hsinchu Science Park over decades.

350
00:14:38,533 --> 00:14:40,066
He can buy equipment

351
00:14:40,066 --> 00:14:43,366
But I can’t buy TSMC, ASML, and Applied Materials

352
00:14:43,366 --> 00:14:46,900
Long-term synergy between Lam Research and Tokyo Electron

353
00:14:46,900 --> 00:14:48,533
He can find cooperation with Intel

354
00:14:48,533 --> 00:14:52,500
But we cannot copy Apple, NVIDIA, AMD, and Qualcomm in the short term.

355
00:14:52,500 --> 00:14:56,700
Generations of MediaTek products help TSMC develop process databases

356
00:14:56,700 --> 00:14:58,566
He can build a Texas factory

357
00:14:58,566 --> 00:15:00,100
But we can’t build something like Taiwan.

358
00:15:00,100 --> 00:15:04,066
One phone call can handle packaging, testing, substrates, and server foundry

359
00:15:04,066 --> 00:15:07,433
Industrial network with heat dissipation and power supply all on the table

360
00:15:07,433 --> 00:15:10,733
This is why Huang Renxun, Su Zifeng, and Chen Liwu

361
00:15:10,733 --> 00:15:12,500
The reason why everyone is running to Taiwan

362
00:15:12,500 --> 00:15:14,166
Because they know too well

363
00:15:14,166 --> 00:15:17,200
What is truly scarce in the AI era is not a factory building

364
00:15:17,200 --> 00:15:19,266
It's the network behind the factory

365
00:15:19,266 --> 00:15:21,733
TSMC is just the brightest node in this network

366
00:15:21,733 --> 00:15:23,000
truly terrible

367
00:15:23,033 --> 00:15:25,600
It is the entire Taiwan semiconductor industry chain behind it.

368
00:15:25,600 --> 00:15:28,466
It has become the most important infrastructure in the AI era

369
00:15:28,466 --> 00:15:31,066
Jen-Hsun Huang says Taiwan is the center of the AI revolution

370
00:15:31,066 --> 00:15:32,800
Not because he likes to shout slogans

371
00:15:32,800 --> 00:15:36,066
It’s because NVIDIA’s AI throne requires an entire supply chain:

372
00:15:36,133 --> 00:15:38,933
Cooperate with MediaTek to develop RTX Spark

373
00:15:38,933 --> 00:15:40,966
Cooperate with TSMC to manufacture chips

374
00:15:41,033 --> 00:15:45,333
Advanced packaging connects GPU, CPU and high-speed memory

375
00:15:45,400 --> 00:15:49,366
Quanta, Wistron, and Hon Hai install them into servers and cabinets

376
00:15:49,366 --> 00:15:52,233
Finally, we must rely on optical communications, electricity, and liquid cooling.

377
00:15:52,233 --> 00:15:54,666
Really run the entire AI factory

378
00:15:54,666 --> 00:15:57,366
Su Zifeng’s US$10 billion is not just random money.

379
00:15:57,366 --> 00:16:00,766
AMD names ASE, Silicon Products, and Power Technologies

380
00:16:00,766 --> 00:16:02,266
It's because she knows very well

381
00:16:02,266 --> 00:16:05,566
If you are forever stuck at the back of the TSMC CoWoS queue

382
00:16:05,666 --> 00:16:08,433
It will be difficult for AMD to really catch up with Nvidia

383
00:16:08,433 --> 00:16:11,333
So she wants to support more advanced packaging paths

384
00:16:11,333 --> 00:16:15,133
Let MI450, EPYC Venice, Helios

385
00:16:15,133 --> 00:16:18,000
These system-level products have a chance of being delivered on time

386
00:16:18,000 --> 00:16:19,466
Chen Liwu is more realistic

387
00:16:19,733 --> 00:16:21,600
Intel wants to revive chip manufacturing

388
00:16:21,600 --> 00:16:24,033
But the AI PC window is already open

389
00:16:24,166 --> 00:16:27,733
He cannot wait for 18A and 14A to fully mature before starting a war

390
00:16:27,966 --> 00:16:29,766
So he wants TSMC’s production capacity

391
00:16:29,766 --> 00:16:32,166
Ask Taiwan Ecology to help him buy time

392
00:16:32,433 --> 00:16:34,266
So this round is really good

393
00:16:34,266 --> 00:16:35,800
not an isolated concept

394
00:16:35,800 --> 00:16:38,266
It’s about several hard links of AI infrastructure:

395
00:16:38,333 --> 00:16:41,733
Advanced process, advanced packaging, AI server ODM

396
00:16:41,733 --> 00:16:43,733
High-speed interconnect, power liquid cooling

397
00:16:43,733 --> 00:16:46,133
TSMC relies on deterministic delivery

398
00:16:46,166 --> 00:16:50,566
ASE, Silicon Products, and Licheng suffer from advanced packaging spillovers

399
00:16:50,566 --> 00:16:55,333
Quanta, Wistron, and Hon Hai focus on AI servers and full cabinet delivery

400
00:16:55,333 --> 00:16:58,733
Corning, Marvell, Coherent, Lumentum

401
00:16:58,733 --> 00:17:01,033
Eat optical communications and high-speed interconnection

402
00:17:01,033 --> 00:17:04,933
This is the change in AI investment from the first half to the second half.

