1
00:00:02,480 --> 00:00:03,560
Speaker 1: As media.

2
00:00:04,640 --> 00:00:20,160
Speaker 2: Hello you, it's better offline. I'm your host ed Zeitron. Now,

3
00:00:20,160 --> 00:00:22,000
Speaker 2: I've talked about the pay all horses of the AI

4
00:00:22,079 --> 00:00:24,119
Speaker 2: apocalypse in the past, and these are the events that

5
00:00:24,160 --> 00:00:26,680
Speaker 2: signify that the general VAI era is coming to an end.

6
00:00:26,920 --> 00:00:29,840
Speaker 2: And the subject to this episode and its second part, well,

7
00:00:29,880 --> 00:00:33,280
Speaker 2: I believe represents the biggest paylest horses of the entire flock.

8
00:00:34,040 --> 00:00:36,680
Speaker 2: So on February twenty first analyst TD Cohen revealed that

9
00:00:36,680 --> 00:00:40,120
Speaker 2: Microsoft had canceled leases and I quote totaling a couple

10
00:00:40,200 --> 00:00:43,159
Speaker 2: hundred megawatts with at least two private data center operators

11
00:00:43,200 --> 00:00:46,879
Speaker 2: across multiple US markets canceled. The report also detailed how

12
00:00:46,920 --> 00:00:50,479
Speaker 2: Microsoft pulled back on converting negotiated and signed statements of

13
00:00:50,560 --> 00:00:54,920
Speaker 2: qualifications sqq's, which are added was precursors to a data

14
00:00:54,920 --> 00:00:57,640
Speaker 2: center lease, so effectively the first step before you really

15
00:00:57,680 --> 00:01:00,320
Speaker 2: agreed to something, well the last step, I guess. Although

16
00:01:00,360 --> 00:01:03,200
Speaker 2: the analyst, which is part of the TD Bank company,

17
00:01:03,480 --> 00:01:06,520
Speaker 2: added it was unclear whether Microsoft might convert these sqqs

18
00:01:06,520 --> 00:01:10,200
Speaker 2: in the future. These generally close close to one hundred

19
00:01:10,240 --> 00:01:13,480
Speaker 2: percent of the time. According to them, canceling these was

20
00:01:13,480 --> 00:01:18,640
Speaker 2: effectively rather unusual. Now TD Cohen also added that Microsoft

21
00:01:18,680 --> 00:01:22,080
Speaker 2: was reallocating a considerable portion of its projected international spend

22
00:01:22,080 --> 00:01:24,759
Speaker 2: to the US, which suggests to DD Cohen that there

23
00:01:24,800 --> 00:01:27,880
Speaker 2: was a material slow down in international leasing for Microsoft,

24
00:01:29,319 --> 00:01:32,720
Speaker 2: But one crucial, teeny tiny part of the report was

25
00:01:32,760 --> 00:01:36,040
Speaker 2: missed by just about everybody. I'm going to read directly

26
00:01:36,080 --> 00:01:37,800
Speaker 2: from the report, and you'll probably be able to tell

27
00:01:37,800 --> 00:01:40,480
Speaker 2: from the tone of my voice which the most pertinent

28
00:01:40,520 --> 00:01:44,520
Speaker 2: part is. Ah. As we highlighted in our recent takeaway

29
00:01:44,560 --> 00:01:48,720
Speaker 2: from PTC, the Pacific Telecommunications Council Conference, we learned via

30
00:01:48,760 --> 00:01:51,400
Speaker 2: our channel checks that Microsoft one walked away from multiple

31
00:01:51,400 --> 00:01:53,920
Speaker 2: one hundred plus make what deals in multiple markets that

32
00:01:53,920 --> 00:01:57,680
Speaker 2: were in early to mid stages and negotiations to let

33
00:01:57,680 --> 00:02:01,640
Speaker 2: one gigawatt of loys on larger footoprint sites expire, and

34
00:02:01,720 --> 00:02:04,200
Speaker 2: three walked away from at least five land parcels that

35
00:02:04,280 --> 00:02:07,800
Speaker 2: it had under contract in multiple Tier one markets. What

36
00:02:07,880 --> 00:02:10,560
Speaker 2: TD Cohen is saying is not just that Microsoft canceled

37
00:02:10,560 --> 00:02:13,919
Speaker 2: some data centers, but the Microsoft also effectively canceled over

38
00:02:13,960 --> 00:02:16,840
Speaker 2: a gigawatt of data center operations on top of the

39
00:02:16,840 --> 00:02:21,160
Speaker 2: previously reported multiple one hundred plus what megawat deals. If

40
00:02:21,160 --> 00:02:24,040
Speaker 2: we add in the land under contract, which is indeterminate.

41
00:02:24,080 --> 00:02:26,240
Speaker 2: Based on what tdcens has said and the deals that

42
00:02:26,280 --> 00:02:29,160
Speaker 2: were in flight, the total capacity likely amounts to even

43
00:02:29,200 --> 00:02:32,480
Speaker 2: more than a gigawat. For some context, data sent to

44
00:02:32,560 --> 00:02:35,360
Speaker 2: Dynamics reported that Microsoft had five gigawats of data center

45
00:02:35,400 --> 00:02:38,360
Speaker 2: capacity in April twenty twenty four, saying that Microsoft had

46
00:02:38,360 --> 00:02:40,400
Speaker 2: also planned to add one gigawat of capacity by the

47
00:02:40,400 --> 00:02:42,760
Speaker 2: October twenty twenty four and another one and a half

48
00:02:42,840 --> 00:02:45,440
Speaker 2: gigawats of capacity by the first half of twenty twenty five.

49
00:02:46,080 --> 00:02:48,520
Speaker 2: Based on this reporting, one can estimate the Microsoft as

50
00:02:48,520 --> 00:02:51,200
Speaker 2: somewhere between six and seven and a half gigawatts of

51
00:02:51,280 --> 00:02:54,399
Speaker 2: capacity at this time. As a result, based on td

52
00:02:54,480 --> 00:02:58,959
Speaker 2: Cohen's analyst analysis, even Microsoft has thro recombination of canceled leases,

53
00:02:59,040 --> 00:03:02,800
Speaker 2: pullbacks and statements of qualifications, cancelations of land parcels, and

54
00:03:02,840 --> 00:03:06,920
Speaker 2: deliberate expiration of letters of intent. That's those lois. They've

55
00:03:06,919 --> 00:03:10,800
Speaker 2: effectively abandoned data center expansion equivalent to over fourteen percent

56
00:03:10,840 --> 00:03:14,880
Speaker 2: of their current capacity. It's completely bloody insane, and this

57
00:03:14,960 --> 00:03:17,120
Speaker 2: story just kind of sat there. It kind of not

58
00:03:17,280 --> 00:03:20,320
Speaker 2: the market's confidence. But I don't know, I'm a lot

59
00:03:20,360 --> 00:03:22,920
Speaker 2: more worried about this than I think people are. And

