AI产业已秘密转向！趁市场未动，这3大机会抓紧布局！

1
00:00:00,080 --> 00:00:01,260
美股波华尔街专利

2
00:00:01,260 --> 00:00:01,679
大家好

3
00:00:01,679 --> 00:00:02,319
我是米诺君

4
00:00:02,319 --> 00:00:04,740
你们的美股探路者

5
00:00:04,940 --> 00:00:06,419
在刚刚过去的这周

6
00:00:06,419 --> 00:00:08,300
美股最大的几家科技公司

7
00:00:08,300 --> 00:00:11,250
纷纷发布了他们最新一季度的财报

8
00:00:11,250 --> 00:00:12,310
这次财报季

9
00:00:12,310 --> 00:00:15,369
最重要的还真不是这些公司业绩的高低

10
00:00:15,369 --> 00:00:17,089
也不是他们股价的变动

11
00:00:17,089 --> 00:00:19,070
而是所有这些大科技

12
00:00:19,070 --> 00:00:22,269
几乎不约而同地做出了一个变化

13
00:00:22,269 --> 00:00:24,609
这个变化隐秘不易察觉

14
00:00:24,609 --> 00:00:27,870
但却可能彻底改变整个AI产业

15
00:00:27,870 --> 00:00:31,030
它预示着AI产业中的一场方向之争

16
00:00:31,030 --> 00:00:32,429
终于尘埃落定了

17
00:00:32,429 --> 00:00:37,098
也意味着接下来的AI投资将会发生彻底的变化

18
00:00:37,499 --> 00:00:40,098
作为跟踪AI产业最早的博主之一

19
00:00:40,098 --> 00:00:41,859
这么重要的一次产业转向

20
00:00:41,859 --> 00:00:44,640
美投君一定得第一时间跟大家分享

21
00:00:44,640 --> 00:00:46,399
趁他还没有彻底坐实

22
00:00:46,399 --> 00:00:49,909
我们普通投资者或许还有不错的投资机会

23
00:00:49,909 --> 00:00:52,170
所以本期视频我们就来好好聊聊

24
00:00:52,170 --> 00:00:54,689
这个AI产业的最新变化

25
00:00:54,689 --> 00:00:55,210
视频

26
00:00:55,210 --> 00:00:56,130
最后我也会分享

27
00:00:56,130 --> 00:00:59,979
我个人具体的投资方向和投资标的

28
00:01:06,900 --> 00:01:08,219
在本次财报集中

29
00:01:08,219 --> 00:01:11,500
有这么一个重要的细节被很多人忽略了

30
00:01:11,500 --> 00:01:12,920
在谷歌的财报中

31
00:01:12,920 --> 00:01:16,579
市场一向最关心的AI大模型GEMINI的月活用户数

32
00:01:16,579 --> 00:01:19,030
这一次公司居然没有披露

33
00:01:19,030 --> 00:01:20,950
要知道詹姆乃的月活

34
00:01:20,950 --> 00:01:23,200
可是去年谷歌股价起飞的关键

35
00:01:23,200 --> 00:01:24,959
也正是这个数据的公布

36
00:01:24,959 --> 00:01:28,459
当时吹起了GEMINI向chat gbt反攻的号角

37
00:01:28,459 --> 00:01:31,340
然而就是这么重要的一个数据公司

38
00:01:31,340 --> 00:01:32,519
没有做出任何解释

39
00:01:32,519 --> 00:01:34,310
直接就不公布了

40
00:01:34,310 --> 00:01:35,069
管理层

41
00:01:35,069 --> 00:01:38,409
转而高调宣布了另一个令人耐人寻味的数据

42
00:01:38,409 --> 00:01:43,239
GEMINI企业端agent服务Gemini enterprise的月活数

43
00:01:43,239 --> 00:01:47,060
该数字在最新一个季度环比增长了40%

44
00:01:47,060 --> 00:01:50,239
一个是有7.5亿规模的超大数据

45
00:01:50,239 --> 00:01:53,459
另一个是只有几百万付费席位的小数

46
00:01:53,459 --> 00:01:56,120
那公司为什么会选择抛弃前者

47
00:01:56,120 --> 00:01:58,340
而公布后者呢

48
00:01:58,899 --> 00:01:59,759
无独有偶

49
00:01:59,759 --> 00:02:02,599
OpenAI那边也做出了一个类似的变化

50
00:02:02,599 --> 00:02:05,459
而他们则走的更加极端

51
00:02:05,799 --> 00:02:07,200
很多看官可能还不知道

52
00:02:07,200 --> 00:02:08,419
在过去这一个月里

53
00:02:08,419 --> 00:02:11,680
OpenAI悄悄做出了许多战略上的改变

54
00:02:11,680 --> 00:02:15,460
他们彻底关停了风靡一时的视频生成软件SORA

55
00:02:15,460 --> 00:02:17,400
暂停了成人模型的研发

56
00:02:17,400 --> 00:02:19,860
连之前高调推出的科研项目AI

57
00:02:19,860 --> 00:02:21,919
for science也彻底被砍掉

58
00:02:21,919 --> 00:02:25,060
为此公司将内部人士做了一次大换血

59
00:02:25,060 --> 00:02:28,099
所有的资源现在都集中在了一个产品当中

60
00:02:28,099 --> 00:02:30,150
这就是codex

61
00:02:30,550 --> 00:02:33,590
codex是OpenAI版的AI agent

62
00:02:33,590 --> 00:02:36,860
它直接对标的是athropic cloud code

63
00:02:36,860 --> 00:02:39,860
起初他就是单纯帮程序员写代码的软件

64
00:02:39,860 --> 00:02:42,460
后来随着MCP skill等能力的加入

65
00:02:42,460 --> 00:02:44,860
现在他也能够完成除了编程之外

66
00:02:44,860 --> 00:02:46,620
很多其他的工作流

67
00:02:46,629 --> 00:02:49,729
要知道编程原本可是cloud的专场

68
