1
00:00:03,160 --> 00:00:06,760
Speaker 1: Media. Hello, and welcome to a very special episode of

2
00:00:06,800 --> 00:00:09,440
Speaker 1: Better Offline. I'm, of course, your host ed ze trunk.

3
00:00:17,880 --> 00:00:20,400
Speaker 1: For years, I've been hunting down the core details behind

4
00:00:20,440 --> 00:00:23,159
Speaker 1: open aiyes costs and revenues, and today I'm going to

5
00:00:23,160 --> 00:00:25,159
Speaker 1: bring you some of them. A lot of what I

6
00:00:25,160 --> 00:00:27,120
Speaker 1: say today is going to be reflected in my newsletter,

7
00:00:27,120 --> 00:00:29,840
Speaker 1: which I'll link to in the notes. Based on documents

8
00:00:29,920 --> 00:00:32,839
Speaker 1: viewed by my newsletter, I'm able to report open aiy's

9
00:00:32,840 --> 00:00:35,760
Speaker 1: inference spend on Microsoft asure in addition to its payments

10
00:00:35,800 --> 00:00:38,640
Speaker 1: to Microsoft as part of its twenty percent revenue share agreement,

11
00:00:38,800 --> 00:00:41,480
Speaker 1: which was reported in October twenty twenty four by The Information.

12
00:00:42,159 --> 00:00:44,920
Speaker 1: In simple terms, that last bit means that Microsoft receives

13
00:00:44,960 --> 00:00:48,120
Speaker 1: twenty percent of open eyes revenue in addition to whatever

14
00:00:48,200 --> 00:00:52,159
Speaker 1: it spends on GPUs and servers. As a reminder, influence

15
00:00:52,240 --> 00:00:54,240
Speaker 1: is the process through which a model creates an output,

16
00:00:54,240 --> 00:00:56,480
Speaker 1: which I'll be reminding you of a few times because

17
00:00:56,560 --> 00:01:00,120
Speaker 1: it's actually kind of important. Now a few notes. I

18
00:01:00,160 --> 00:01:02,520
Speaker 1: don't have open AI's training spend, nor do I have

19
00:01:02,600 --> 00:01:05,920
Speaker 1: information on the entire extent of open AI's revenues, as

20
00:01:05,959 --> 00:01:08,760
Speaker 1: it appears that Microsoft shares some percentage of its revenue

21
00:01:08,760 --> 00:01:10,840
Speaker 1: from being as well as twenty percent of the revenue

22
00:01:10,880 --> 00:01:13,600
Speaker 1: Microsoft receives from selling open AI's models on as you.

23
00:01:14,480 --> 00:01:17,720
Speaker 1: What I do have, as I've mentioned, is its inference spend.

24
00:01:18,240 --> 00:01:20,280
Speaker 1: And if you're new to this, like I said, this

25
00:01:20,400 --> 00:01:23,280
Speaker 1: means all the computation's open ai does when processing requests

26
00:01:23,319 --> 00:01:26,640
Speaker 1: sent to its services like chat, GPT, and Sora. Now,

27
00:01:26,640 --> 00:01:29,640
Speaker 1: before publishing, I asked a financial time supporter to help

28
00:01:29,680 --> 00:01:32,080
Speaker 1: corroborate some of the data in the documents. They reached

29
00:01:32,080 --> 00:01:36,160
Speaker 1: out to Microsoft and open Ai, who both declined to comment. Now,

30
00:01:36,200 --> 00:01:37,960
Speaker 1: the following will be a lot of numbers, and it

31
00:01:38,040 --> 00:01:40,520
Speaker 1: might be easier for you to read them. However, I'm

32
00:01:40,520 --> 00:01:42,160
Speaker 1: going to try and make things as easy and clear

33
00:01:42,160 --> 00:01:44,720
Speaker 1: as possible, because the documents I've seen call into question

34
00:01:44,800 --> 00:01:47,240
Speaker 1: what we actually knew about open AI's business and the

35
00:01:47,280 --> 00:01:51,760
Speaker 1: sustainability of said business. To keep things simple, all the

36
00:01:51,800 --> 00:01:55,000
Speaker 1: years in this piece are calendar years. Microsoft has fiscal years.

37
00:01:55,040 --> 00:01:57,000
Speaker 1: I'm not going to play that game. It's impossible to

38
00:01:57,000 --> 00:02:01,200
Speaker 1: follow along with nobody thinks this way. Now we've done that,

39
00:02:01,400 --> 00:02:04,720
Speaker 1: let's get to him. According to the document's view by

40
00:02:04,760 --> 00:02:07,840
Speaker 1: my newsletter, open Ai spent five point oh two billion

41
00:02:07,880 --> 00:02:10,880
Speaker 1: dollars on inference alone with Microsoft Azure in the first

42
00:02:10,880 --> 00:02:14,680
Speaker 1: half of calendar year twenty twenty five. This is a

43
00:02:14,720 --> 00:02:17,040
Speaker 1: pattern that has continued through the end of September twenty

44
00:02:17,080 --> 00:02:19,440
Speaker 1: twenty five, by which point open ai had spent eight

45
00:02:19,480 --> 00:02:23,920
Speaker 1: point six seven billion dollars just on inference. Open AI's

46
00:02:23,919 --> 00:02:27,120
Speaker 1: inference costs have risen consistently over the past eighteen months too.