60
00:03:23,480 --> 00:03:25,320
Speaker 2: really I don't like telling people how to feel, but

61
00:03:25,720 --> 00:03:28,200
Speaker 2: this kind of worries me. Okay, now I've thrown a

62
00:03:28,240 --> 00:03:30,520
Speaker 2: lot of new terminology at you. But before we move on,

63
00:03:30,639 --> 00:03:33,440
Speaker 2: let's explain some terms, because they're essential to understanding why

64
00:03:33,480 --> 00:03:35,880
Speaker 2: this is all such a big deal. First of all,

65
00:03:36,200 --> 00:03:39,560
Speaker 2: letter of intent or LOI in this context is a

66
00:03:39,600 --> 00:03:42,280
Speaker 2: statement that an entity intends to lease or buy land

67
00:03:42,360 --> 00:03:44,840
Speaker 2: or power from a data center. These can be binding

68
00:03:44,840 --> 00:03:47,800
Speaker 2: on non binding. A letter of intent is serious, though,

69
00:03:47,880 --> 00:03:50,160
Speaker 2: and walking away from one is not something you do idly.

70
00:03:50,200 --> 00:03:53,240
Speaker 2: It's not like not answering an email. Now we'll go

71
00:03:53,280 --> 00:03:56,480
Speaker 2: on to SQQS statements of qualifications. These set the terms

72
00:03:56,520 --> 00:03:59,920
Speaker 2: and conditions of a lease. While they do not themselves constitutally,

73
00:04:00,120 --> 00:04:02,480
Speaker 2: they convert into sign leases, as I mentioned, at an

74
00:04:02,480 --> 00:04:05,280
Speaker 2: almost one hundred percent rate according to TD Cohen, and

75
00:04:05,320 --> 00:04:07,480
Speaker 2: are generally used as a signal to the landowner to

76
00:04:07,520 --> 00:04:11,520
Speaker 2: start construction. Basically, they're the green light before the green lane.

77
00:04:12,280 --> 00:04:15,360
Speaker 2: As for Tier one markets, these are markets for hyperscale growth,

78
00:04:15,440 --> 00:04:18,760
Speaker 2: helped by favorable conditions like power, land and cabling. From

79
00:04:18,800 --> 00:04:20,760
Speaker 2: what I can tell, there's no fixed list of which

80
00:04:20,800 --> 00:04:23,120
Speaker 2: cities are Tier one in which aren't, but they include

81
00:04:23,120 --> 00:04:26,400
Speaker 2: obvious candidates like London, Singapore, as well as Northern Virginia,

82
00:04:26,400 --> 00:04:29,480
Speaker 2: which is the largest tube of data centers in the world. Finally,

83
00:04:29,600 --> 00:04:33,640
Speaker 2: we're talking power, megawatt and gigawe. This one is really important,

84
00:04:33,640 --> 00:04:37,719
Speaker 2: but it's also really confusing. Data center capacity is measured

85
00:04:37,760 --> 00:04:40,400
Speaker 2: not by the amount of computations the facility can handle,

86
00:04:40,440 --> 00:04:43,160
Speaker 2: but rather by power capacity, and that makes sense because

87
00:04:43,200 --> 00:04:46,000
Speaker 2: power capacity is directly linked to the capabilities of the facility,

88
00:04:46,160 --> 00:04:49,080
Speaker 2: with more power capacity allowing for more servers or more

89
00:04:49,080 --> 00:04:52,120
Speaker 2: power hungry chips, of course, and because chips themselves are

90
00:04:52,120 --> 00:04:55,120
Speaker 2: constantly getting faster and more power efficient. Power makes a

91
00:04:55,160 --> 00:04:58,320
Speaker 2: little more sense. If you mentioned in terms of computations

92
00:04:58,320 --> 00:04:59,920
Speaker 2: per second, you'd likely have a number that fluck you

93
00:05:00,240 --> 00:05:03,320
Speaker 2: as hardware as upgraded and decommissioned. When you hear megawat

94
00:05:03,360 --> 00:05:05,559
Speaker 2: or gigawatt in this episode, assume that we're talking about

95
00:05:05,560 --> 00:05:09,680
Speaker 2: capacity and not power generation unless I say otherwise. Now,

96
00:05:09,680 --> 00:05:11,720
Speaker 2: with that out of the way, let's talk more about

97
00:05:11,720 --> 00:05:12,240
Speaker 2: this report.

98
00:05:24,200 --> 00:05:24,360
Speaker 1: Now.

99
00:05:24,360 --> 00:05:26,719
Speaker 2: The numbers in the TD Cohen Report, which I'll link

100
00:05:26,760 --> 00:05:29,960
Speaker 2: in the episode spreadsheet, of course, heavily suggest that Microsoft,

101
00:05:30,000 --> 00:05:33,280
Speaker 2: the biggest purchaser of Nvidio GPUs and, according to Ted

102
00:05:33,440 --> 00:05:36,159
Speaker 2: Cohen and I quote, the most active data center LESE

103
00:05:36,320 --> 00:05:38,640
Speaker 2: of capacity in twenty twenty three and the first half

104
00:05:38,680 --> 00:05:41,440
Speaker 2: of twenty twenty four, does not believe that there's future

105
00:05:41,640 --> 00:05:44,560
Speaker 2: growth in generative AI, nor does it have faith in,

106
00:05:44,680 --> 00:05:47,680
Speaker 2: nor does it want responsibility for the future of open AI.

107
00:05:48,480 --> 00:05:51,240
Speaker 2: Data center buildouts take three to six years to complete,

108
00:05:51,279 --> 00:05:54,960
Speaker 2: and the largest hyperscalar facilities can easily cost several billion dollars,

109
00:05:55,120 --> 00:05:58,320
Speaker 2: meaning that these moves are extremely forward looking. You don't

110
00:05:58,360 --> 00:06:00,440
Speaker 2: just build a data center for the demand GI now,

111
00:06:00,480 --> 00:06:02,640
Speaker 2: but for the demand you expect further down the line.

112
00:06:03,040 --> 00:06:05,920
Speaker 2: This suggests that Microsoft believes its current infrastructure and its

113
00:06:06,000 --> 00:06:08,960
Speaker 2: likely scaled back plans for expansion will be sufficient for

114
00:06:09,080 --> 00:06:12,480
Speaker 2: a movement that Sachynadella once called a golden age for systems.