00:02:49,729 --> 00:02:51,509
而随着OpenAI的战略转向

69
00:02:51,509 --> 00:02:53,009
codex奋起直追

70
00:02:53,009 --> 00:02:55,009
现在的势头非常凶猛

71
00:02:55,009 --> 00:02:58,399
年初时codex大概只有50万的周货用户

72
00:02:58,399 --> 00:03:01,579
而4月初这个数字便已经突破了300万

73
00:03:01,579 --> 00:03:03,158
而又过了仅仅两周

74
00:03:03,158 --> 00:03:06,069
最新的数字已经冲上了400万

75
00:03:06,069 --> 00:03:08,389
OpenAI在4月初发推文表示

76
00:03:08,389 --> 00:03:09,330
企业端的收入

77
00:03:09,330 --> 00:03:11,800
现在已经超过了总收入的40%

78
00:03:11,800 --> 00:03:13,719
并且还在持续的提升

79
00:03:13,719 --> 00:03:17,280
预计年底前就能和消费端收入持平

80
00:03:17,889 --> 00:03:19,569
从谷歌到OpenAI

81
00:03:19,569 --> 00:03:21,560
你看出了什么端倪吗

82
00:03:21,560 --> 00:03:22,340
很显然

83
00:03:22,340 --> 00:03:24,099
现在所有AI大模型公司

84
00:03:24,099 --> 00:03:27,080
都不约而同的在商业战略上做出了

85
00:03:27,080 --> 00:03:29,349
从C端向B端的转向

86
00:03:29,349 --> 00:03:33,659
而且能看出来乞爷们这次是下定了决心

87
00:03:34,139 --> 00:03:37,400
这一切的根源都要来自于另一家大模型公司

88
00:03:37,400 --> 00:03:38,599
Anthropic

89
00:03:38,599 --> 00:03:39,560
一个月前

90
00:03:39,560 --> 00:03:41,919
ATHROPIC公布了最新的a arr

91
00:03:41,919 --> 00:03:44,919
企业经常性收入从年初的90亿美元

92
00:03:44,919 --> 00:03:48,069
一个季度的时间直接飙升至了300亿

93
00:03:48,069 --> 00:03:50,770
这倒不是说cloud模型本身有多强

94
00:03:50,770 --> 00:03:53,110
而是ASTROPIC率先证明了一件事

95
00:03:53,110 --> 00:03:57,710
那就是企业端agent的商业模式彻底跑通了

96
00:03:57,710 --> 00:04:01,889
说白了就是agent真的可以帮企业完成实际的工作

97
00:04:01,889 --> 00:04:05,960
而为此企业们也愿意不惜余力的花钱

98
00:04:06,520 --> 00:04:10,259
要知道过去这两年大模型公司都在拼C端入口

99
00:04:10,259 --> 00:04:11,580
都在拼他们的用户规模

100
00:04:11,580 --> 00:04:13,459
还在拼他们的聊天体验

101
00:04:13,459 --> 00:04:17,100
为此他们不惜花重金去补贴自己的用户

102
00:04:17,100 --> 00:04:19,579
可是这些用户虽然基数庞大

103
00:04:19,579 --> 00:04:21,399
但是付费意愿却极差

104
00:04:21,399 --> 00:04:23,079
使用场景也非常分散

105
00:04:23,079 --> 00:04:24,990
粘性更是完全谈不上

106
00:04:24,990 --> 00:04:26,810
所以我们看到过去这半年

107
00:04:26,810 --> 00:04:28,949
OpenAI不仅财务状况令人担忧

108
00:04:28,949 --> 00:04:31,829
而且用户也被C端生态更好的詹姆乃

109
00:04:31,829 --> 00:04:33,310
抢走了不少

110
00:04:33,920 --> 00:04:35,759
但ANTHROPIC的变现成功

111
00:04:35,759 --> 00:04:39,129
则让这两家大模型公司彻底醒悟了过来

112
00:04:39,129 --> 00:04:41,990
与其在不确定的C端去砸钱拼杀

113
00:04:41,990 --> 00:04:45,519
为什么不在已经验证的B端去哗哗挣钱呢

114
00:04:45,519 --> 00:04:49,420
于是OpenAI和谷歌也纷纷跟上ANTHROPIC的步伐

115
00:04:49,420 --> 00:04:51,480
向B端进行战略转移

116
00:04:51,480 --> 00:04:56,079
而重点就放在了AI agent的应用能力上

117
00:04:59,620 --> 00:05:02,100
那对于咱投资者而言

118
00:05:02,100 --> 00:05:04,339
大模型向B端战略转移

119
00:05:04,339 --> 00:05:06,769
这究竟意味着什么呢

120
00:05:07,129 --> 00:05:09,550
首先这会带来整个大模型企业

121
00:05:09,550 --> 00:05:11,538
商业模式上的变化

122
00:05:11,538 --> 00:05:14,579
大模型的商业模式将会从之前的订阅收费

123
00:05:14,579 --> 00:05:16,939
转而向用量收费

124
00:05:17,529 --> 00:05:20,529
之前的大模型主要都是以按月订阅为主

125
00:05:20,529 --> 00:05:23,288
但订阅制其实并不适合大模型

126
00:05:23,288 --> 00:05:26,988
订阅制只有在成本相对稳定的时候才比较合理

127
00:05:26,988 --> 00:05:27,908
比如健身卡

128
00:05:27,908 --> 00:05:29,788
你一个用户甭管举多少铁

129
00:05:29,788 --> 00:05:32,600
健身房的成本都不会发生什么太大变化

130
00:05:32,600 --> 00:05:34,060
但AI订阅不同

131
00:05:34,060 --> 00:05:36,680
它可是要实实在在的烧算力的

132
00:05:36,680 --> 00:05:39,500
而不同的用户算力消耗又大不相同

133
00:05:39,500 --> 00:05:41,769
成本可能是天差地别

134
00:05:41,769 --> 00:05:43,889
