AI Bosses Want to Slow Down. Could Money Be One Reason? — Transcript

Full transcript

1. 0:00AI labs have spent hundreds of billions
2. 0:02of dollars competing to build ever more
3. 0:04powerful models, but now they're warning
4. 0:06us about the dangers of their creations
5. 0:08and saying it could be time to slow
6. 0:10down. They say it's about safety,
7. 0:12staving off Armageddon, but could there
8. 0:14be another reason? Might it also be
9. 0:17about the money? And might the huge
10. 0:19financial commitment already made to AI
11. 0:22pose its own threat, if not to the
12. 0:24species, to the global economic system?
13. 0:26To understand, we need to follow the
14. 0:29money, and there is a lot to follow. AI
15. 0:31has triggered one of the biggest capital
16. 0:33investment cycles in history. Estimates
17. 0:36like this one put the cost of the data
18. 0:38centers and associated technology over
19. 0:40the next 25 years at somewhere between
20. 0:42$20 at the slowest rate and $50 at the
21. 0:46fastest. We are in the opening stages of
22. 0:48a technological and financial arms race,
23. 0:52and it is being led by a big five of
24. 0:54compute providers, dubbed the
25. 0:56hyperscalers. You'll have heard of all
26. 0:58of them, Amazon, Google, Microsoft,
27. 1:00Meta, and Oracle. They are spending at
28. 1:03least $800 billion this year on the
29. 1:06land, energy, grid connections, and
30. 1:08cooling technology, and crucially the
31. 1:10chips that power data centers. And look,
32. 1:13that investment has almost doubled in
33. 1:15each of the last three years. But the
34. 1:18extraordinary thing isn't just how much
35. 1:20these companies are spending, it's even
36. 1:22the richest companies in the world are
37. 1:23struggling to pay for it themselves.
38. 1:26Let's take Google as an example. This is
39. 1:28one of the most profitable and
40. 1:30cash-generating businesses that has ever
41. 1:32existed, but even its cash flow can't
42. 1:35keep up. Look, five years ago Google
43. 1:36generated more than $90 billion of cash,
44. 1:39that's the blue bar, of which it
45. 1:40invested $25 billion, around a third,
46. 1:44and it issued less than $15 of debt,
47. 1:46that's the light blue. If we fast
48. 1:48forward to this year, in the first half
49. 1:50of 2026, it had already generated almost
50. 1:53as much cash, but it's plunged almost
51. 1:55all of it into capital expenditure and
52. 1:58borrowed another 50, believe 52 billion
53. 2:01by issuing bonds. And it's the same
54. 2:04story for many of the hyperscalers.
55. 2:06These wildly profitable cash machines
56. 2:09have had to turn to the debt markets
57. 2:10offering long-term corporate bonds to
58. 2:12institutional investors on a scale that
59. 2:15was unimaginable a few years ago. The
60. 2:17acceleration really is remarkable. This
61. 2:19is the US market and hyperscaler debt
62. 2:22issuance has gone from less than 20
63. 2:24billion dollars three years ago to more
64. 2:25than 130 billion this year already. And
65. 2:28look at the black line. That's the share
66. 2:31of the US bond market that this
67. 2:32represents. Borrowing for AI
68. 2:34infrastructure now makes up more than
69. 2:3610%, which means it really matters to
70. 2:40financial stability. Now, there is
71. 2:42another way of looking at this. The debt
72. 2:44market is also where governments go to
73. 2:46borrow. In the first half of this year,
74. 2:47the hyperscalers have issued 170 billion
75. 2:50dollars of debt once you add in
76. 2:52non-dollar issuance. Compare that to the
77. 2:54UK, which issued gilts worth the
78. 2:56equivalent of 174 billion dollars this
79. 2:59year. They're basically the same. AI
80. 3:01infrastructure has the same appetite for
81. 3:04debt as Britain, a G7 economy. And that
82. 3:08matters not least because these
83. 3:09companies may now be competing with
84. 3:11governments for the same pool of money.
85. 3:13It may partly explain why government
86. 3:15borrowing costs have spiked recently.