115
00:06:12,640 --> 00:06:15,160
Speaker 2: He did that less than a year ago. To quote

116
00:06:15,160 --> 00:06:18,520
Speaker 2: TD Cohen again, the magnitude of both potential data center

117
00:06:18,560 --> 00:06:21,280
Speaker 2: capacity Microsoft walked away from the decision to pull back

118
00:06:21,279 --> 00:06:25,320
Speaker 2: on land acquisition, which supports core long term capacity growth,

119
00:06:25,360 --> 00:06:27,560
Speaker 2: in our view, indicates the loss of a major demand

120
00:06:27,560 --> 00:06:30,359
Speaker 2: signal that Microsoft was originally responding to, and that we

121
00:06:30,440 --> 00:06:33,320
Speaker 2: believe the shift is in their appetite for capacity is

122
00:06:33,360 --> 00:06:36,400
Speaker 2: tied to open AI. To explain here, TD Cowen is

123
00:06:36,400 --> 00:06:39,800
Speaker 2: effectively saying that Microsoft is responding to a major demand signal,

124
00:06:39,880 --> 00:06:42,240
Speaker 2: and said major demand signal is saying you do not

125
00:06:42,400 --> 00:06:45,919
Speaker 2: need more data centers. Said demand signal that Microsoft was

126
00:06:46,160 --> 00:06:48,880
Speaker 2: responding to, in TD Cohen's words, is its appetite for

127
00:06:48,920 --> 00:06:52,320
Speaker 2: capacity to provide servers to open AI. And it seems

128
00:06:52,320 --> 00:06:54,720
Speaker 2: that said appetite is waning and Microsoft no longer wants

129
00:06:54,760 --> 00:06:58,160
Speaker 2: to build out data centers for America's most swagged out

130
00:06:58,200 --> 00:07:00,719
Speaker 2: AI guy. Now, I say that kind of as a joke,

131
00:07:01,160 --> 00:07:03,440
Speaker 2: and I do think that he's more swagged out than

132
00:07:03,440 --> 00:07:06,919
Speaker 2: Mark Zuckerberg. Mark Zuckerberg's trying too hard. However, Sam Moltman

133
00:07:07,040 --> 00:07:10,600
Speaker 2: is more damp than Mark Zuckerberg, so ultimately Zuckerberg wins. Now,

134
00:07:10,600 --> 00:07:12,560
Speaker 2: I want to make it clear that Microsoft is effectively

135
00:07:12,600 --> 00:07:15,760
Speaker 2: cutting its data center expansion by over a gigawatt of capacity,

136
00:07:15,840 --> 00:07:18,920
Speaker 2: if not more, and it's impossible to reconcile these cuts

137
00:07:19,040 --> 00:07:21,920
Speaker 2: with the expectation that generative AI will be this massive,

138
00:07:22,000 --> 00:07:26,240
Speaker 2: transformative technological phenomenon. I believe that the reason that Microsoft

139
00:07:26,280 --> 00:07:28,240
Speaker 2: is coming back is that it does not have the

140
00:07:28,280 --> 00:07:31,120
Speaker 2: appetite to provide further data center expansion for open Ai,

141
00:07:31,360 --> 00:07:34,120
Speaker 2: and it's having doubts about the future of generative AI

142
00:07:34,160 --> 00:07:37,120
Speaker 2: as a whole. If Microsoft believed that there was a

143
00:07:37,160 --> 00:07:40,440
Speaker 2: massive opportunity in supporting open AI's further growth, or that

144
00:07:40,520 --> 00:07:43,800
Speaker 2: it had massive demand for generative AI services, there'd be

145
00:07:43,840 --> 00:07:46,640
Speaker 2: no reason to cancel capacity, let alone cats or such

146
00:07:46,680 --> 00:07:50,440
Speaker 2: a significant amount. These moves also suggest that Microsoft is

147
00:07:50,480 --> 00:07:53,560
Speaker 2: walking away from building and training further large frontier models

148
00:07:53,800 --> 00:07:57,320
Speaker 2: like chat GPT's GPT four point five now and from

149
00:07:57,320 --> 00:08:01,640
Speaker 2: supporting doing so for others. Remember, Microsoft has significantly more

150
00:08:01,680 --> 00:08:04,240
Speaker 2: insight into the current health and growth of GENERATIVEAI than

151
00:08:04,280 --> 00:08:07,880
Speaker 2: any other company. Remember they have full access to all

152
00:08:07,920 --> 00:08:10,880
Speaker 2: of open AI's tech, probably the future stuff too, not

153
00:08:10,920 --> 00:08:13,840
Speaker 2: that there's much they know. All their research too, Microsoft

154
00:08:13,880 --> 00:08:17,080
Speaker 2: knows something we don't. As open AI's largest backer and

155
00:08:17,120 --> 00:08:19,920
Speaker 2: infrastructural partner and the owners of the server architecture where

156
00:08:19,960 --> 00:08:22,600
Speaker 2: they train them ultra expensive models, not to mention, the

157
00:08:22,680 --> 00:08:26,280
Speaker 2: largest shareholder in OpenAI, Microsoft can see exactly what is

158
00:08:26,400 --> 00:08:28,560
Speaker 2: or isn't coming down the pike on top of having

159
00:08:28,600 --> 00:08:30,920
Speaker 2: a view into both the sales of its own GENERATIVEAI

160
00:08:31,000 --> 00:08:34,280
Speaker 2: powered software such as Microsoft three sixty five copiler, and

161
00:08:34,320 --> 00:08:36,800
Speaker 2: sales of both model services and cloud compute for other

162
00:08:36,840 --> 00:08:39,400
Speaker 2: models run on Microsoft A zero, which is their cloud

163
00:08:39,440 --> 00:08:43,640
Speaker 2: platform in plain English, Microsoft, which arguably has more data

164
00:08:43,679 --> 00:08:45,800
Speaker 2: than anybody else about the health of the generative AI

165
00:08:45,840 --> 00:08:48,160
Speaker 2: industry and its potential for growth, has decided that it

166
00:08:48,200 --> 00:08:50,239
Speaker 2: needs to dramatically slow down its expansion.

167
00:08:51,440 --> 00:08:51,520
Speaker 1: Now.

168
00:08:51,600 --> 00:08:54,760
Speaker 2: To be clear, this expansion, I really am hammering this

169
00:08:54,800 --> 00:08:56,360
Speaker 2: home a lot, but I need you to understand this

170
00:08:56,720 --> 00:09:00,600
Speaker 2: is absolutely necessary for generative AI to continue evolving expanding,

171
00:09:00,679 --> 00:09:02,600
Speaker 2: even if it only does so in ways that kind

172
00:09:02,600 --> 00:09:05,480
Speaker 2: of do the same thing again and again. Now, before

173
00:09:05,520 --> 00:09:06,960
Speaker 2: we move on, I want to make it clear that

174
00:09:07,000 --> 00:09:09,800
Speaker 2: I'm not saying that Microsoft has stopped building data centers.

175
00:09:09,880 --> 00:09:11,800
Speaker 2: I've said it a few times, but these projects take

176
00:09:11,960 --> 00:09:14,720
Speaker 2: years three to six years to complete and are far

177
00:09:15,320 --> 00:09:18,200
Speaker 2: far in advance with their planning, and Microsoft does have

178
00:09:18,240 --> 00:09:20,520
Speaker 2: a few big projects in the works. One plan three

179
00:09:20,640 --> 00:09:23,480
Speaker 2: hundred and twenty four megawatt Microsoft Data center in Atlanta

180
00:09:23,559 --> 00:09:26,000
Speaker 2: is expected to cost one zero point eight billion dollars

181
00:09:26,120 --> 00:09:27,680
Speaker 2: and as far as I know, this deal is still

182
00:09:27,720 --> 00:09:32,520
Speaker 2: in flight, however, and this was cited separately by TD Cohen.