尤其现在agent模式推出

135
00:05:43,889 --> 00:05:45,970
AI可以24小时的无休工作

136
00:05:45,970 --> 00:05:49,158
那对于算力的消耗更可能是天量级别的

137
00:05:49,158 --> 00:05:51,238
如果说还是所有人都按照

138
00:05:51,238 --> 00:05:53,038
固定订阅费的模式来收费

139
00:05:53,038 --> 00:05:56,379
那大模型企业就会越来越吃不消

140
00:05:56,899 --> 00:05:58,279
那怎么办呢

141
00:05:58,279 --> 00:06:01,480
这就注定会向按量收费来转移

142
00:06:01,480 --> 00:06:03,600
用户先给一个基础的订阅费

143
00:06:03,600 --> 00:06:04,980
但算力是有限的

144
00:06:04,980 --> 00:06:06,180
如果需要更多算力

145
00:06:06,180 --> 00:06:07,600
就需要订阅更贵的服务

146
00:06:07,600 --> 00:06:09,500
或者直接按量收费

147
00:06:09,500 --> 00:06:11,920
这便是现在cloud的收费模式

148
00:06:11,920 --> 00:06:13,139
也是已经被验证

149
00:06:13,139 --> 00:06:15,470
变现效率极高的商业模式

150
00:06:15,470 --> 00:06:18,290
它不仅会给大模型企业带来收入的增加

151
00:06:18,290 --> 00:06:21,550
而且利润也会有大幅的增长

152
00:06:21,949 --> 00:06:22,709
关于这一点

153
00:06:22,709 --> 00:06:24,410
我自己是深有体会

154
00:06:24,410 --> 00:06:26,569
我平时工作会大量应用cloud

155
00:06:26,569 --> 00:06:28,759
做资料收集和数据分析

156
00:06:28,759 --> 00:06:31,399
cloud在算力上可是太抠门了

157
00:06:31,399 --> 00:06:32,980
总是token不够用

158
00:06:32,980 --> 00:06:34,899
而一旦任务已经开始执行了

159
00:06:34,899 --> 00:06:37,779
你就很难因为token不够而终止任务

160
00:06:37,779 --> 00:06:40,540
那这时我就不得不再去购买额外的token

161
00:06:40,540 --> 00:06:41,860
一个月花个几百美元

162
00:06:41,860 --> 00:06:43,199
那都是常有的事

163
00:06:43,199 --> 00:06:46,029
而要说我还算是一个比较低端的用户

164
00:06:46,029 --> 00:06:47,370
我们美投的程序员

165
00:06:47,370 --> 00:06:49,350
那才是重度的token消耗者

166
00:06:49,350 --> 00:06:51,129
24小时无休消耗

167
00:06:51,129 --> 00:06:53,629
尽管他们比我有更多的手段来提升他们

168
00:06:53,629 --> 00:06:54,699
token的使用效率

169
00:06:54,699 --> 00:06:57,180
但当我每次看到公司每月的AI账单

170
00:06:57,180 --> 00:06:57,579
那一刻

171
00:06:57,579 --> 00:07:01,230
我就知道这帮人又开始要整活了

172
00:07:01,550 --> 00:07:02,949
那你说作为老板

173
00:07:02,949 --> 00:07:04,490
我愿意花这钱吗

174
00:07:04,490 --> 00:07:05,889
肯定愿意啊

175
00:07:05,889 --> 00:07:09,158
token再贵也没有再招个程序员贵对吧

176
00:07:09,158 --> 00:07:11,218
我的员工用AI用的越多

177
00:07:11,218 --> 00:07:12,538
我就越开心

178
00:07:12,538 --> 00:07:15,029
说明他们帮我省了越多钱

179
00:07:15,029 --> 00:07:17,509
所以说AIA阵的模式的推出

180
00:07:17,509 --> 00:07:19,810
非常适合企业端用户

181
00:07:19,810 --> 00:07:22,269
它会勾引着你花越来越多的钱

182
00:07:22,269 --> 00:07:24,410
而且啊你还心甘情愿

183
00:07:24,410 --> 00:07:25,769
对于大模型而言

184
00:07:25,769 --> 00:07:28,410
这就是巨大的变现潜力

185
00:07:28,949 --> 00:07:31,889
所以你看现在有了验证的变现道路

186
00:07:31,889 --> 00:07:33,720
有了按量收费的商业模式

187
00:07:33,720 --> 00:07:35,279
我想很快我们就会看到

188
00:07:35,279 --> 00:07:39,048
这三大模型在收入端和利润端的集体释放

189
00:07:39,048 --> 00:07:42,560
这个爆发节点现在已经离我们非常接近了

190
00:07:42,560 --> 00:07:46,160
而这对于我们普通投资者而言可太重要了

191
00:07:46,160 --> 00:07:48,660
因为它会带来很多投资逻辑的改变

192
00:07:48,660 --> 00:07:51,519
以及大量的投资机会

193
00:07:56,300 --> 00:07:59,579
到这里如果我们把这上面两个变化结合起来看

194
00:07:59,579 --> 00:08:02,500
你就会发现一件非常重要的事

195
00:08:02,819 --> 00:08:04,600
你琢磨这过去大半年

196
00:08:04,600 --> 00:08:07,360
市场对于AI最大的担忧是什么呢

197
00:08:07,360 --> 00:08:09,420
其实就是AI变不了

198
00:08:09,420 --> 00:08:13,120
现这些大模型公司每年收那么多钱去训练

199
00:08:13,120 --> 00:08:13,839
去买芯片

200
00:08:13,839 --> 00:08:15,459
去扩充他们的数据中心

201
00:08:15,459 --> 00:08:18,658
到底什么时候才能把这些钱给赚回来呢

202
00:08:18,658 --> 00:08:20,778
