1
00:00:00,000 --> 00:00:04,160
Neither OpenAI nor Anthropic are profitable.

2
00:00:04,520 --> 00:00:06,580
That fact cannot be stressed enough.

3
00:00:06,740 --> 00:00:09,340
Now, they both have massive customer bases,

4
00:00:09,560 --> 00:00:11,640
and they both have wildly popular products,

5
00:00:12,000 --> 00:00:17,780
and yet they are burning through capital like crazy.

6
00:00:18,040 --> 00:00:21,580
Ed Zittron does a fantastic breakdown of this on his blog,

7
00:00:21,720 --> 00:00:24,180
but real quick, let's just do a rough overview.

8
00:00:24,180 --> 00:00:36,160
so open ai lost 20 billion dollars in the year 2025 on 13 billion in revenue now they have over

9
00:00:36,160 --> 00:00:45,460
800 billion dollars in obligations that will come due by 2030 and 146 billion dollars of that

10
00:00:45,460 --> 00:00:52,760
is due at the end of next year 2027 if they do not do something astonishing they have no real

11
00:00:52,760 --> 00:00:59,980
way to pay that bill unless someone else foots it for them. Now, this is on top of both Anthropic

12
00:00:59,980 --> 00:01:06,980
and OpenAI getting some of the biggest discounts possible on compute for training and inference.

13
00:01:06,980 --> 00:01:14,620
So this isn't just bad. It's also being subsidized and is still this bad for OpenAI.

14
00:01:14,620 --> 00:01:22,060
Now, in quarter one of 2026, they burned $12 billion to make $5.7 billion.

15
00:01:22,660 --> 00:01:28,600
Again, this is the biggest AI company out there with a wildly popular product.

16
00:01:29,120 --> 00:01:32,060
How are they going to make up that difference?

17
00:01:32,400 --> 00:01:37,360
Are they somehow going to double and triple their customer base without doubling and tripling their costs?

18
00:01:37,620 --> 00:01:41,040
Are they going to double and triple their prices and stay competitive?

19
00:01:41,040 --> 00:01:50,540
Or are they somehow going to more than have all of their costs and do so sustainably within like a year?

20
00:01:50,840 --> 00:01:53,280
Anthropic doesn't seem to be a whole lot better off.

21
00:01:53,780 --> 00:01:57,980
So, quote, Anthropic's gross margins, they've supposedly made a profit for two quarters.

22
00:01:58,300 --> 00:02:10,040
Anthropic's gross margins are above 80% before accounting for revenue shared with distribution partners, including Amazon, and the cost of training its models.

23
00:02:10,540 --> 00:02:15,960
Not to mention this includes a huge one-time discount that Elon gave them on compute.

24
00:02:16,140 --> 00:02:21,880
That's like saying if you take out the cost of making the product and the cost of distributing the product, we're making money.

25
00:02:22,160 --> 00:02:23,760
Hey, we're making enough money, right?

26
00:02:24,080 --> 00:02:26,860
You think their training costs are going to get any less?

27
00:02:27,440 --> 00:02:31,760
That this isn't just something that they're always going to have ongoing?

28
00:02:32,460 --> 00:02:49,984
That they just going to finish and there just going to be a model that done and stops costing them money Training cost is a part of the business They will always be there They also signed on for billion in commitments to multiple other massive corporations for compute and training

29
00:02:49,984 --> 00:02:55,984
Both of these companies are expected to IPO soon, and both of them want larger-than-trillion-dollar valuations.

30
00:02:56,904 --> 00:03:01,364
Yet neither seems to have a concrete way to pay all of this stuff.

31
00:03:01,364 --> 00:03:05,704
The accounting is so bad that every customer they add, they lose more money.

32
00:03:06,224 --> 00:03:12,144
And their losses are entangled in the entire tech industry.

33
00:03:12,484 --> 00:03:15,344
Enter China's open source models.

34
00:03:15,764 --> 00:03:22,084
So Kimi K3 is one of the most recent frontier models, open source models to come out of China.

35
00:03:22,304 --> 00:03:23,624
And it's amazing.

36
00:03:24,164 --> 00:03:24,604
It's amazing.

37
00:03:24,724 --> 00:03:28,164
I use Fable 5 and Kimi basically in tandem.

38
00:03:28,164 --> 00:03:34,024
I get Kimmy to check the work of Fable and then Fable to respond to the checked work of Kimmy.

39
00:03:34,244 --> 00:03:35,764
I basically go back and forth.

40
00:03:35,984 --> 00:03:37,604
And they're both fantastic models.

41
00:03:38,064 --> 00:03:40,724
I can't tell a massive difference between them.