403
00:17:04,933 --> 00:17:08,700
In the first half, the market only asked who has the strongest GPU

404
00:17:08,700 --> 00:17:11,400
In the second half, the market will start asking:

405
00:17:11,400 --> 00:17:15,300
Who can turn a GPU into an entire operational AI factory?

406
00:17:15,300 --> 00:17:18,300
And the answer is not just at Nvidia’s press conference

407
00:17:18,300 --> 00:17:20,533
Also included in the wafer manufacturing capacity table

408
00:17:20,600 --> 00:17:22,466
In advanced packaging production lines

409
00:17:22,466 --> 00:17:23,966
In the server factory

410
00:17:24,200 --> 00:17:26,766
In orders from optical communications and liquid cooling companies

411
00:17:26,766 --> 00:17:28,466
How did you lay it out?

412
00:17:28,733 --> 00:17:30,200
Welcome to share your views

413
00:17:30,366 --> 00:17:32,600
I am Olivia who will accompany you to unlock the code of wealth.

414
00:17:32,700 --> 00:00:04,400
Like and follow, let's go together~

415
00:00:04,400 --> 00:00:06,566
“Terafab”

416
00:00:34,733 --> 00:00:37,100
"I may not be able to handle the chip supply."

417
00:01:45,200 --> 00:01:47,666
Olivia

418
00:02:53,466 --> 00:03:02,733
888

419
00:03:19,533 --> 00:03:23,633
18A

420
00:04:09,066 --> 00:04:18,566
Unpack tasks, adjust tools, and read files

421
00:04:52,666 --> 00:05:11,466
Client Computing Group: $7.7 billion

422
00:06:54,233 --> 00:07:01,933
18A fully mature

423
00:09:54,433 --> 00:10:00,200
Customer

424
00:10:52,500 --> 00:10:55,633
technology equipment

425
00:11:45,766 --> 00:11:59,966
Deposition, etching, ion implantation

426
00:12:03,533 --> 00:12:12,233
On time, complete set, stable

427
00:12:53,966 --> 00:12:58,533
make a chip

428
00:13:06,433 --> 00:13:10,833
Stable yield

429
00:13:42,366 --> 00:13:48,366
Customer requirements cannot be ignored, and the yield rate is adjusted every day.

430
00:14:09,700 --> 00:14:27,500
There are EDA and IP design rules in front.

431
00:16:38,333 --> 00:00:05,333
Advanced process, advanced packaging, AI server ODM

432
00:00:05,333 --> 00:00:06,566
Turn off the engine?

433
00:03:20,166 --> 00:03:23,633
14A

434
00:04:11,566 --> 00:04:18,566
Run code, access databases, manage permissions, and handle network communications

435
00:04:58,900 --> 00:05:11,466
Data Center and AI: $5.1 billion

436
00:09:55,700 --> 00:10:00,200
Technology

437
00:11:49,066 --> 00:11:59,966
CMP, cleaning

438
00:12:05,700 --> 00:12:12,233
Prioritize access to your production line

439
00:12:55,300 --> 00:12:58,533
Stable mass production of a batch of chips

440
00:13:07,433 --> 00:13:10,833
Cost controllable

441
00:13:44,966 --> 00:13:48,366
The problem must be traced to the root cause, and engineers and suppliers must work together.

442
00:14:12,700 --> 00:14:27,500
In the middle there are equipment, materials, photoresist, silicon wafers, gases, chemicals

443
00:16:41,733 --> 00:04:15,700
High-speed interconnect, power liquid cooling

444
00:04:15,700 --> 00:04:18,566
Coordinate multiple models and multiple applications simultaneously

445
00:09:56,766 --> 00:10:00,200
Equipment

446
00:11:50,933 --> 00:11:59,966
Inspection, measurement, defect analysis, automatic handling

447
00:13:08,333 --> 00:13:10,833
Reliable delivery

448
00:14:17,600 --> 00:09:57,933
Behind it are advanced packaging, testing, substrate, and HBM integration.

449
00:09:57,933 --> 00:10:00,200
Talent

450
00:11:54,533 --> 00:11:59,966
Cleaning system, chemical supply

451
00:14:22,233 --> 00:09:59,066
Then there are Quanta, Wistron, Inventec, and Hon Hai

452
00:09:59,066 --> 00:10:00,200
money