这个问题就像一块大石头

203
00:08:20,778 --> 00:08:23,178
一直压在整个AI板块的头上

204
00:08:23,178 --> 00:08:25,059
谁都给不出答案

205
00:08:25,769 --> 00:08:28,149
但现在情况发生了明显的变化

206
00:08:28,149 --> 00:08:31,750
大模型的变现和利润提升已经一触即发了

207
00:08:31,750 --> 00:08:35,889
那么市场对于AI最大的担忧很快也会被掀开

208
00:08:35,889 --> 00:08:39,409
那么那些此前被变现担忧耽误的最狠的股票

209
00:08:39,409 --> 00:08:42,850
未来可能会迎来最猛烈的反弹

210
00:08:43,428 --> 00:08:47,109
首当其冲的我认为就是所有的云计算企业

211
00:08:47,109 --> 00:08:48,509
过去这大半年

212
00:08:48,509 --> 00:08:51,590
大部分的云计算企业的股价都有所承压

213
00:08:51,590 --> 00:08:53,029
核心原因就是因为

214
00:08:53,029 --> 00:08:56,899
市场担心他们巨大的资本性开支能否看到回报

215
00:08:56,899 --> 00:08:58,860
只要他们稍微一提资本开支

216
00:08:58,860 --> 00:09:00,659
市场就会给予他们惩罚

217
00:09:00,659 --> 00:09:03,289
即便到了现在也仍然是如此

218
00:09:03,289 --> 00:09:06,009
这也让他们的估值受到了很大的压缩

219
00:09:06,009 --> 00:09:09,250
现在三大云厂商的远期市盈率只有24倍

220
00:09:09,250 --> 00:09:10,029
可以说

221
00:09:10,029 --> 00:09:13,690
他们是受AI变现担忧压力最大的公司之一

222
00:09:13,690 --> 00:09:17,370
因而在未来也有望迎来较大的复苏

223
00:09:18,009 --> 00:09:22,078
另一大机会就是OpenAI阵营的一系列公司

224
00:09:22,078 --> 00:09:23,899
上一轮AI变现担忧中

225
00:09:23,899 --> 00:09:26,429
OpenAI绝对是处于漩涡中心

226
00:09:26,429 --> 00:09:29,049
由于公司非常激进的到处签单

227
00:09:29,049 --> 00:09:31,940
他们的财务状况也备受市场质疑

228
00:09:31,940 --> 00:09:34,779
因而受连累的还有一众相关公司

229
00:09:34,779 --> 00:09:35,759
比如英伟达

230
00:09:35,759 --> 00:09:36,299
Amd

231
00:09:36,299 --> 00:09:36,779
微软

232
00:09:36,779 --> 00:09:38,080
甲骨文等等等等

233
00:09:38,080 --> 00:09:38,820
人数众多

234
00:09:38,820 --> 00:09:40,389
我就不一一点名了

235
00:09:40,389 --> 00:09:42,830
现在随着大模型变现能力的提升

236
00:09:42,830 --> 00:09:45,490
这些担忧的环节也将给相应的公司

237
00:09:45,490 --> 00:09:47,830
带来不错的上行潜力

238
00:09:48,330 --> 00:09:51,109
最后便是AI基础层的公司

239
00:09:51,109 --> 00:09:53,708
之前市场担心大公司的资本性支出

240
00:09:53,708 --> 00:09:55,969
在今年之后就将有所放缓了

241
00:09:55,969 --> 00:09:58,620
原因也是因为AI变现不足

242
00:09:58,620 --> 00:10:02,240
但是有了刚刚我们说的弊端变现的这个前提

243
00:10:02,240 --> 00:10:06,470
我相信大公司们会更有底气去进一步的投入AI

244
00:10:06,470 --> 00:10:08,029
所以我们看到本季度

245
00:10:08,029 --> 00:10:10,750
大科技们依旧在疯狂的增加资本性开支

246
00:10:10,750 --> 00:10:13,919
就连印象最为保守的微软也开始激进了起来

247
00:10:13,919 --> 00:10:15,019
最新的财报显示

248
00:10:15,019 --> 00:10:17,519
他们已经将资本性支出从1300亿

249
00:10:17,519 --> 00:10:19,818
大幅提升到了1900亿

250
00:10:19,818 --> 00:10:20,599
很显然

251
00:10:20,599 --> 00:10:22,778
管理层内部已经先于华尔街

252
00:10:22,778 --> 00:10:25,389
看到了足够的变现潜力

253
00:10:25,389 --> 00:10:29,909
而这便能够进一步延续基础层公司的增长动能

254
00:10:34,559 --> 00:10:36,620
说到这可能有人会好奇了

255
00:10:36,620 --> 00:10:38,820
你美投君一下说了三个大的方向

256
00:10:38,820 --> 00:10:40,820
每个方向中又有那么多的公司

257
00:10:40,820 --> 00:10:43,179
那具体应该如何选择呢

258
00:10:43,179 --> 00:10:46,360
这个问题的答案就隐藏在下一个问题当中

259
00:10:46,360 --> 00:10:50,269
既然现在三大模型全部都转向了B端AI agent应用

260
00:10:50,269 --> 00:10:52,330
那么在竞争如此拥挤的领域

261
00:10:52,330 --> 00:10:54,818
最终谁会先跑出来呢

262
00:10:54,818 --> 00:10:57,219
什么才是决定胜负的关键呢

263
00:10:57,219 --> 00:11:01,009
这个问题决定了我们应该具体如何布局

264
00:11:01,009 --> 00:11:02,909
而要想搞清楚问题的答案

265
00:11:02,909 --> 00:11:05,740
我们得先分层级来讨论

266
00:11:06,059 --> 00:11:08,519
第一层是模型技术层

267
00:11:08,519 --> 00:11:09,860
如果是两年前

268
00:11:09,860 --> 00:11:12,519
模型技术能力可能是决定性因素

269
00:11:12,519 --> 00:11:14,299