47
00:02:27,560 --> 00:02:30,560
Speaker 1: For example, open ai spent three point seven six billion

48
00:02:30,600 --> 00:02:33,600
Speaker 1: dollars on inference in twenty twenty four, meaning that open

49
00:02:33,639 --> 00:02:36,240
Speaker 1: ai has already more than doubled its inference costs in

50
00:02:36,360 --> 00:02:39,919
Speaker 1: just the first nine months of twenty twenty five. These

51
00:02:39,960 --> 00:02:43,520
Speaker 1: costs are dramatic and significantly higher than has been previously reported.

52
00:02:44,320 --> 00:02:47,639
Speaker 1: According to the Information, open AI's computer run models, which

53
00:02:47,680 --> 00:02:50,320
Speaker 1: I understand to mean inference, was two billion dollars in

54
00:02:50,320 --> 00:02:54,040
Speaker 1: twenty twenty four. Additionally, an April twenty twenty five piece

55
00:02:54,080 --> 00:02:56,920
Speaker 1: from the Information stated that open AI's inference costs for

56
00:02:56,960 --> 00:03:00,480
Speaker 1: twenty twenty five would be around six billion dollars, or

57
00:03:00,560 --> 00:03:03,200
Speaker 1: roughly two billion dollars less than open ai appears to

58
00:03:03,200 --> 00:03:05,600
Speaker 1: have spent through the end of September. I want to

59
00:03:05,600 --> 00:03:08,000
Speaker 1: be clear as well, I'm just reporting what these documents

60
00:03:08,040 --> 00:03:10,880
Speaker 1: have said. This is not a statement about the information.

61
00:03:10,960 --> 00:03:14,800
Speaker 1: They do great reporting. But then there's the issue of

62
00:03:14,840 --> 00:03:18,440
Speaker 1: the revenue share. As I've previously stated, the following numbers

63
00:03:18,440 --> 00:03:20,640
Speaker 1: are based on the revenue share paid to Microsoft as

64
00:03:20,720 --> 00:03:23,600
Speaker 1: part of its deal with open Ai, where it gives

65
00:03:23,600 --> 00:03:27,040
Speaker 1: Microsoft twenty percent of its revenues. According to the documents,

66
00:03:27,080 --> 00:03:30,200
Speaker 1: Microsoft received four hundred and ninety three point eight million

67
00:03:30,240 --> 00:03:32,720
Speaker 1: dollars in revenue share payments in twenty twenty four from

68
00:03:32,760 --> 00:03:36,320
Speaker 1: open Ai, implying revenues for twenty twenty four for open

69
00:03:36,360 --> 00:03:38,680
Speaker 1: Ai of at least two point four to six nine

70
00:03:38,720 --> 00:03:41,400
Speaker 1: billion dollars, or around one point twenty three billion dollars

71
00:03:41,520 --> 00:03:44,400
Speaker 1: less than the three point seven billion dollar number that's

72
00:03:44,440 --> 00:03:48,160
Speaker 1: been previously reported in multiple outlets. Similarly, for the first

73
00:03:48,200 --> 00:03:51,040
Speaker 1: half of twenty twenty five, Microsoft received four hundred and

74
00:03:51,080 --> 00:03:54,200
Speaker 1: fifty four point seven million dollars as part of its

75
00:03:54,200 --> 00:03:57,240
Speaker 1: revenue share agreement, implying open AI's revenues for that six

76
00:03:57,280 --> 00:03:59,600
Speaker 1: month period or at least two point twenty seven three

77
00:03:59,640 --> 00:04:02,640
Speaker 1: billion dollars or around two billion dollars less than the

78
00:04:02,680 --> 00:04:05,480
Speaker 1: four point three billion dollars previously reported for that period.

79
00:04:06,640 --> 00:04:10,080
Speaker 1: Through September, Microsoft's revenue share payments total eight hundred and

80
00:04:10,120 --> 00:04:13,240
Speaker 1: sixty five point eight million dollars, implying open AI's revenues

81
00:04:13,280 --> 00:04:15,800
Speaker 1: are at least four point three to two nine billion

82
00:04:15,840 --> 00:04:18,120
Speaker 1: dollars through the end of Q three twenty twenty five.

83
00:04:19,080 --> 00:04:22,040
Speaker 1: To be clear, and I'm going to say this, Microsoft

84
00:04:22,120 --> 00:04:24,440
Speaker 1: also pays open Ai a cut of bing's revenues under

85
00:04:24,440 --> 00:04:27,479
Speaker 1: certain circumstances I could not confirm, as well as a

86
00:04:27,520 --> 00:04:30,800
Speaker 1: cut of about twenty percent of all open ai models

87
00:04:30,800 --> 00:04:33,120
Speaker 1: sold through a zore. Just to be clear, Microsoft is

88
00:04:33,160 --> 00:04:35,640
Speaker 1: the only party that can sell open AI's models other

89
00:04:35,680 --> 00:04:38,560
Speaker 1: than open Ai. I don't have the details on those payments,

90
00:04:38,640 --> 00:04:40,480
Speaker 1: like I said, but I am skeptical that they can

91
00:04:40,480 --> 00:04:42,720
Speaker 1: account for the massive difference between those numbers that have

92
00:04:42,760 --> 00:04:45,560
Speaker 1: been leaked and the ones in the documents in question.