42
00:03:41,004 --> 00:03:47,044
I would probably say Fable is better at larger coding problems and larger context windows.

43
00:03:47,464 --> 00:03:54,484
But I don't have the greatest confidence or way to benchmark that for myself.

44
00:03:54,484 --> 00:03:58,264
because I'm not a developer and I'm not going to read hundreds, thousands of lines of codes

45
00:03:58,264 --> 00:04:00,404
for all the stupid projects that I'm working on.

46
00:04:00,484 --> 00:04:01,884
Holy God, I'm not reading all that.

47
00:04:01,884 --> 00:04:08,104
The one big difference is that Kimi K3 is like a tenth of the price

48
00:04:08,104 --> 00:04:11,104
and I can't really tell the difference in the output.

49
00:04:11,384 --> 00:04:13,724
That's not good for open AI and Anthropic.

50
00:04:13,964 --> 00:04:20,004
And the open source models only seem to be like two to four weeks behind the frontier US companies.

51
00:04:20,384 --> 00:04:21,924
They have no moat.

52
00:04:21,924 --> 00:04:30,944
And the bubble bursting on these companies and their obligations will be spectacular if something doesn't change.

53
00:04:31,144 --> 00:04:31,944
And that's the point.

54
00:04:32,764 --> 00:04:37,864
I think this is an existential crisis for them.

55
00:04:37,864 --> 00:04:57,924
Be honest with yourself. If you had a multi-hundred billion dollar company that was the talk of the whole world right now and may end up being at the heart of one of the worst investment failures in history that makes the dot-com bubble look like child's play, what wouldn't you do to protect that?

56
00:04:58,024 --> 00:05:06,064
You think they wouldn't use their own tools to plan a campaign and run a bunch of bots to try to convince everyone that AI is the most dangerous thing in the world?

57
00:05:06,064 --> 00:05:07,804
We're doomed. Doomed.

58
00:05:09,924 --> 00:05:25,788
That they wouldn run news articles that they wouldn create institutions to check their work that said yeah it is super dangerous Industry leaders and some lawmakers have been calling for new regulation saying there a need to slow down after former AI researchers warned

59
00:05:25,788 --> 00:05:27,888
the technology is advancing so quickly,

60
00:05:28,248 --> 00:05:32,348
it could eventually outsmart the companies building it and pose a threat to humanity.

61
00:05:32,348 --> 00:05:35,828
That they wouldn't try to get direct relationships with the government

62
00:05:35,828 --> 00:05:38,628
and joint ownership for the good of humanity

63
00:05:38,628 --> 00:05:42,228
because of just how important the thing that they're building really is.

64
00:05:42,348 --> 00:05:44,048
Now, I'm not saying that AI isn't important.

65
00:05:44,148 --> 00:05:45,648
I think it's vastly important.

66
00:05:45,768 --> 00:05:52,328
But I don't think you're going to be able to maintain a monopoly on it unless you change the way the market works.

67
00:05:52,328 --> 00:05:56,848
And of course this will come at the cost of massive subsidies for the good of everyone.

68
00:05:57,328 --> 00:06:00,348
And huge government contracts for the good of everyone.

69
00:06:00,668 --> 00:06:03,748
And liability protection for the good of everyone.

70
00:06:03,988 --> 00:06:08,988
And a guaranteed customer in many of the world's largest governments for the good of everyone.

71
00:06:09,208 --> 00:06:15,588
And they will of course do whatever those governments want in surveilling and controlling and manipulating their people.

72
00:06:15,768 --> 00:06:20,908
And so they will walk away with guaranteed customers, a market that's almost impossible to enter,

73
00:06:21,088 --> 00:06:26,648
compliance costs that are so enormous that no small company can actually meet them.

74
00:06:26,728 --> 00:06:28,788
I think there's a reason they compare it to nuclear.

75
00:06:29,048 --> 00:06:31,348
Protection from open source competition.

76
00:06:31,688 --> 00:06:34,788
I mean, foreign competition, excuse me, dangerous foreign models.

77
00:06:34,788 --> 00:06:40,468
And a cozy relationship with the world's largest decatrillion dollar institution

78
00:06:40,468 --> 00:06:43,668
that gets to put their customers in prison for not paying the bill.

79
00:06:43,748 --> 00:06:44,768
Straight to jail.

80
00:06:44,768 --> 00:06:46,828
Ha! Now you got burned. No bail.

81
00:06:47,128 --> 00:06:49,768
I say keep an eye out for three things as all of this unfolds.