当时的领先者是OpenAI

270
00:11:14,299 --> 00:11:16,440
他们也正是靠着模型能力的领先

271
00:11:16,440 --> 00:11:19,210
一度拿下了90%的市场份额

272
00:11:19,210 --> 00:11:21,309
但现在在模型能力上

273
00:11:21,309 --> 00:11:24,049
各大模型已经很难再拉开差距了

274
00:11:24,049 --> 00:11:25,230
他确实有影响

275
00:11:25,230 --> 00:11:28,049
但已经不再是决定性的了

276
00:11:28,438 --> 00:11:31,178
第二层是agent应用层

277
00:11:31,178 --> 00:11:32,479
过去这半年

278
00:11:32,479 --> 00:11:33,078
我们看到

279
00:11:33,078 --> 00:11:37,009
拉开差距的决定性因素就在于这个agent应用层

280
00:11:37,009 --> 00:11:39,230
其中的领先者就是ANTHROPIC

281
00:11:39,230 --> 00:11:40,649
和另外两家大模型

282
00:11:40,649 --> 00:11:42,009
专注C端用户不同

283
00:11:42,009 --> 00:11:44,799
ATHROPIC始终专注在B端应用层

284
00:11:44,799 --> 00:11:48,149
从cloud code到后来的cold work都是如此

285
00:11:48,149 --> 00:11:51,210
这也让他们从今年年初开始迎来了爆发

286
00:11:51,210 --> 00:11:54,509
也引领了整个AI产业的战略转向

287
00:11:55,000 --> 00:11:57,600
但最近我们也发现了一个有趣的现象

288
00:11:57,600 --> 00:12:00,990
应用层的先发优势也很难一直保持

289
00:12:00,990 --> 00:12:02,450
就像是编程领域

290
00:12:02,450 --> 00:12:05,110
cloud code一直是毫无疑问的领先者

291
00:12:05,110 --> 00:12:08,919
但是随着OpenAI的战略重心彻底向codex转移

292
00:12:08,919 --> 00:12:10,340
仅仅一个季度的时间

293
00:12:10,340 --> 00:12:12,080
他们便已迎头赶上

294
00:12:12,080 --> 00:12:13,500
为了制作本期视频

295
00:12:13,500 --> 00:12:16,399
我还特意强迫自己用了一段时间的codex

296
00:12:16,399 --> 00:12:18,750
我发现体验确实还不错

297
00:12:18,750 --> 00:12:21,309
而我们公司的程序员更是已经主动

298
00:12:21,309 --> 00:12:23,899
将大部分的工作量都转向了codex

299
00:12:23,899 --> 00:12:27,639
这点啊从实际的codex增长数据中也能够看出

300
00:12:27,639 --> 00:12:29,840
他从50万的用户到400万

301
00:12:29,840 --> 00:12:32,419
也就用了不到三个月的时间

302
00:12:32,860 --> 00:12:33,759
多说一句

303
00:12:33,759 --> 00:12:36,549
咱这儿啊没提Gemini enterprise

304
00:12:36,549 --> 00:12:40,590
它的AI应用端和cloud和codex走的路线不太相同

305
00:12:40,590 --> 00:12:44,149
它主要是走企业级的AIA制的平台

306
00:12:44,149 --> 00:12:47,110
而cloud和codex还是从企业员工

307
00:12:47,110 --> 00:12:49,909
个人层面去提高他们的生产效率

308
00:12:49,909 --> 00:12:52,470
尽管最终他们肯定会在某个地方相遇

309
00:12:52,470 --> 00:12:54,019
形成正面的竞争

310
00:12:54,019 --> 00:12:55,899
但就当前的情况而言

311
00:12:55,899 --> 00:12:57,320
Germana enterprise

312
00:12:57,320 --> 00:13:00,328
暂时和另外两家没有太大的竞争关系

313
00:13:00,328 --> 00:13:02,149
现在的重点还是要放在

314
00:13:02,149 --> 00:13:05,580
OpenAI和ANTHROPIC的正面硬刚上

315
00:13:05,860 --> 00:13:08,139
那么既然从agent的应用层面来看

316
00:13:08,139 --> 00:13:11,580
OpenAI和ANTHROPIC似乎也很难彻底拉开差距

317
00:13:11,580 --> 00:13:14,769
那么最终的胜负手究竟在哪呢

318
00:13:14,769 --> 00:13:17,149
答案就隐藏在第三层

319
00:13:17,149 --> 00:13:19,940
也就是算力层上

320
00:13:20,299 --> 00:13:21,600
AIA阵的时代

321
00:13:21,600 --> 00:13:24,509
算力的消耗会呈指数级增加

322
00:13:24,509 --> 00:13:26,450
如果应用层难以拉开差距

323
00:13:26,450 --> 00:13:28,409
那么谁掌握了更大的算力

324
00:13:28,409 --> 00:13:31,909
谁就更有可能在竞争中获得优势

325
00:13:32,559 --> 00:13:34,500
我们看到在agent出来之后

326
00:13:34,500 --> 00:13:37,080
ATHROPIC的a arr迅速起飞

327
00:13:37,080 --> 00:13:40,700
其实这点就连AROPIC自己都没有料到

328
00:13:40,700 --> 00:13:42,639
现在他们明显已经发现

329
00:13:42,639 --> 00:13:45,039
自己的算力开始跟不上了

330
00:13:45,039 --> 00:13:47,100
ANTHROPIC一直是三大模型中

331
00:13:47,100 --> 00:13:49,460
算力投入上最为克制的那个

332
00:13:49,460 --> 00:13:52,679
但是却最先跑通了商业模式

333
00:13:52,679 --> 00:13:54,019
尴尬的就在这

334
00:13:54,019 --> 00:13:55,840
