WEBVTT

1
00:00:00.040 --> 00:00:04.960
<v Speaker 1>Ai Daily briefing, I'm ed sharp, thanks for joining me today.

2
00:00:05.280 --> 00:00:10.560
<v Speaker 1>China's Ai surprise timeline just collapsed. China just collapsed the timeline.

3
00:00:10.880 --> 00:00:14.519
<v Speaker 1>Moonshot AI's Kimmi K three launched this week, ranking number

4
00:00:14.519 --> 00:00:17.160
<v Speaker 1>one on Arena got AI's front end Code Arena with

5
00:00:17.239 --> 00:00:20.839
<v Speaker 1>a seventy six percent pairwise win rate, outperforming both clawed

6
00:00:20.879 --> 00:00:24.079
<v Speaker 1>in GPT on key benchmarks. The expert consensus was that

7
00:00:24.160 --> 00:00:26.960
<v Speaker 1>Chinese frontier models were at least six months away from

8
00:00:27.039 --> 00:00:31.359
<v Speaker 1>matching US leaders. That assumption is now gone. The market

9
00:00:31.399 --> 00:00:35.920
<v Speaker 1>reaction was immediate. In telling TSMC dropped seven percent, Nvidia

10
00:00:36.000 --> 00:00:38.920
<v Speaker 1>briefly lost its position as the world's most valuable company.

11
00:00:39.359 --> 00:00:42.840
<v Speaker 1>The Nasdaq one hundred fell one percent. Investors weren't just

12
00:00:42.880 --> 00:00:46.200
<v Speaker 1>reacting to one model release. They were reprising their assumptions

13
00:00:46.200 --> 00:00:49.200
<v Speaker 1>about how long American labs hold the lead. Here's what

14
00:00:49.320 --> 00:00:53.240
<v Speaker 1>makes Kimmi K three significant Beyond the benchmarks. It's open weight.

15
00:00:53.679 --> 00:00:56.759
<v Speaker 1>That means it can be deployed, fine tuned, and adapted

16
00:00:56.840 --> 00:01:01.320
<v Speaker 1>without Moonshot's permission or infrastructure. Xx Sport controls were designed

17
00:01:01.359 --> 00:01:05.760
<v Speaker 1>to slow Chinese capability development by restricting advanced chips. What

18
00:01:05.879 --> 00:01:09.239
<v Speaker 1>Kimmi K three suggests is that Chinese labs responded by

19
00:01:09.239 --> 00:01:13.519
<v Speaker 1>optimizing for efficiency rather than raw compute, and that strategy

20
00:01:13.640 --> 00:01:17.640
<v Speaker 1>is now producing frontier level results. The important caveat is

21
00:01:17.680 --> 00:01:21.519
<v Speaker 1>that benchmark rankings don't always translate cleanly to real world performance.

22
00:01:22.079 --> 00:01:26.640
<v Speaker 1>Arena dot AIS leaderboard measures specific coding tasks. Whether that

23
00:01:26.760 --> 00:01:29.760
<v Speaker 1>lead holds across broader applications is still an open question,

24
00:01:30.400 --> 00:01:34.040
<v Speaker 1>But the gap closing this fast, this suddenly is the signal,

25
00:01:34.439 --> 00:01:38.120
<v Speaker 1>not the benchmark number itself. The NVIDIAs story deserves a

26
00:01:38.159 --> 00:01:41.920
<v Speaker 1>direct read. Apple reclaimed the top market cap position this week,

27
00:01:42.200 --> 00:01:45.439
<v Speaker 1>approaching five trillion dollars in value. That's not just a

28
00:01:45.480 --> 00:01:50.680
<v Speaker 1>market curiosity, it reflects a genuine investor repricing. Consumer distribution

29
00:01:50.879 --> 00:01:53.840
<v Speaker 1>and device ownership are starting to look like more durable

30
00:01:53.879 --> 00:01:58.319
<v Speaker 1>advantages than chip infrastructure scarcity. When Kimi K three landed

31
00:01:58.560 --> 00:02:02.000
<v Speaker 1>and demonstrated that frontier a capability can emerge from constrained

32
00:02:02.040 --> 00:02:05.120
<v Speaker 1>hardware in Nvidia's mote looked narrower than it did a

33
00:02:05.120 --> 00:02:09.159
<v Speaker 1>week ago. Google is moving in the opposite direction. Alphabet

34
00:02:09.199 --> 00:02:12.479
<v Speaker 1>shares fell over four percent after reports confirmed that Gemini

35
00:02:12.520 --> 00:02:15.879
<v Speaker 1>three point five pro missed its scheduled release. The stated

36
00:02:15.919 --> 00:02:19.400
<v Speaker 1>reason is refinement of coding and ergentic capabilities. That's a

37
00:02:19.520 --> 00:02:23.400
<v Speaker 1>plausible explanation. It's also exactly the kind of explanation that

38
00:02:23.479 --> 00:02:27.439
<v Speaker 1>becomes concerning when competitors are shipping weekly. Google has scale

39
00:02:27.479 --> 00:02:31.479
<v Speaker 1>advantages that won't disappear overnight, but execution risk is real

40
00:02:31.680 --> 00:02:34.800
<v Speaker 1>and markets are starting to price it in. Here's the thing.

