WEBVTT

1
00:00:00.080 --> 00:00:03.680
<v Speaker 1>AI daily briefing. I made sharp thanks for joining me

2
00:00:04.440 --> 00:00:07.519
<v Speaker 1>today the four hundred and twelve billion dollars AI bubble

3
00:00:07.759 --> 00:00:11.560
<v Speaker 1>when mega rounds hired a smaller market, four hundred and

4
00:00:11.560 --> 00:00:14.519
<v Speaker 1>twelve billion dollars flowed into US South venture capital in

5
00:00:14.560 --> 00:00:17.160
<v Speaker 1>the first half of this year. That number sounds like

6
00:00:17.160 --> 00:00:20.879
<v Speaker 1>a boon. The reality underneath it is more complicated. Eighty

7
00:00:20.920 --> 00:00:23.800
<v Speaker 1>six percent of that capital, roughly three hundred and fifty

8
00:00:23.839 --> 00:00:27.320
<v Speaker 1>six billion dollars went to AI. But strip away the

9
00:00:27.359 --> 00:00:30.800
<v Speaker 1>mega rounds and the picture shifts fast. Deals below one

10
00:00:30.839 --> 00:00:33.280
<v Speaker 1>hundred million dollars now account for just twelve and a

11
00:00:33.320 --> 00:00:36.960
<v Speaker 1>half percent of total venture value. Two years ago, that

12
00:00:37.000 --> 00:00:40.359
<v Speaker 1>figure was forty three point eight percent. The market hasn't

13
00:00:40.359 --> 00:00:44.560
<v Speaker 1>expanded broadly. It's concentrated sharply. A small number of very

14
00:00:44.640 --> 00:00:48.240
<v Speaker 1>large bets are inflating the headline number, while the broader

15
00:00:48.280 --> 00:00:52.359
<v Speaker 1>startup ecosystem quietly shrinks as a share of deployed capital.

16
00:00:52.960 --> 00:00:56.759
<v Speaker 1>That's the signal here, not the total the distribution. The

17
00:00:56.799 --> 00:00:59.439
<v Speaker 1>clearest example of that concentration sits at the top of

18
00:00:59.479 --> 00:01:03.159
<v Speaker 1>the frontier lab race. Anthropic just closed a sixty five

19
00:01:03.200 --> 00:01:06.159
<v Speaker 1>billion dollar funding round at a post money valuation of

20
00:01:06.239 --> 00:01:09.680
<v Speaker 1>nine hundred and sixty five billion dollars three months ago,

21
00:01:09.799 --> 00:01:12.640
<v Speaker 1>its pre money valuation was three hundred and fifty billion.

22
00:01:13.079 --> 00:01:15.439
<v Speaker 1>That's one hundred and fifty seven percent step up in

23
00:01:15.480 --> 00:01:18.680
<v Speaker 1>a quarter. The important distinction is that this now puts

24
00:01:18.719 --> 00:01:22.560
<v Speaker 1>Anthipic ahead of Open AI by valuation. Both companies have

25
00:01:22.640 --> 00:01:26.120
<v Speaker 1>filed confidentially to go public. What that means in practice

26
00:01:26.200 --> 00:01:28.319
<v Speaker 1>is that the race to a trillion dollar AI lab

27
00:01:28.400 --> 00:01:32.920
<v Speaker 1>valuation isn't herpathetical anymore. It's a matter of timing. The

28
00:01:33.040 --> 00:01:37.079
<v Speaker 1>unresolved question is whether the underlying revenue justifies these numbers,

29
00:01:37.560 --> 00:01:40.560
<v Speaker 1>or whether the valuation is pricing in AGI outcomes that

30
00:01:40.640 --> 00:01:44.799
<v Speaker 1>haven't materialized yet. While capital concentrates in the US, South

31
00:01:45.239 --> 00:01:48.120
<v Speaker 1>Europe is tightening the rules on how that capital gets deployed.

32
00:01:48.560 --> 00:01:52.120
<v Speaker 1>On July eighth, the EU Data Protection Board adopted Guidelines

33
00:01:52.200 --> 00:01:56.040
<v Speaker 1>zero three slash twenty twenty six. These aren't voluntary commitments,

34
00:01:56.959 --> 00:02:00.280
<v Speaker 1>they're an enforceable pan EU framework, requiring legal US view

35
00:02:00.359 --> 00:02:04.200
<v Speaker 1>before any scraping begins, data minimization at the point of collection,

36
00:02:04.480 --> 00:02:07.680
<v Speaker 1>and special handling for sensitive category DTA in training sets.