现在他们再想去补算力

335
00:13:55,840 --> 00:13:57,659
却发现买不到了

336
00:13:57,659 --> 00:13:59,000
即便就是想买

337
00:13:59,000 --> 00:14:01,318
价格也比之前贵了一大截

338
00:14:01,318 --> 00:14:02,839
那怎么办呢

339
00:14:02,839 --> 00:14:06,499
我们看到ANTHROPIC开始给用户变相的涨价

340
00:14:06,499 --> 00:14:10,059
同样的价格他们降低了可使用的token

341
00:14:10,059 --> 00:14:13,779
也就是说相同的订阅能干的事变得越来越少了

342
00:14:13,779 --> 00:14:15,960
这自然会影响到用户体验

343
00:14:18,059 --> 00:14:20,919
已经有很多程序员因为抱怨cloud的算力不足

344
00:14:20,919 --> 00:14:23,009
去投奔codex了

345
00:14:23,330 --> 00:14:27,570
相反OpenAI以前一直为人所诟病的到处屯算力

346
00:14:27,570 --> 00:14:30,230
现在恐怕要成为他的优势了

347
00:14:30,230 --> 00:14:31,649
根据公开数据总结

348
00:14:31,649 --> 00:14:35,279
OpenAI未来几年已经锁定了超过30几瓦的算力

349
00:14:35,279 --> 00:14:38,139
而ANTHROPIC现在锁定的就只有6G瓦

350
00:14:38,139 --> 00:14:39,899
加上那些未确定的订单

351
00:14:39,899 --> 00:14:42,419
最多也不会超过十几瓦

352
00:14:42,419 --> 00:14:43,879
而更为重要的是

353
00:14:43,879 --> 00:14:46,320
open ai1早就用非常便宜的价格

354
00:14:46,320 --> 00:14:47,839
囤了一大堆的算力

355
00:14:47,839 --> 00:14:52,039
而这是ATHROPIC现在怎么也实现不了的了

356
00:14:52,360 --> 00:14:53,279
讽刺的是

357
00:14:53,279 --> 00:14:56,039
这些算力当年被open ai1顿瞎撒网

358
00:14:56,039 --> 00:14:56,779
搞SORA

359
00:14:56,779 --> 00:14:57,440
搞science

360
00:14:57,440 --> 00:14:58,480
搞成人模式

361
00:14:58,480 --> 00:15:00,568
最终啊都无疾而终

362
00:15:00,568 --> 00:15:01,708
人们嘲笑他

363
00:15:01,708 --> 00:15:02,408
质疑他

364
00:15:02,408 --> 00:15:06,068
然而现在这些算力全部都集中在了codex上

365
00:15:06,068 --> 00:15:08,250
反而助推了它的起飞

366
00:15:08,250 --> 00:15:10,929
在A阵的应用差距逐渐缩小的背景下

367
00:15:10,929 --> 00:15:11,970
充足的算力

368
00:15:11,970 --> 00:15:16,409
未来还有望继续为codex撬动它的竞争优势

369
00:15:16,769 --> 00:15:18,309
真是世事难料啊

370
00:15:18,309 --> 00:15:21,070
花钱最克制的人却最先跑通了变现

371
00:15:21,070 --> 00:15:22,669
而花钱最疯狂的人

372
00:15:22,669 --> 00:15:25,480
反而最有可能成为最后的赢家

373
00:15:25,480 --> 00:15:26,200
当然了

374
00:15:26,200 --> 00:15:28,100
现在下结论肯定还为时尚早

375
00:15:28,100 --> 00:15:29,120
你也别真较真

376
00:15:29,120 --> 00:15:31,759
我就是单纯感慨一下

377
00:15:32,149 --> 00:15:33,470
虽然最终谁输谁赢

378
00:15:33,470 --> 00:15:34,330
现在还不确定

379
00:15:34,330 --> 00:15:36,129
但是就当前的情况来看

380
00:15:36,129 --> 00:15:40,029
竞争格局似乎很快就要发生彻底的改变了

381
00:15:40,029 --> 00:15:42,990
在现在所有人都看清了变现方向的时刻

382
00:15:42,990 --> 00:15:45,110
在模型能力和agent的应用能力

383
00:15:45,110 --> 00:15:47,019
都很难拉开差距的时刻

384
00:15:47,019 --> 00:15:50,620
现阶段决定胜负的关键就在于算力

385
00:15:50,620 --> 00:15:52,000
谁算力充足

386
00:15:52,000 --> 00:15:55,090
谁就更可能在竞争中获得优势

387
00:15:55,090 --> 00:15:55,950
注意啊

388
00:15:55,950 --> 00:15:57,529
在这里一直强调的是

389
00:15:57,529 --> 00:16:00,909
现阶段未来的长期肯定还会有更多的变量

390
00:16:00,909 --> 00:16:03,750
这个不在咱们现在的讨论范围内

391
00:16:04,110 --> 00:16:05,190
鉴于这个原因

392
00:16:05,190 --> 00:16:06,690
我认为一直被质疑

393
00:16:06,690 --> 00:16:09,730
被打压的OpenAI很有可能会后来居上

394
00:16:09,730 --> 00:16:12,909
在agent的时代重新夺回属于他的优势

395
00:16:12,909 --> 00:16:15,070
或许是OpenAI l r的提升

396
00:16:15,070 --> 00:16:17,690
又或者是codex用户数的提升

397
00:16:17,690 --> 00:16:21,279
总之这个节点应该很快就要到来了

398
00:16:21,279 --> 00:16:22,840
但是根据我的观察

399
00:16:22,840 --> 00:16:25,828
现在的市场还没有完全意识到这一点

400
00:16:25,828 --> 00:16:28,908
而它还将会给咱们投资者带来很多

401
00:16:28,908 --> 00:16:30,428
全新的投资机会

402