183
00:09:32,840 --> 00:09:35,560
Speaker 2: Microsoft has recently paused construction on parts of its three

184
00:09:35,600 --> 00:09:39,040
Speaker 2: point three billion dollar data center campus in Mount Pleasant, Wisconsin.

185
00:09:39,559 --> 00:09:41,920
Speaker 2: While Microsoft had tried to reassure locals that the first

186
00:09:41,920 --> 00:09:43,760
Speaker 2: phase of the project was on course to be complete

187
00:09:43,800 --> 00:09:46,240
Speaker 2: on time, its justification of delaying the rest of it

188
00:09:46,360 --> 00:09:48,960
Speaker 2: was well not brilliant, and it was to give Microsoft

189
00:09:48,960 --> 00:09:52,280
Speaker 2: an opportunity to evaluate and I quote the project's scope

190
00:09:52,960 --> 00:09:55,720
Speaker 2: and recent changes in technology, and consider how this might

191
00:09:55,800 --> 00:09:59,680
Speaker 2: impact the design of its facilities. Oh, bob, oh, buddy,

192
00:09:59,720 --> 00:10:02,800
Speaker 2: that's not good. You don't want it. No. One evaluates

193
00:10:02,800 --> 00:10:05,640
Speaker 2: the scope and then goes, oh, actually the scope is great.

194
00:10:05,720 --> 00:10:08,760
Speaker 2: I love it. Nor do they think about impacts and

195
00:10:08,800 --> 00:10:12,160
Speaker 2: go oh, let's do more. No, no, no, anyway, the

196
00:10:12,200 --> 00:10:15,000
Speaker 2: same registered article I'm citing here adds that and I

197
00:10:15,120 --> 00:10:17,920
Speaker 2: quote the review process may include the need to negotiate

198
00:10:17,960 --> 00:10:20,520
Speaker 2: some building permits, potentially placing another hurdle in the way

199
00:10:20,520 --> 00:10:23,280
Speaker 2: of the project. The register did add that Microsoft said

200
00:10:23,280 --> 00:10:27,040
Speaker 2: it expected to complete one hyperscaler data center in Mount

201
00:10:27,040 --> 00:10:31,000
Speaker 2: Pleasant as originally planned, though its capacity wasn't available. Arguably,

202
00:10:31,040 --> 00:10:34,880
Speaker 2: Microsoft would expand its data center infrastructure. Anyway, as more

203
00:10:34,920 --> 00:10:37,280
Speaker 2: stuff moves to the cloud and our dependence grows on it,

204
00:10:37,559 --> 00:10:40,640
Speaker 2: Microsoft and other providers need to build capacity. The organic

205
00:10:40,640 --> 00:10:44,679
Speaker 2: growth is natural and sadly inevitable. However, Microsoft's plans for

206
00:10:44,760 --> 00:10:47,480
Speaker 2: data center expansion were far far in excess of that

207
00:10:47,559 --> 00:10:50,280
Speaker 2: natural growth, and perhaps we're seeing a pullback from those

208
00:10:50,400 --> 00:10:54,320
Speaker 2: stated extravagances into something perhaps a little more reasonable.

209
00:10:55,800 --> 00:10:55,920
Speaker 3: Now.

210
00:10:55,960 --> 00:10:58,640
Speaker 2: I've talked a lot about megawards and gigawats, and you're

211
00:10:58,800 --> 00:11:01,120
Speaker 2: not in the data center business, and you should be.

212
00:11:01,160 --> 00:11:03,200
Speaker 2: The parties are an absolute laugh. This can all seem

213
00:11:03,240 --> 00:11:06,800
Speaker 2: a bit abstract, so let's put it into context. Without context,

214
00:11:06,800 --> 00:11:09,800
Speaker 2: it's hard to understand how big and one hundred megawat

215
00:11:09,880 --> 00:11:12,720
Speaker 2: data center is. These are some of the biggest. According

216
00:11:12,720 --> 00:11:16,280
Speaker 2: to the International Energy Agency, Small data centers can consume

217
00:11:16,320 --> 00:11:19,160
Speaker 2: anywhere between one and five megawats. These are, for the

218
00:11:19,200 --> 00:11:23,120
Speaker 2: most part, average size facilities, perhaps not for cloud compute giants,

219
00:11:23,160 --> 00:11:25,160
Speaker 2: but for other companies. It's kind of part of the course.

220
00:11:25,520 --> 00:11:28,880
Speaker 2: One hundred megawats, by comparison, is huge. It's the equivalent

221
00:11:28,920 --> 00:11:31,480
Speaker 2: of the annual energy consumption of between three hundred and

222
00:11:31,480 --> 00:11:34,200
Speaker 2: fifty thousand, four hundred thousand electric cars, and I know

223
00:11:34,280 --> 00:11:35,840
Speaker 2: some sort of pedant it's going to say, it's not

224
00:11:35,960 --> 00:11:38,280
Speaker 2: the same thing, shout out a fuck up. Go outside,

225
00:11:38,320 --> 00:11:42,400
Speaker 2: stop listen, go outside, go outside now, go do something anyway,

226
00:11:42,640 --> 00:11:45,040
Speaker 2: Although there are others that will likely dwarf what we

227
00:11:45,240 --> 00:11:48,000
Speaker 2: today considered to be a large facility, META is in

228
00:11:48,040 --> 00:11:50,559
Speaker 2: the process of constructing, for example, a ten billion dollar

229
00:11:50,679 --> 00:11:55,080
Speaker 2: data center campus in Louisiana with a proposed two gigawat capacity. Still,

230
00:11:55,120 --> 00:11:57,760
Speaker 2: whatever way you cut it, and one hundred megawatt facility

231
00:11:57,800 --> 00:12:00,640
Speaker 2: is big and it's a big long term investment. Cushman

232
00:12:00,679 --> 00:12:03,560
Speaker 2: and Wakefield's twenty twenty four global data center market comparison

233
00:12:03,640 --> 00:12:07,520
Speaker 2: gives some chilling context into how significant Microsoft's pullback is.

234
00:12:08,120 --> 00:12:10,480
Speaker 2: A gigawatt of data set to capacity is roughly the

235
00:12:10,640 --> 00:12:13,480
Speaker 2: entire operational IT load of Tokyo, which has a one

236
00:12:13,480 --> 00:12:17,160
Speaker 2: point to eight gigawatt capacity, or London nine hundred and

237
00:12:17,200 --> 00:12:19,679
Speaker 2: ninety six megawatts, or the Bay Area, which only has

238
00:12:19,679 --> 00:12:23,120
Speaker 2: eight hundred and forty two megawatts. These are actually very large.

239
00:12:23,880 --> 00:12:26,480
Speaker 2: It's just that Microsoft got rid of so much more.