87. 3:18So, what's the problem with spending so
88. 3:20much so fast? Maybe there isn't one.
89. 3:22Growing companies have always borrowed
90. 3:24and these are some of the most
91. 3:25profitable in history. They're investing
92. 3:28in infrastructure that will deliver an
93. 3:29industrial revolution and investors and
94. 3:31institutions are falling over themselves
95. 3:34to provide capital. Well, what's not to
96. 3:36like? Well, there are a couple of things
97. 3:37that make this boom unusual and they
98. 3:39have started to worry regulators. First,
99. 3:42there is the concentration of risk. This
100. 3:44is a handful of companies borrowing on a
101. 3:46nation-state scale, buying hardware from
102. 3:49a tiny number of suppliers to enable
103. 3:51development by an even smaller number of
104. 3:53AI labs and almost all of them in the
105. 3:55US. If one of them fails, the effects
106. 3:57could ripple through the global economy.
107. 4:00Earlier this year, the Bank of England
108. 4:02said this, "An adverse shock to AI
109. 4:04companies that results in losses or
110. 4:06affects their ability to service debt
111. 4:08could more materially affect global
112. 4:11financing conditions." To paraphrase, an
113. 4:14AI crash will not be confined to AI.
114. 4:17That's a message the Bank of England
115. 4:18governor, Andrew Bailey, took to G20
116. 4:20leaders uh earlier this month. Another
117. 4:23concern for regulators is the so-called
118. 4:25circularity
119. 4:27of funding. The major players, as we can
120. 4:29see here, are connected by a complex web
121. 4:31of intra-company investment and spending
122. 4:34that moves money between them. This,
123. 4:36mapped by Bloomberg, shows the links
124. 4:38between Anthropic, creators of Claude,
125. 4:40and its major chip and compute
126. 4:43providers. Amazon, Google, Microsoft,
127. 4:45and chip makers, Nvidia and AMD, have
128. 4:47all invested into Anthropic. That's the
129. 4:50light blue lines going in here. In
130. 4:52return, Anthropic has purchased compute
131. 4:55services, which is the royal blue lines
132. 4:57the other way, and hardware, the pink
133. 4:59lines. In effect, the investors are
134. 5:01helping finance demand for their own
135. 5:05products. The companies say this is a
136. 5:06virtuous circle that will accelerate
137. 5:09technologies and revenue. And some of
138. 5:11the biggest private equity investors in
139. 5:12the world agree, but the risk, as we
140. 5:15discovered when the dot-com bubble
141. 5:16burst, is what happens if one part of
142. 5:19this circle stops working. If history
143. 5:22does have a lesson, it may be here. This
144. 5:24chart shows the scale of investment in
145. 5:27previous transforming technologies
146. 5:29expressed as a share of GDP. Railways,
147. 5:32cars, all had a bigger investment cycle
148. 5:36than the current spending in AI, which
149. 5:38is only just bigger than that in the
150. 5:39tech boom. It doesn't look so unusual,
151. 5:42does it? Though, of course, it's worth
152. 5:43bearing in mind the railways and tech
153. 5:46all suffered major financial crashes on
154. 5:48their way to transforming the economy.
155. 5:50So, even if AI is a bubble and it
156. 5:53bursts, it doesn't mean it won't deliver
157. 5:56eventually. Now, there are plenty of AI
158. 5:58optimists, including Donald Trump, who
159. 6:00dismiss recent warnings as
160. 6:01scaremongering that amounts to either
161. 6:03marketing or an excuse to slow down this
162. 6:05arms race. And if AI delivers on its
163. 6:07promise, if US technology remains ahead
164. 6:10of Chinese competition, if public
165. 6:12consent for the technology and its
166. 6:14intrusive infrastructure remains, and
167. 6:16obviously if it doesn't kill us all
168. 6:18first, then the trillions invested in
169. 6:21data centers will be repaid handsomely.
170. 6:23But, it is an if. And while we wait to
171. 6:26find out, it's worth remembering where
172. 6:28the risk lies today.