00:16:30,428 --> 00:16:34,188
也将消除很多市场原本的投资风险

403
00:16:38,659 --> 00:16:40,360
想象一下怎么赚钱

404
00:16:40,360 --> 00:16:41,980
确定了决定胜负的关键

405
00:16:41,980 --> 00:16:42,860
也确定了

406
00:16:42,860 --> 00:16:46,179
那AI企业们会怎么做呢

407
00:16:46,580 --> 00:16:47,799
首先我认为

408
00:16:47,799 --> 00:16:51,700
新一轮的囤算力军备竞赛很可能即将要到来

409
00:16:51,700 --> 00:16:55,799
这对于基础层的公司又将是一轮新的增长动能

410
00:16:55,799 --> 00:16:56,919
而另一方面

411
00:16:56,919 --> 00:17:00,500
这也能够系统性地降低资本性投入的风险

412
00:17:00,500 --> 00:17:02,919
市场不再会因为投入而感到恐惧

413
00:17:02,919 --> 00:17:06,078
反而可能会更加担心投入不足的问题

414
00:17:06,078 --> 00:17:08,999
这些对于资本性开支巨大的大科技而言

415
00:17:08,999 --> 00:17:11,430
都将会是结构性的利好

416
00:17:11,430 --> 00:17:13,970
另一个更为重要的趋势是

417
00:17:13,970 --> 00:17:15,369
解决算力瓶颈

418
00:17:15,369 --> 00:17:18,230
将会成为接下来整个AI产业链上

419
00:17:18,230 --> 00:17:20,589
全部公司的重中之重

420
00:17:20,589 --> 00:17:21,569
算力瓶颈

421
00:17:21,569 --> 00:17:24,509
不再仅仅是基础层公司的一个增长动能了

422
00:17:24,509 --> 00:17:29,079
而是一个需要全体AI参与者解决的巨大商机

423
00:17:29,079 --> 00:17:32,240
那接下来我们就捋着整个AI产业链去看看

424
00:17:32,240 --> 00:17:36,309
哪些公司更可能获得这个商机下的alpha

425
00:17:36,630 --> 00:17:37,970
在基础层公司中

426
00:17:37,970 --> 00:17:40,890
我认为ASIC自研芯片是一个好的方向

427
00:17:40,890 --> 00:17:41,829
它成本低

428
00:17:41,829 --> 00:17:42,470
效率高

429
00:17:42,470 --> 00:17:45,079
能够有效的缓解通用算力的瓶颈

430
00:17:45,079 --> 00:17:46,900
相关公司包括博通

431
00:17:46,900 --> 00:17:47,460
Marvel

432
00:17:47,460 --> 00:17:48,220
谷歌

433
00:17:48,220 --> 00:17:49,660
同时咱也不能忽视

434
00:17:49,660 --> 00:17:52,059
英伟达下一代ruin芯片的迭代

435
00:17:52,059 --> 00:17:55,099
也将会是突破算力瓶颈的一个关键

436
00:17:55,730 --> 00:17:58,230
再往上一层是模型层面

437
00:17:58,230 --> 00:17:59,450
性能的重要性

438
00:17:59,450 --> 00:18:00,730
现在会逐渐降低

439
00:18:00,730 --> 00:18:03,369
算力的重要性会逐渐提高

440
00:18:03,369 --> 00:18:04,890
哪怕就是性能较差

441
00:18:04,890 --> 00:18:06,809
但是能够提供更多算力的公司

442
00:18:06,809 --> 00:18:09,319
也将会受到市场额外的青睐

443
00:18:09,319 --> 00:18:12,420
同时在算法层面解决算力瓶颈

444
00:18:12,420 --> 00:18:15,950
也将会成为很重要的一个模型迭代的方向

445
00:18:15,950 --> 00:18:17,230
而在这两部分

446
00:18:17,230 --> 00:18:19,289
中国的模型都做得相当不错

447
00:18:19,289 --> 00:18:23,338
嗯再往上一层便是云计算公司

448
00:18:23,338 --> 00:18:27,619
云计算是我认为最先能够被市场认可的大方向

449
00:18:27,619 --> 00:18:29,259
因为它离变线端最近

450
00:18:29,259 --> 00:18:31,509
而且此前还遭受了不少的压力

451
00:18:31,509 --> 00:18:32,589
一旦估值修复

452
00:18:32,589 --> 00:18:34,109
再加上AI变现的东风

453
00:18:34,109 --> 00:18:37,029
能够给该行业带来不小的增长动能

454
00:18:37,029 --> 00:18:38,690
而所有的云计算公司

455
00:18:38,690 --> 00:18:41,049
我认为都能获得不俗的收益

456
00:18:41,049 --> 00:18:44,549
而现阶段我个人会非常看好甲骨文

457
00:18:44,549 --> 00:18:48,289
甲骨文是这轮资本性开支惩罚中承压最大的

458
00:18:48,289 --> 00:18:49,029
同时

459
00:18:49,029 --> 00:18:52,710
他也是和坏孩子OpenAI绑定最深的公司之一

460
00:18:52,710 --> 00:18:56,670
也因此公司的股价遭受到了前所未有的压力

461
00:18:56,670 --> 00:18:58,430
和之前那些基础层的公司

462
00:18:58,430 --> 00:19:00,029
已经普遍表现优异不同

463
00:19:00,029 --> 00:19:02,929
甲骨文现在的股价都仍趴在地上

464
00:19:02,929 --> 00:19:04,449
然而压力越大

465
00:19:04,449 --> 00:19:05,638
弹性就越强

466
00:19:05,638 --> 00:19:07,078
随着AI变现的验证

467
00:19:07,078 --> 00:19:09,179
以及OpenAI竞争优势的回归

468
00:19:09,179 --> 00:19:10,499
双重压力的释放

469
00:19:10,499 --> 00:19:13,799
或许能够给公司带来更高的alpha

470
00:19:14,160 --> 00:19:15,779