240
00:12:26,960 --> 00:12:29,800
Speaker 2: And again, the total figure of canceled or abandoned capacity

241
00:12:29,840 --> 00:12:32,760
Speaker 2: is likely far higher than a gigawatt. That number only

242
00:12:32,800 --> 00:12:35,640
Speaker 2: accounts for the letters of intent that Microsoft allowed to expire.

243
00:12:35,840 --> 00:12:38,120
Speaker 2: It doesn't include everything else that the two data sentences

244
00:12:38,280 --> 00:12:40,800
Speaker 2: already killed, or the land pass was it abandoned, or

245
00:12:40,840 --> 00:12:43,559
Speaker 2: the deals that were in early to mid stages of negotiation.

246
00:12:44,080 --> 00:12:46,520
Speaker 2: Imagine walking away from two London's or two Tokyos of

247
00:12:46,520 --> 00:12:48,880
Speaker 2: capacity and it not being a massive deal. This is

248
00:12:48,920 --> 00:12:53,000
Speaker 2: a huge flipping deal. Microsoft is not simply walking back

249
00:12:53,040 --> 00:12:56,920
Speaker 2: some future plans. It's effectively canceling what it loudly insisted

250
00:12:57,040 --> 00:13:00,320
Speaker 2: was the future. If you think this sounds hyperbolic, consider this.

251
00:13:00,760 --> 00:13:03,840
Speaker 2: London and Tokyo are respectively the biggest data center markets

252
00:13:03,840 --> 00:13:05,920
Speaker 2: in Europe and Asia, according to the same Cushman and

253
00:13:05,960 --> 00:13:09,439
Speaker 2: Wakefield report. Canceling city's worth of capacity at a time when

254
00:13:09,520 --> 00:13:14,760
Speaker 2: artificial intelligence is supposedly revolutionizing everything certainly suggests that artificial

255
00:13:14,800 --> 00:13:19,520
Speaker 2: intelligence isn't really revolutionizing anything now. One other detail in

256
00:13:19,559 --> 00:13:22,280
Speaker 2: TD Cohen's report really stood out to me. While this

257
00:13:22,440 --> 00:13:25,400
Speaker 2: pullback in Microsoft's data center leasing, It's also seen a

258
00:13:25,440 --> 00:13:29,240
Speaker 2: commensurate rise in demand from Oracle related to the Stargate project,

259
00:13:29,400 --> 00:13:32,760
Speaker 2: a relatively new partnership of up to five hundred billion

260
00:13:32,760 --> 00:13:35,640
Speaker 2: dollars stopped saying it's five hundred billion dollars to build

261
00:13:35,679 --> 00:13:39,120
Speaker 2: massive new data centers for Ai, specifically for one company

262
00:13:39,400 --> 00:13:42,080
Speaker 2: led by soft Bank and of course open Ai with

263
00:13:42,160 --> 00:13:44,640
Speaker 2: investment from Oracle and MGX and one hundred billion dollar

264
00:13:44,720 --> 00:13:48,080
Speaker 2: investment fund backed by the United Arab Emirates. Open Ai

265
00:13:48,120 --> 00:13:51,160
Speaker 2: has committed eighteen to nineteen billion dollars to the Stargate project,

266
00:13:51,720 --> 00:13:55,360
Speaker 2: money it doesn't have, meaning that part of the twenty

267
00:13:55,360 --> 00:13:57,920
Speaker 2: five to forty billion dollars that they're raising at the

268
00:13:57,920 --> 00:14:00,160
Speaker 2: moment will be committed to funding these data centers, and

269
00:14:00,240 --> 00:14:03,080
Speaker 2: less as I'll get to later. Open Ai raises more

270
00:14:03,080 --> 00:14:05,960
Speaker 2: in debt. Leading the round is SoftBank, which is also

271
00:14:06,000 --> 00:14:08,679
Speaker 2: committing eighteen to nineteen billion dollars, as well as creating

272
00:14:08,720 --> 00:14:11,880
Speaker 2: a joint venture fund called sb open Ai Japan to

273
00:14:11,920 --> 00:14:14,920
Speaker 2: offer open Ai services to the Japanese market, something that

274
00:14:14,960 --> 00:14:17,360
Speaker 2: I thought was already happening, as well as spending three

275
00:14:17,400 --> 00:14:20,480
Speaker 2: billion dollars annually to use open AI's technology across its

276
00:14:20,520 --> 00:14:24,000
Speaker 2: group businesses. According to The Wall Street Journal. In simpler terms,

277
00:14:24,040 --> 00:14:26,640
Speaker 2: soft Bank is investing as much as I think it's

278
00:14:26,640 --> 00:14:28,760
Speaker 2: going to be like thirty billion dollars in open Ai,

279
00:14:28,880 --> 00:14:31,760
Speaker 2: then spending another three billion dollars a year on software

280
00:14:31,800 --> 00:14:34,480
Speaker 2: that only loses money and still hallucinates and shows no

281
00:14:34,680 --> 00:14:38,160
Speaker 2: side of getting meaningfully better or more reliable. Well a

282
00:14:38,160 --> 00:14:40,600
Speaker 2: soft bank actually sees value in open AI's tech, or

283
00:14:40,600 --> 00:14:42,720
Speaker 2: whether this purchase dealer is a subsidy by the back

284
00:14:42,760 --> 00:14:45,720
Speaker 2: door is open to debate. Given that three billion dollars

285
00:14:45,800 --> 00:14:48,640
Speaker 2: is equivalent to open AI's entire revenue from selling premium

286
00:14:48,640 --> 00:14:51,520
Speaker 2: access to Chatch GPT in twenty twenty four, which included

287
00:14:51,560 --> 00:14:53,880
Speaker 2: some major deals with the likes of Price Waterhouse Coopers,

288
00:14:54,560 --> 00:14:57,240
Speaker 2: I'm inclined to believe the latter. Even then, how is

289
00:14:57,280 --> 00:15:00,760
Speaker 2: it feasible that SoftBank can continue paying to get the

290
00:15:00,800 --> 00:15:03,760
Speaker 2: deal done? Microsoft changed the terms of its exclusive relationship

291
00:15:03,760 --> 00:15:05,960
Speaker 2: with open Ai to allow it to work with Oracle

292
00:15:05,960 --> 00:15:08,440
Speaker 2: to build out further data centers full of GPUs necessary

293
00:15:08,440 --> 00:15:12,360
Speaker 2: to power open AI's big, shitty, unprofitable and unsustainable models.

294
00:15:12,880 --> 00:15:16,080
Speaker 2: The open Ai Oracle stargate situation was a direct result,

295
00:15:16,240 --> 00:15:19,200
Speaker 2: according to reporting from the Information of open Ai becoming

296
00:15:19,200 --> 00:15:22,240
Speaker 2: frustrated with Microsoft and not providing it with service fast enough,

297
00:15:22,440 --> 00:15:25,520
Speaker 2: including an allotment of three hundred thousand of nvidious GBT

298
00:15:25,640 --> 00:15:27,720
Speaker 2: two hundred chips by the end of twenty twenty five.