41
00:02:35.280 --> 00:02:39.360
<v Speaker 1>Meta is doubling down on infrastructure as its core competitive strategy.

42
00:02:39.800 --> 00:02:43.479
<v Speaker 1>The company hired Dave Brown, an eighteen year AWS veteran,

43
00:02:43.719 --> 00:02:48.120
<v Speaker 1>to lead AI data center expansion. Meta's annual infrastructure budget

44
00:02:48.159 --> 00:02:50.840
<v Speaker 1>sits between one hundred and twenty five billion and one

45
00:02:50.960 --> 00:02:55.199
<v Speaker 1>hundred and forty five billion dollars that covers servers, networking,

46
00:02:55.439 --> 00:02:58.599
<v Speaker 1>and power. The signal here is that Meta isn't just

47
00:02:58.680 --> 00:03:01.960
<v Speaker 1>building models, it's building the platform those models run on,

48
00:03:02.240 --> 00:03:05.080
<v Speaker 1>and it wants someone who has operated cloud infrastructure at

49
00:03:05.080 --> 00:03:08.879
<v Speaker 1>Amazon's scale to lead that effort. New York Governor Hockle

50
00:03:09.199 --> 00:03:12.680
<v Speaker 1>signed the country's first executive order imposing a data centre

51
00:03:12.759 --> 00:03:17.280
<v Speaker 1>moratorium this week, citing power consumption, water use, and local pollution.

52
00:03:18.039 --> 00:03:21.520
<v Speaker 1>Major projects already underway appear to continue, but the order

53
00:03:21.639 --> 00:03:25.039
<v Speaker 1>establishes a precedent. The question now is whether other states

54
00:03:25.080 --> 00:03:27.599
<v Speaker 1>follow and how that reshapes where the next wave of

55
00:03:27.639 --> 00:03:31.759
<v Speaker 1>AI infrastructure actually gets built. The backlash against AI isn't

56
00:03:31.759 --> 00:03:35.800
<v Speaker 1>staying abstract. Executives at leading AI companies are now hiring

57
00:03:35.919 --> 00:03:40.599
<v Speaker 1>armed security and limiting public appearances following documented threats, including

58
00:03:40.639 --> 00:03:43.680
<v Speaker 1>an attempted fire bombing at an Open AI executive's home.

59
00:03:44.199 --> 00:03:47.400
<v Speaker 1>That's not a policy debate anymore. It's a different category

60
00:03:47.439 --> 00:03:51.639
<v Speaker 1>of pressure entirely. Two product moves worth tracking. Open AI

61
00:03:51.759 --> 00:03:55.680
<v Speaker 1>deployed GPT Live, a full duplex voice model that lets

62
00:03:55.719 --> 00:03:59.800
<v Speaker 1>chat GPT listen and speak simultaneously rather than taking turns.

63
00:04:00.479 --> 00:04:03.599
<v Speaker 1>That removes a fundamental constraint in voice into action and

64
00:04:03.680 --> 00:04:08.280
<v Speaker 1>opens real applications in customer service and accessibility. And Microsoft

65
00:04:08.319 --> 00:04:12.199
<v Speaker 1>is preparing Project Perception, a cyber security product that routes

66
00:04:12.280 --> 00:04:17.120
<v Speaker 1>queries across models from Microsoft, Open Ai, and Anthropic. The

67
00:04:17.199 --> 00:04:21.319
<v Speaker 1>framing is multi model flexibility. The subtext is a direct

68
00:04:21.360 --> 00:04:25.639
<v Speaker 1>challenge to anthropics growing security market position. The clearest watch

69
00:04:25.680 --> 00:04:30.000
<v Speaker 1>points from here how Kimi K three performs outside controlled benchmarks,

70
00:04:30.360 --> 00:04:33.600
<v Speaker 1>whether Google's Gemini delay extends beyond the weeks in two months,

71
00:04:34.040 --> 00:04:38.079
<v Speaker 1>and how metas infrastructure ambitions translate once Dave Brown is operational.

72
00:04:38.680 --> 00:04:41.120
<v Speaker 1>The deeper question running through all of it is whether

73
00:04:41.160 --> 00:04:44.240
<v Speaker 1>the competitive advantage in AI is shifting from who builds

74
00:04:44.240 --> 00:04:47.399
<v Speaker 1>the best model to who controls the infrastructure those models

75
00:04:47.399 --> 00:04:51.120
<v Speaker 1>depend on. This week pushed that question forward without fully

76
00:04:51.160 --> 00:04:55.040
<v Speaker 1>answering it. Thanks for listening. This podcast was built using

77
00:04:55.079 --> 00:04:57.480
<v Speaker 1>AI technology, a Yes We production