咱们再往上倒一层

471
00:19:15,779 --> 00:19:17,400
就是软件层

472
00:19:17,400 --> 00:19:19,119
如果要解决算力瓶颈

473
00:19:19,119 --> 00:19:21,969
软件其实是至关重要的一环

474
00:19:21,969 --> 00:19:23,048
在AI时代

475
00:19:23,048 --> 00:19:24,969
软件公司的优势不在于技术

476
00:19:24,969 --> 00:19:27,750
而在于他们定义需求的能力

477
00:19:27,750 --> 00:19:31,150
现在agent的执行任务之所以会消耗那么多的token

478
00:19:31,150 --> 00:19:33,630
是因为很多任务都得从零开始跑

479
00:19:33,630 --> 00:19:36,599
需要不断的去打磨才能完成最终的交付

480
00:19:36,599 --> 00:19:39,429
这个过程就需要消耗大量的token

481
00:19:39,429 --> 00:19:41,368
而一旦流程已经固定下来

482
00:19:41,368 --> 00:19:43,509
下一次再执行相同任务的时候

483
00:19:43,509 --> 00:19:45,720
token的消耗就会大幅降低

484
00:19:45,720 --> 00:19:47,799
而软件公司最擅长的就是

485
00:19:47,799 --> 00:19:50,019
根据应用场景来定义流程

486
00:19:50,019 --> 00:19:50,940
理论上讲

487
00:19:50,940 --> 00:19:53,859
软件公司就非常有机会撬动这个优势

488
00:19:53,859 --> 00:19:56,769
为用户提供更高效的算力使用

489
00:19:56,769 --> 00:19:59,410
而且他们也能够一定程度上解决

490
00:19:59,410 --> 00:20:02,589
非AI专业人士应用门槛的问题

491
00:20:03,069 --> 00:20:05,230
由于这部分scale实在是太大

492
00:20:05,230 --> 00:20:06,490
牵扯的公司也众多

493
00:20:06,490 --> 00:20:07,990
一句两句肯定说不清

494
00:20:07,990 --> 00:20:11,130
我之后会专门做视频跟大家详细分享

495
00:20:11,130 --> 00:20:13,890
如果你不满足于等着美投君跟大家分享标的

496
00:20:13,890 --> 00:20:16,180
想要自己筛选看好的软件股的话

497
00:20:16,180 --> 00:20:17,740
那么在我的美图pro上

498
00:20:17,740 --> 00:20:20,579
我给大家分享了这么一套AI时代软件股

499
00:20:20,579 --> 00:20:21,240
谁会受益

500
00:20:21,240 --> 00:20:23,150
谁会被颠覆的判断方法

501
00:20:23,150 --> 00:20:26,848
感兴趣的看官欢迎移步查看好了

502
00:20:26,848 --> 00:20:28,588
关于AI产业的全新变化

503
00:20:28,588 --> 00:20:30,489
就跟各位看官都分析完了

504
00:20:30,489 --> 00:20:32,288
如果你喜欢美投君的分析

505
00:20:32,288 --> 00:20:34,138
别忘了订阅本频道

506
00:20:34,138 --> 00:20:36,538
频道内除了这样的深度分析视频之外

507
00:20:36,538 --> 00:20:38,439
每天我还会在YOUTUBE社区中

508
00:20:38,439 --> 00:20:41,099
分享大量的投资机会和投资观点

509
00:20:41,099 --> 00:20:42,619
你只有订阅了本频道

510
00:20:42,619 --> 00:20:46,069
才能不错过这些最关键的投资信息

511
00:20:46,630 --> 00:20:48,849
如果你还想了解更多股票研究

512
00:20:48,849 --> 00:20:50,910
也欢迎来我的美图pro

513
00:20:50,910 --> 00:20:54,549
美图pro是我专门为咱普通美股投资者打造的

514
00:20:54,549 --> 00:20:56,549
专业股票研究产品

515
00:20:56,549 --> 00:20:58,430
今天给出的所有机会中

516
00:20:58,430 --> 00:21:01,190
云计算是我认为最先会出机会的

517
00:21:01,190 --> 00:21:04,200
也是最适合咱们普通散户去布局的

518
00:21:04,200 --> 00:21:05,640
四大云计算公司

519
00:21:05,640 --> 00:21:06,339
亚马逊

520
00:21:06,339 --> 00:21:06,799
微软

521
00:21:06,799 --> 00:21:07,240
谷歌

522
00:21:07,240 --> 00:21:07,880
甲骨文

523
00:21:07,880 --> 00:21:09,160
关于他们的基本面

524
00:21:09,160 --> 00:21:11,500
风险以及估值等各种重要信息

525
00:21:11,500 --> 00:21:13,990
在美特pro上我都有详尽的分析

526
00:21:14,059 --> 00:21:16,180
其中亚马逊微软和谷歌

527
00:21:16,180 --> 00:21:17,980
最近这周刚刚都出了财报

528
00:21:17,980 --> 00:21:21,220
下周我也会陆续更新他们最新的分析

529
00:21:21,220 --> 00:21:23,480
如果你持有这几家云计算公司

530
00:21:23,480 --> 00:21:25,079
或者打算投资他们

531
00:21:25,079 --> 00:21:29,109
那么这些专业且易懂的分析你一定不能错过

532
00:21:29,109 --> 00:21:30,130
感兴趣的看官

533
00:21:30,130 --> 00:21:33,069
欢迎订阅美特pro新用户前七天免费试用

534
00:21:33,069 --> 00:21:34,869
快来体验一下吧

535
00:21:35,440 --> 00:21:37,099
如果你喜欢美投君的视频

536
00:21:37,099 --> 00:21:38,960
别忘了点赞关注和转发哦

537
00:21:38,960 --> 00:21:40,240
谢谢大家

共 20 条，按点赞数排序。

共 24 条，按视频时间排序。