82
00:06:50,488 --> 00:06:56,148
And I promise I would bet money, good money, that you will see all three of these

83
00:06:56,148 --> 00:07:00,288
in whatever regulatory or joint initiative comes out of this,

84
00:07:00,388 --> 00:07:02,348
or it will not be supported by them.

85
00:07:02,528 --> 00:07:08,188
Number one is a government partnership or a mini-government partnership that involves a lot of money.

86
00:07:08,368 --> 00:07:10,988
Two, liability protection for the AI companies.

87
00:07:11,408 --> 00:07:13,448
It's the pharma game. They know how to play it.

88
00:07:13,448 --> 00:07:21,188
and three, an AI regulatory board or some sort of joint institution that they sit at the center of

89
00:07:21,188 --> 00:07:26,688
making all the decisions. That's what this is about. They want to protect a failing business

90
00:07:26,688 --> 00:07:33,948
model on the back of a wildly successful and wildly useful product by making it look like

91
00:07:33,948 --> 00:07:39,528
what they're doing is so dangerous and so important that it will literally kill us all

92
00:07:39,528 --> 00:07:49,148
Unless they can protect us. Even Mark Zuckerberg said, we don't need alignment regulation because a misaligned model is a model that nobody wants to use.

93
00:07:49,328 --> 00:08:07,592
If a model doesn do what you want you not going to use that model And liability is also right where it needs to be on the companies that make the models It just so happens that creating an AI model that lies to customers behaves erratically and unpredictably and kills all of humanity might be bad for business

94
00:08:10,512 --> 00:08:13,592
Alignment is dealt with by customers who are free to leave

95
00:08:13,592 --> 00:08:15,972
and liability that is paid for by the company.

96
00:08:15,972 --> 00:08:21,592
The only way that a company can be sustained by aligning against the interests of the people,

97
00:08:21,772 --> 00:08:24,832
if we start focusing on making offensive military operations

98
00:08:24,832 --> 00:08:29,912
and models entirely focused on infiltration, and importantly, put up guardrails for the

99
00:08:29,912 --> 00:08:36,332
everyday person so that they can't even make their software and servers more secure, because

100
00:08:36,332 --> 00:08:38,652
that's dangerous territory.

101
00:08:38,652 --> 00:08:43,552
Only if we heavily control the industry so that everyone is stuck in just a few giant

102
00:08:43,552 --> 00:08:50,712
corporate products that will guarantee the largest and most comprehensive and most profitable

103
00:08:50,712 --> 00:08:53,992
surveillance system ever devised by mankind.

104
00:08:53,992 --> 00:08:57,812
And of course if we have that small group of governments and huge corporations build

105
00:08:57,812 --> 00:09:02,472
themselves a veritable monopoly on a technology that is aggressively trying to be free and

106
00:09:02,472 --> 00:09:04,012
open for everyone to use.

107
00:09:04,012 --> 00:09:11,232
In effect, creating a very big and violent artificial moat so they can maintain permanent

108
00:09:11,232 --> 00:09:12,232
control.

109
00:09:12,232 --> 00:09:15,352
That's what this is about.

110
00:09:15,352 --> 00:09:17,052
We pay the bill.

111
00:09:17,052 --> 00:09:19,312
We pay for the liability.

112
00:09:19,312 --> 00:09:25,452
We lose access to the models that would protect us from hacking and infiltration.

113
00:09:26,152 --> 00:09:32,072
We lose the ability to have alternatives and get around their massive surveillance system.

114
00:09:32,732 --> 00:09:41,492
And yet we will still get to watch AI become violent and less predictable and more geared toward destruction and not serving the customers.

115
00:09:41,492 --> 00:09:47,712
But treating customers and people as things to control and to make sure they don't get out of line.

116
00:09:47,712 --> 00:09:51,232
or get access to these super dangerous weapons.

117
00:09:51,832 --> 00:09:57,532
We will see models embedded with all sorts of ideological and political nonsense and inconsistencies

118
00:09:57,532 --> 00:10:00,592
that do behave erratic and unpredictably.

119
00:10:00,592 --> 00:10:07,632
And we will get exactly the giant centralized infrastructure of control and surveillance

120
00:10:07,632 --> 00:10:13,352
that the gaslighting bastards claim only they can protect us from.

121
00:10:13,952 --> 00:10:17,552
Sorry, I don't believe you.

122
00:10:17,712 --> 00:10:21,352
Let them pay the bill.

123
00:10:22,032 --> 00:10:24,192
They are out to save their own asses,

124
00:10:24,432 --> 00:10:26,012
and I want no part of it.

125
00:10:26,532 --> 00:10:28,072
Don't fall for it.

126
00:10:29,532 --> 00:10:31,072
Well, that's just my two cents.