299
00:15:29,280 --> 00:15:31,120
Speaker 2: For what it's worth, The Wall Street Journal reports that

300
00:15:31,160 --> 00:15:34,560
Speaker 2: Microsoft was getting increasingly frustrated with open AI's constant demands

301
00:15:34,560 --> 00:15:37,480
Speaker 2: for more compute. The relationship between the two entities had

302
00:15:37,520 --> 00:15:40,600
Speaker 2: start to fray, with both sides feeling kind of aggrieved. This,

303
00:15:40,720 --> 00:15:44,120
Speaker 2: combined with Microsoft's data center pullback, heavily suggests that Microsoft

304
00:15:44,160 --> 00:15:47,160
Speaker 2: is no longer interested in being open AI's infrastructure paypig

305
00:15:47,760 --> 00:15:50,960
Speaker 2: long term, at least after all it was if it was,

306
00:15:51,000 --> 00:15:54,240
Speaker 2: I mean it funded and support open AI's expansion rather

307
00:15:54,280 --> 00:15:56,240
Speaker 2: than doing the literal opposite.

308
00:15:56,280 --> 00:15:56,960
Speaker 3: And you have to.

309
00:15:56,960 --> 00:16:00,360
Speaker 2: Wonder if when that whole non exclusisive thing came along,

310
00:16:00,400 --> 00:16:03,200
Speaker 2: whether Microsoft was kind of like, no, no, you couldn't

311
00:16:03,200 --> 00:16:06,200
Speaker 2: possibly like the papers already at the pen they've already

312
00:16:06,200 --> 00:16:08,640
Speaker 2: got a stamp Ford sad woman's signature. No, don't sign

313
00:16:08,680 --> 00:16:12,720
Speaker 2: it. It would be so bad. No, I really don't. I

314
00:16:12,760 --> 00:16:29,760
Speaker 2: don't think that Microsoft's too cut up about that. And

315
00:16:29,840 --> 00:16:32,040
Speaker 2: here's a question for you. If general, if AI had

316
00:16:32,080 --> 00:16:35,520
Speaker 2: so much demand, why is Microsoft canceling data centrare contracts.

317
00:16:35,760 --> 00:16:39,120
Speaker 2: Why is Microsoft Oracle's largest customer as at the end

318
00:16:39,160 --> 00:16:41,880
Speaker 2: of twenty twenty three, allowing softmank and open Ai to

319
00:16:41,920 --> 00:16:44,880
Speaker 2: work with Oracle to build the future rather than Microsoft.

320
00:16:46,120 --> 00:16:49,400
Speaker 2: As mentioned previously, TD Cohen specifically noted in its report

321
00:16:49,400 --> 00:16:53,040
Speaker 2: that microsoft shifting appetite for capacity was tied to open Ai, which,

322
00:16:53,080 --> 00:16:55,800
Speaker 2: as I've said already, heavily suggests that Microsoft is at

323
00:16:55,880 --> 00:16:58,720
Speaker 2: best less invested in the future of the company, a

324
00:16:58,760 --> 00:17:01,280
Speaker 2: statement confirmed by the information which adds that Microsoft have

325
00:17:01,320 --> 00:17:03,800
Speaker 2: been trying to and I quote lessen its reliance and

326
00:17:03,840 --> 00:17:07,119
Speaker 2: open ai technology as they increasingly compete in selling AI

327
00:17:07,160 --> 00:17:12,040
Speaker 2: products well. At worse, this situation could suggest that Microsoft

328
00:17:12,119 --> 00:17:14,560
Speaker 2: is actively trying to dump open ai and it's having

329
00:17:14,640 --> 00:17:18,840
Speaker 2: questions about the fundamentals of this industry. Writ large in

330
00:17:19,000 --> 00:17:22,520
Speaker 2: very plain terms, Microsoft, despite its excitement around AI and

331
00:17:22,560 --> 00:17:25,560
Speaker 2: its dogged insistence that it's the future, has canceled data

332
00:17:25,600 --> 00:17:28,040
Speaker 2: center leases of over a gig or what of other

333
00:17:28,320 --> 00:17:31,600
Speaker 2: data center infrastructure. Doing so heavily suggests that they do

334
00:17:31,640 --> 00:17:34,040
Speaker 2: not intend to expand further, or at least to the

335
00:17:34,040 --> 00:17:37,399
Speaker 2: extraordinary levels that they'd initially promised. I know, I'm being

336
00:17:37,480 --> 00:17:40,280
Speaker 2: kind of repetitive. I know that I'm saying some of

337
00:17:40,320 --> 00:17:43,680
Speaker 2: these things repeatedly, and you might think why. I need

338
00:17:43,720 --> 00:17:47,200
Speaker 2: you to understand how significant this is because this did

339
00:17:47,240 --> 00:17:50,440
Speaker 2: not get covered enough. The coverage of this was dogshit.

340
00:17:51,160 --> 00:17:54,880
Speaker 2: I'm saying it, frankly, everyone missed this detail. I am

341
00:17:54,960 --> 00:17:58,040
Speaker 2: just one guy. I do a podcast and a newsletter

342
00:17:58,080 --> 00:18:00,119
Speaker 2: and I run a PR firm. Why am I the

343
00:18:00,280 --> 00:18:04,080
Speaker 2: person every not every time? But it just drives me

344
00:18:04,119 --> 00:18:07,600
Speaker 2: a little insane because this was seeing that how many

345
00:18:07,960 --> 00:18:11,359
Speaker 2: reporters actually read this report too. This is what drives

346
00:18:11,400 --> 00:18:13,480
Speaker 2: me insane with my work. But I really do enjoy

347
00:18:13,480 --> 00:18:16,879
Speaker 2: it was all right. While Microsoft has reiterated that it

348
00:18:16,920 --> 00:18:20,040
Speaker 2: intends to spend a ridiculous eighty billion dollars in capital

349
00:18:20,080 --> 00:18:23,680
Speaker 2: expenditures on AI in twenty twenty five, per CNBC article,

350
00:18:23,840 --> 00:18:25,679
Speaker 2: it's unclear how it intends to do so if it's

351
00:18:25,720 --> 00:18:28,359
Speaker 2: pulling back on data center expansion at such a large scale,

352
00:18:28,480 --> 00:18:31,359
Speaker 2: and the company has provided no tangible explanation or elaboration

353
00:18:31,400 --> 00:18:34,080
Speaker 2: as to how it might do so. While hardware upgrades

354
00:18:34,080 --> 00:18:36,439
Speaker 2: could account for some of those CAPEX, it would be

355
00:18:36,520 --> 00:18:40,320
Speaker 2: nowhere near the eighty billion dollar figure. Again, hyperscale data

356
00:18:40,320 --> 00:18:43,520
Speaker 2: centers aren't cheap. They're massive billion dollar or multi billion

357
00:18:43,520 --> 00:18:47,720
Speaker 2: dollar ventures. Microsoft, according to CNBC, also leases data center