37
00:02:08.240 --> 00:02:11.960
<v Speaker 1>The catch that matters most the guidelines don't grandfather existing

38
00:02:12.039 --> 00:02:15.800
<v Speaker 1>data sets. Developers who collected training data between twenty twenty

39
00:02:15.800 --> 00:02:19.439
<v Speaker 1>three and twenty twenty five now face retroactive scrutiny under

40
00:02:19.479 --> 00:02:22.759
<v Speaker 1>the current balancing test. Most large language models weren't built

41
00:02:22.759 --> 00:02:26.120
<v Speaker 1>with that standard in mind. Compliance teams at Frontier Lapse

42
00:02:26.199 --> 00:02:28.840
<v Speaker 1>are now working backwards through data sets that were never

43
00:02:28.879 --> 00:02:32.599
<v Speaker 1>designed to be audited this way. That's not a theoretical risk,

44
00:02:33.039 --> 00:02:37.479
<v Speaker 1>it's an immediate operational burden. Here's the thing. The geographic

45
00:02:37.520 --> 00:02:40.680
<v Speaker 1>story of AI is also shifting. More than thirty founding

46
00:02:40.719 --> 00:02:43.240
<v Speaker 1>teams have relocated from Asia to the US South via

47
00:02:43.319 --> 00:02:47.039
<v Speaker 1>Antler since twenty twenty five. Asia's share of global venture

48
00:02:47.039 --> 00:02:49.240
<v Speaker 1>capital dropped to nine point six percent in the first

49
00:02:49.319 --> 00:02:52.400
<v Speaker 1>quarter of this year. The US self holds eighty percent.

50
00:02:53.120 --> 00:02:56.080
<v Speaker 1>Southeast Asian funding has collapsed from ten point one billion

51
00:02:56.120 --> 00:02:58.639
<v Speaker 1>dollars in twenty twenty two to two point two billion

52
00:02:58.680 --> 00:03:02.520
<v Speaker 1>in twenty twenty four. That's an eighty percent decline. The

53
00:03:02.560 --> 00:03:08.199
<v Speaker 1>incentive structure is straightforward, fragmented regulatory environments, limited exit opportunities,

54
00:03:08.479 --> 00:03:10.719
<v Speaker 1>and a customer base that doesn't scale the way Silicon

55
00:03:10.800 --> 00:03:14.479
<v Speaker 1>Valley does. Founders are voting with their feet. The visa

56
00:03:14.520 --> 00:03:17.560
<v Speaker 1>picture adds friction, but a court block in July ruled

57
00:03:17.599 --> 00:03:20.560
<v Speaker 1>unauthorized the recent fee hike that had pushed h one

58
00:03:20.599 --> 00:03:24.639
<v Speaker 1>B costs toward one hundred thousand dollars. Mercor, the ar

59
00:03:24.800 --> 00:03:27.719
<v Speaker 1>unicorn that hit two billion dollars in annual recurring revenue

60
00:03:27.759 --> 00:03:32.199
<v Speaker 1>in June, acquired simulation platform depect Tune. The strategic logic

61
00:03:32.280 --> 00:03:36.919
<v Speaker 1>is direct. The merkll brings five million domain experts. Depectune

62
00:03:36.960 --> 00:03:40.159
<v Speaker 1>brings enterprise software environments where AI agents can train on

63
00:03:40.240 --> 00:03:45.759
<v Speaker 1>tools like Salesforce, Slack, and Excel before touching production systems. Together,

64
00:03:45.879 --> 00:03:50.039
<v Speaker 1>its full stack reinforcement learning infrastructure for frontier labs building agents.

65
00:03:50.599 --> 00:03:53.800
<v Speaker 1>Mercor doubled its arr in twelve months. The d chap

66
00:03:53.840 --> 00:03:57.039
<v Speaker 1>to acquisition is a bet that agent training infrastructure becomes

67
00:03:57.039 --> 00:04:01.240
<v Speaker 1>the next critical layer. The final thread worth tracking is infrastructure.

68
00:04:01.719 --> 00:04:04.599
<v Speaker 1>Asian investors are pivoting away from growth at all costs

69
00:04:04.639 --> 00:04:08.800
<v Speaker 1>AI toward what they're calling resilient AI. Capitalists flow into

70
00:04:08.800 --> 00:04:13.439
<v Speaker 1>liquid coaling, small modular reactors, and edge computing. The power

71
00:04:13.479 --> 00:04:16.399
<v Speaker 1>and thermal constraints of running large AI systems are no

72
00:04:16.480 --> 00:04:21.160
<v Speaker 1>longer secondary concerns. They're becoming the primary bottlemech. This isn't

73
00:04:21.199 --> 00:04:23.920
<v Speaker 1>just an Asian story. It's an early signal of where

74
00:04:23.920 --> 00:04:28.079
<v Speaker 1>the global infrastructure conversation is heading. Consider this. The near

75
00:04:28.160 --> 00:04:32.040
<v Speaker 1>term watch points are clear. Watch whether anthropics valuation holds

76
00:04:32.079 --> 00:04:36.199
<v Speaker 1>as public market scrutiny approaches. Watch how EU regulators begin

77
00:04:36.319 --> 00:04:40.120
<v Speaker 1>enforcing the new scraping guidelines against non compliant training data sets.

78
00:04:40.720 --> 00:04:44.199
<v Speaker 1>And watch whether the concentration in venture funding accelerates further

79
00:04:44.680 --> 00:04:47.560
<v Speaker 1>or triggers the correction that some analysts are already modeling.

80
00:04:48.040 --> 00:04:50.959
<v Speaker 1>The headline number is four hundred and twelve billion. The

81
00:04:51.040 --> 00:04:54.639
<v Speaker 1>real story is who's capturing it and who isn't. Thanks

82
00:04:54.639 --> 00:04:58.360
<v Speaker 1>for listening. This podcast was built using AI technology, a

83
00:04:58.439 --> 00:04:59.240
<v Speaker 1>Yes We production