93
00:04:46,320 --> 00:04:48,320
Speaker 1: I do not know, nor will I speculate on why

94
00:04:48,360 --> 00:04:51,480
Speaker 1: these differences are so distinct. What was important about today

95
00:04:51,520 --> 00:04:53,240
Speaker 1: was getting you these numbers and shedding light on the

96
00:04:53,279 --> 00:04:56,039
Speaker 1: differences I see between the story told about open ai

97
00:04:56,240 --> 00:04:59,720
Speaker 1: and the reality of its spend and potential revenues. You've

98
00:04:59,720 --> 00:05:01,800
Speaker 1: also I probably noticed that this podcast is a bit

99
00:05:01,839 --> 00:05:05,000
Speaker 1: of a different tone to the usual no insults, no jokes,

100
00:05:05,200 --> 00:05:07,680
Speaker 1: haven't called anyone clammy, haven't even said a swear word

101
00:05:07,720 --> 00:05:10,400
Speaker 1: for the first time in maybe one hundred episodes. The

102
00:05:10,440 --> 00:05:14,400
Speaker 1: reason simple. These numbers are serious and seriously different to

103
00:05:14,440 --> 00:05:18,600
Speaker 1: those reported. Open AI's costs are dramatically higher than previously

104
00:05:18,640 --> 00:05:22,000
Speaker 1: reported in thought, and based on the extrapolations from Microsoft's

105
00:05:22,000 --> 00:05:26,119
Speaker 1: revenue share, its implied revenues are also seemingly dramatically lower

106
00:05:26,160 --> 00:05:29,480
Speaker 1: than we knew. The ramifications of these numbers are severe.

107
00:05:30,120 --> 00:05:33,480
Speaker 1: Open AI's influence costs are incredibly high, absorbing any and

108
00:05:33,520 --> 00:05:36,320
Speaker 1: all revenues and seemingly scaling with every increase in chat

109
00:05:36,400 --> 00:05:40,200
Speaker 1: GPT's user numbers. As revenue goes up, so does their

110
00:05:40,240 --> 00:05:44,240
Speaker 1: inference costs. Conversely, if these implied revenues are indicative of

111
00:05:44,279 --> 00:05:47,880
Speaker 1: the larger financial picture, open ai is not as successful

112
00:05:47,880 --> 00:05:51,479
Speaker 1: as company as we previously believed. In any case, the

113
00:05:51,520 --> 00:05:54,760
Speaker 1: reality of the AI bubble is becoming clearer. Inference, the

114
00:05:54,800 --> 00:05:57,240
Speaker 1: process of creating outputs for a model appears to be

115
00:05:57,279 --> 00:06:00,159
Speaker 1: an incredibly burdensome cost. And if these implied revenue is

116
00:06:00,160 --> 00:06:04,400
Speaker 1: there any indicator the actual business of selling generative AI

117
00:06:04,560 --> 00:06:07,960
Speaker 1: services and models doesn't really seem to be as good

118
00:06:07,960 --> 00:06:11,760
Speaker 1: a business as we thought either. It's all looking a

119
00:06:11,760 --> 00:06:15,720
Speaker 1: little bleak out there. I don't want to editorialize too

120
00:06:15,800 --> 00:06:17,960
Speaker 1: much because I want this information to sit on its

121
00:06:18,000 --> 00:06:22,320
Speaker 1: own own self. But it's it's strange being here. It's

122
00:06:22,360 --> 00:06:25,320
Speaker 1: strange getting these numbers and seeing them myself, and I

123
00:06:25,360 --> 00:06:27,919
Speaker 1: have to wonder how things work out from here. I

124
00:06:27,960 --> 00:06:30,440
Speaker 1: truthfully have no idea, but I do know I'll be

125
00:06:30,440 --> 00:06:32,920
Speaker 1: happy to do this every week, and I will tell

126
00:06:32,960 --> 00:06:36,600
Speaker 1: you what happens now. These numbers allow us to kind

127
00:06:36,600 --> 00:06:39,840
Speaker 1: of see the real picture of the AI bubble, and

128
00:06:39,880 --> 00:06:42,320
Speaker 1: I have to wonder what other companies look like now

129
00:06:42,320 --> 00:06:46,720
Speaker 1: that I've seen these numbers. Email me contact me Esitron

130
00:06:46,800 --> 00:06:49,080
Speaker 1: dots seventy six on signal. If you ever want to

131
00:06:49,080 --> 00:06:51,680
Speaker 1: tell me anything, do you ever want to show me anything.

132
00:06:52,440 --> 00:06:55,159
Speaker 1: I'm always interested to hear, and I'm honored to do

133
00:06:55,240 --> 00:06:55,360
Speaker 1: this