358
00:18:47,720 --> 00:18:50,159
Speaker 2: capacity through core Weave and other providers, though at that

359
00:18:50,200 --> 00:18:52,160
Speaker 2: point the reporter has stopped being curious enough to ask

360
00:18:52,240 --> 00:18:55,760
Speaker 2: how much or who those other providers might be. Now, Luckily,

361
00:18:55,760 --> 00:18:57,960
Speaker 2: with the power of research, I found that the information

362
00:18:58,000 --> 00:19:00,840
Speaker 2: report that Microsoft plants spend about tenion dollars renting core

363
00:19:00,840 --> 00:19:03,400
Speaker 2: Weave service between twenty twenty three and twenty thirty, which

364
00:19:03,440 --> 00:19:08,080
Speaker 2: was also reported by CNBC and otherwise anyway planned past

365
00:19:08,119 --> 00:19:10,440
Speaker 2: tense being the operative word, because last week we learned

366
00:19:10,440 --> 00:19:13,360
Speaker 2: that Microsoft plans to scale back its purchase of capacity

367
00:19:13,359 --> 00:19:16,120
Speaker 2: from core Weave. Core Weave is, as you'll find out

368
00:19:16,119 --> 00:19:19,840
Speaker 2: in a future episode and newsletter probably before that. They're

369
00:19:19,840 --> 00:19:22,159
Speaker 2: a very odd company which is closely tied to another

370
00:19:22,200 --> 00:19:25,560
Speaker 2: company called core Scientific, which it actually rents service from,

371
00:19:25,960 --> 00:19:29,480
Speaker 2: and that company exited Chapter eleven bankruptcy only last year.

372
00:19:30,240 --> 00:19:32,800
Speaker 2: Core Scientific's financial statements, by the way, are very confusing.

373
00:19:32,840 --> 00:19:34,840
Speaker 2: They're a mess, and it mostly makes its money not

374
00:19:34,880 --> 00:19:37,840
Speaker 2: from selling high performance computer services, but from mining bitcoin.

375
00:19:38,320 --> 00:19:40,040
Speaker 2: It's a weird relationship, and it's weird to still that

376
00:19:40,119 --> 00:19:43,159
Speaker 2: Microsoft is even entangled with a company connected to Core Scientific.

377
00:19:43,400 --> 00:19:47,560
Speaker 2: But again, that's a future episode, future newsletter, future panic

378
00:19:47,600 --> 00:19:49,600
Speaker 2: attack that I'll give myself as I read s once

379
00:19:49,640 --> 00:19:52,720
Speaker 2: all day. Moving on, Microsoft also added in a comment

380
00:19:52,760 --> 00:19:54,960
Speaker 2: to CNBC in the same article that it continues to

381
00:19:54,960 --> 00:19:59,119
Speaker 2: grow at a record pace to meet customer demand. Okay,

382
00:19:59,240 --> 00:20:03,919
Speaker 2: excuse me. What customer demand? What customer demand? What is it?

383
00:20:03,960 --> 00:20:07,000
Speaker 2: What is the customer demand? What is going on? Microsoft

384
00:20:07,000 --> 00:20:09,560
Speaker 2: said back in April twenty twenty four that AI demand

385
00:20:09,640 --> 00:20:13,000
Speaker 2: was exceeding supply even after a seventy nine percent surgeon

386
00:20:13,040 --> 00:20:16,919
Speaker 2: capital expenditures, and Cfoamyhood send in their next quarterly earnings

387
00:20:16,920 --> 00:20:19,800
Speaker 2: in July twenty twenty four that demand remained higher than

388
00:20:19,840 --> 00:20:23,680
Speaker 2: microsoft available capacity. On its most recent January twenty twenty

389
00:20:23,680 --> 00:20:27,080
Speaker 2: five earning score, Cfoamyhood once again said, as your growth

390
00:20:27,119 --> 00:20:30,760
Speaker 2: included thirteen points from AI services, which grew one hundred

391
00:20:30,760 --> 00:20:33,159
Speaker 2: and fifty seven percent year of a year and was

392
00:20:33,200 --> 00:20:36,480
Speaker 2: ahead of expectations even as demand continued to be higher

393
00:20:36,520 --> 00:20:39,960
Speaker 2: than our available capacity. Riddle me this, batman, Why does

394
00:20:40,000 --> 00:20:42,760
Speaker 2: a company that keeps talking about having demand that exceeds

395
00:20:42,800 --> 00:20:46,320
Speaker 2: capacity decide to cancel multiple data centers which collectively account

396
00:20:46,400 --> 00:20:50,320
Speaker 2: for a significant chunk of its existing capacity. I don't know.

397
00:20:50,400 --> 00:20:52,320
Speaker 2: Let's see what Microsoft had to say when asked a

398
00:20:52,320 --> 00:20:55,760
Speaker 2: week or two ago. Ahem. Thanks to the significant investments

399
00:20:55,800 --> 00:20:57,600
Speaker 2: we've made up until this point, we are well positioned

400
00:20:57,600 --> 00:21:01,520
Speaker 2: to beat our current and increasing customer demand. Last year alone,

401
00:21:01,560 --> 00:21:03,800
Speaker 2: we added more capacity than any prior year in history.

402
00:21:03,880 --> 00:21:07,320
Speaker 2: What we may strategically pay just our infrastructure in some areas,

403
00:21:07,359 --> 00:21:09,760
Speaker 2: we will continue to grow strongly in all regions. This

404
00:21:09,800 --> 00:21:12,200
Speaker 2: allows us to invest in, allocate resources to grow areas

405
00:21:12,200 --> 00:21:14,720
Speaker 2: and for our future. The fuck are you talking about?

406
00:21:14,840 --> 00:21:17,200
Speaker 2: Sounds like Microsoft built too much capacity and in fact

407
00:21:17,200 --> 00:21:19,200
Speaker 2: as yet to see the customer demand that actually could

408
00:21:19,200 --> 00:21:22,159
Speaker 2: reach it. In fact, a couple weeks ago, Microsoft's CEO

409
00:21:22,240 --> 00:21:24,760
Speaker 2: Saturnadella said in a podcast interview that one of the

410
00:21:24,800 --> 00:21:27,359
Speaker 2: things is that there will be as a result of

411
00:21:27,440 --> 00:21:31,040
Speaker 2: data center expansion related to AI overbuild. Why is the

412
00:21:31,080 --> 00:21:35,960
Speaker 2: CEO of Microsoft saying that if nothing's changed. Microsoft also

413
00:21:36,000 --> 00:21:38,439
Speaker 2: in late January said it had thirteen billion dollars in

414
00:21:38,480 --> 00:21:40,840
Speaker 2: annual recurring revenue from AI. And by the way, that's

415
00:21:41,280 --> 00:21:46,000
Speaker 2: that's revenue. That's revenue. It's revenue, it's not profit and

416
00:21:46,040 --> 00:21:47,920
Speaker 2: On top of that, they don't. It's not like they've

417
00:21:47,920 --> 00:21:51,600
Speaker 2: made thirteen billion dollars. They're multiplying like a month's revenue

418
00:21:51,640 --> 00:21:54,480
Speaker 2: by twelve. And by the way, AI is also not

419
00:21:54,600 --> 00:21:57,360
Speaker 2: a line item on Microsoft's earnings, meaning that all of

420
00:21:57,359 --> 00:22:00,879
Speaker 2: this is just related revenue put into a hatamari and

421
00:22:00,960 --> 00:22:03,359
Speaker 2: rolled around by satching a Dell or picking up chairs

422
00:22:03,359 --> 00:22:05,720
Speaker 2: and shit in the office. Either way, this is a

423
00:22:05,760 --> 00:22:07,880
Speaker 2: piss poor amount that works out to about three point

424
00:22:07,880 --> 00:22:10,840
Speaker 2: twenty five billion dollars a quarter. These are mediocre numbers.

425
00:22:10,840 --> 00:22:13,520
Speaker 2: Their bush league and Microsoft's data center pullback suggests that

426
00:22:13,560 --> 00:22:17,600
Speaker 2: they're not going to improve. But wait, wait, or perhaps

427
00:22:17,680 --> 00:22:20,479
Speaker 2: Microsoft's pullback has something to do with Stargate open aiy's

428
00:22:20,520 --> 00:22:24,760
Speaker 2: big infrastructure project. In fact, I think that might have

429
00:22:24,840 --> 00:22:27,240
Speaker 2: something to do with a great deal. Let's take a look.

430
00:22:27,720 --> 00:22:30,640
Speaker 2: According to the information, open ai plans to have stargate

431
00:22:30,680 --> 00:22:33,280
Speaker 2: handle three quarters of its computing needs by twenty thirty.

432
00:22:33,960 --> 00:22:36,840
Speaker 2: Look they'll make it, which heavily suggests that Microsoft canceling

433
00:22:36,880 --> 00:22:39,240
Speaker 2: so much capacity is on some level linked to open

434
00:22:39,280 --> 00:22:42,520
Speaker 2: AI's future plans and not being part of them. Well,

435
00:22:42,520 --> 00:22:45,200
Speaker 2: the information reports that open ai is still forecasting to

436
00:22:45,240 --> 00:22:47,639
Speaker 2: spend thirteen billion dollars in twenty twenty five and as

437
00:22:47,680 --> 00:22:49,800
Speaker 2: much as twenty eight billion dollars in twenty twenty eight

438
00:22:49,800 --> 00:22:52,560
Speaker 2: on Microsoft's cloud compute in addition to whatever capacity it

439
00:22:52,560 --> 00:22:55,240
Speaker 2: gets from Stargate or Oracle. One has to wonder how

440
00:22:55,280 --> 00:22:57,359
Speaker 2: it intends to do so if it needs capacity that

441
00:22:57,400 --> 00:23:02,359
Speaker 2: Microsoft isn't building and doesn't have. And these are two

442
00:23:02,480 --> 00:23:05,240
Speaker 2: huge numbers, by the way. For context, thirteen billion dollars

443
00:23:05,320 --> 00:23:07,800
Speaker 2: is about ten percent of Microsoft's cloud revenue in the

444
00:23:07,800 --> 00:23:10,720
Speaker 2: twenty twenty four fiscal year, though it's unclear whether Microsoft

445
00:23:10,720 --> 00:23:15,360
Speaker 2: counts open AI's compute spenders its revenue. Nevertheless, I got

446
00:23:15,359 --> 00:23:18,080
Speaker 2: some real concerns about whether open ai is even capable

447
00:23:18,119 --> 00:23:21,280
Speaker 2: of expanding further, starting with a fairly obvious one. Open

448
00:23:21,320 --> 00:23:24,280
Speaker 2: AI's only source of money SoftBank, the world's worst tech

449
00:23:24,320 --> 00:23:27,240
Speaker 2: investor and the only company in the world better incinerating

450
00:23:27,320 --> 00:23:30,760
Speaker 2: huge piles of cash, and open Ai, well, they got

451
00:23:30,800 --> 00:23:33,080
Speaker 2: some money issues that I'll get into, and funnily enough,

452
00:23:33,080 --> 00:23:35,280
Speaker 2: that is going to be the topic of the next episode.

453
00:23:35,560 --> 00:23:38,600
Speaker 2: I'm not convinced that Microsoft's pullback is driven by Stargate

454
00:23:38,760 --> 00:23:41,040
Speaker 2: largly because I don't think Stargate at least with the

455
00:23:41,040 --> 00:23:45,000
Speaker 2: current spending goals and capacity targets is actually viable. The

456
00:23:45,040 --> 00:23:47,520
Speaker 2: real motivations, I believe have far more to do with

457
00:23:47,560 --> 00:23:49,880
Speaker 2: the fact that Microsoft is recognizing that maybe it got

458
00:23:49,920 --> 00:23:53,680
Speaker 2: generative AI and a large part of its future all wrong.

459
00:23:54,480 --> 00:24:04,879
Speaker 2: See you in the next episode. Thank you for listening

460
00:24:04,920 --> 00:24:05,840
Speaker 2: to Better Offline.

461
00:24:05,960 --> 00:24:08,400
Speaker 3: The editor and composer of the Better Offline theme song

462
00:24:08,480 --> 00:24:11,080
Speaker 3: is Matasowski. You can check out more of his music

463
00:24:11,160 --> 00:24:14,800
Speaker 3: and audio projects at Matasowski dot com, M A T

464
00:24:14,800 --> 00:24:19,280
Speaker 3: T O S O W s ki dot com. You

465
00:24:19,280 --> 00:24:21,800
Speaker 3: can email me at easy at Better Offline dot com

466
00:24:21,880 --> 00:24:24,200
Speaker 3: or visit Better Offline dot com to find more podcast

467
00:24:24,240 --> 00:24:27,560
Speaker 3: links and of course my newsletter. I also really recommend

468
00:24:27,600 --> 00:24:29,520
Speaker 3: you go to chat dot Where's youreaed dot at to

469
00:24:29,640 --> 00:24:32,080
Speaker 3: visit the discord, and go to our slash Better.

470
00:24:31,960 --> 00:24:35,120
Speaker 2: Offline to check out our reddit. Thank you so much

471
00:24:35,160 --> 00:24:35,679
Speaker 2: for listening.

472
00:24:36,520 --> 00:24:39,199
Speaker 1: Better Offline is a production of cool Zone Media. For

473
00:24:39,320 --> 00:24:42,480
Speaker 1: more from cool Zone Media, visit our website cool Zonemedia

474
00:24:42,560 --> 00:24:45,400
Speaker 1: dot com, or check us out on the iHeartRadio app,

475
00:24:45,440 --> 00:25:06,679
Speaker 1: Apple Podcasts, or wherever you get your podcasts.

476
00:25:00,880 --> 00:25:04,400
Speaker 2: SPO