# OpenAI, DeepSeek Chase Billions as AI Funding Boom Rolls On

An episode of [Bloomberg Tech](https://invisiblenetwork.ai/podcast/1611767434), 6 Oct 2026, 42 min.

Audio: https://podtrac.com/pts/redirect.mp3/traffic.omny.fm/d/clips/e73c998e-6e60-432f-8610-ae210140c5b1/41764a4f-fc64-4e11-89ba-ae7c0030ab5e/5cc92204-cfdc-4592-bbb7-b4db01204906/audio.mp3?utm_source=Podcast&in_playlist=9caafc41-289c-4115-995d-ae7c0030ab75

Bloomberg’s Ed Ludlow breaks down OpenAI's talks with multiple investors to raise at least $30 billion at a valuation of about $1.4 trillion. Plus, DeepSeek is also close to securing at least 12 billion in its latest round of funding ahead of its own IPO next year, and JPMorganChase CEO Jamie Dimon weighs in on data centers and AI risks from London's Tech Stars event. See omnystudio.com/listener for privacy information.

## Transcript

[0:02] Speaker 1: Bloomberg Audio Studios. Podcasts.
[0:05] Speaker 2: Radio.
[0:06] Speaker 3: News.
[0:13] Speaker 1: Bloomberg Tech is live from the heart of Silicon Valley
[0:17] Speaker 1: with Ed Ludlow in San Francisco. This is Bloomberg Tech
[0:23] Speaker 1: coming up.
[0:24] Speaker 2: Updates on OpenAI's talks with multiple investors to raise at
[0:28] Speaker 2: least $ 30 billion at a valuation of about $ 1. 4 trillion.
[0:32] Speaker 2: The UAE and BlackRock in the mix. Plus more funding
[0:35] Speaker 2: news as DeepSeek is also close to securing at least
[0:38] Speaker 2: $ 12 billion in its latest round of funding ahead of
[0:41] Speaker 2: its own IPO next year. And we'll hear from JPMorgan
[0:44] Speaker 2: Chase CEO Jamie Dimon on data centers, AI risks, and
[0:48] Speaker 2: a lot more from London's Techstars event. That's later this hour.
[0:52] Speaker 2: Let's get right to our top story, an update on
[0:54] Speaker 2: OpenAI's talks with multiple investors to raise at least $ 30 billion.
[0:58] Speaker 2: Sources say that includes speaking with multiple funds from the
[1:02] Speaker 2: United Arab Emirates to help anchor that round. Bloomberg Senior
[1:06] Speaker 2: Tech Editor Mike Shepard is with us. Let's go through
[1:08] Speaker 2: what we're reporting, right? Some of this is new.
[1:12] Speaker 4: Some of this is new, including one of the key participants, Ed,
[1:15] Speaker 4: and that would be the Abu Dhabi-based fund MGX, which
[1:19] Speaker 4: is looking to play a substantial role in the syndicate
[1:22] Speaker 4: of UAE-based funds. that are looking to invest as much
[1:26] Speaker 4: as $ 10 billion in this total $ 30 billion round. Now,
[1:30] Speaker 4: for MGXEd, as you know, this is not their first
[1:32] Speaker 4: rodeo when it comes to AI investments. They put into
[1:35] Speaker 4: previous rounds by both OpenAI and its rival, Anthropic, and
[1:39] Speaker 4: they've also raised earlier this year as much as $ 50
[1:42] Speaker 4: billion for AI infrastructure-related investments. And the key thing about this,
[1:47] Speaker 4: of course, is that the round itself would take OpenAI's
[1:51] Speaker 4: book value on paper to $ 1. 4 trillion, and that would
[1:55] Speaker 4: be up from the $ 122 billion raise that had it
[1:57] Speaker 4: at $ 852 billion just back in March.
[2:06] Speaker 2: A lot of people, when we first broke the news
[2:08] Speaker 2: of the idea of a round, which was early last week,
[2:10] Speaker 2: were saying, well, why would OpenAI do this?
[2:13] Speaker 1: And like, what's going on with the IPO?
[2:15] Speaker 2: When we first broke that initial report last Tuesday, we
[2:18] Speaker 2: didn't have names. But I was able to ask OpenAI
[2:21] Speaker 2: CFO Sarah Fryer about a round.
[2:23] Speaker 1: This is what she said.
[2:25] Speaker 3: We're really well capitalized. We did a $ 122 billion raise
[2:30] Speaker 3: in Q1. And that sits on the balance sheet. That
[2:32] Speaker 3: sits on the balance sheet. That was record-breaking. I always
[2:35] Speaker 3: want to make sure we're well capitalized so that we
[2:38] Speaker 3: have maximum optionality as a company. We don't know what
[2:41] Speaker 3: DOTS is going to do. We're super optimistic. The worst
[2:44] Speaker 3: thing would be not to have the compute to go
[2:45] Speaker 3: serve it to customers.
[2:46] Speaker 1: So you could do a private market round if you
[2:47] Speaker 1: need to? Absolutely.
[2:49] Speaker 3: There's demand in the marketplace if that becomes the right
[2:51] Speaker 3: thing to do.
[2:54] Speaker 2: She makes it seem like, you know, this is the
[2:56] Speaker 2: investors pushing to do a private market round, right, that
[2:59] Speaker 2: they want to do it. Just catch us up on
[3:00] Speaker 2: where OpenAI is at with its journey to the public markets.
[3:05] Speaker 4: Well, they have actually slowed their role, unlike Anthropic, which
[3:09] Speaker 4: seems to be marching closer to an offering as soon
[3:13] Speaker 4: as November. But OpenAI has taken a different track. They
[3:16] Speaker 4: have decided to tap the brakes a little bit and
[3:19] Speaker 4: wait until at least next year before they go public.
[3:22] Speaker 4: But what was interesting in her comments there, Ed, is
[3:25] Speaker 4: that also this whole question of compute. They want to
[3:27] Speaker 4: make sure that they have enough infrastructure in place so
[3:31] Speaker 4: that perhaps they are ready on both sides, not just
[3:34] Speaker 4: with investors, but also with their ability to deliver and
[3:38] Speaker 4: to keep developing the models that not only their investors,
[3:41] Speaker 4: but customers out there are going to be looking for.
[3:43] Speaker 4: So Interesting comments from her saying that, look, there's demand
[3:47] Speaker 4: on the investor side, but also demand on their own
[3:50] Speaker 4: internal side for funding to be able to sustain all
[3:53] Speaker 4: the infrastructure that they will need.
[3:55] Speaker 2: That's on the financial news cycle side. Actually, in technology,
[3:59] Speaker 2: we continue to get the release of models, and one
[4:01] Speaker 2: of them is from NVIDIA-backed Reflection, which has released an
[4:05] Speaker 2: open weight model.
[4:06] Speaker 1: What do we need to know?
[4:08] Speaker 4: Well, they just put out this new model, Ed, called Beam,
[4:11] Speaker 4: and it is something that they say competes with some
[4:15] Speaker 4: of the best open offerings from rivals in China, including
[4:19] Speaker 4: GLM 5.2 from Z.AI, which is one of the rising
[4:23] Speaker 4: stars in China. And of course, China's models lag behind
[4:26] Speaker 4: the U.S. at the very cutting edge, the frontier. But
[4:30] Speaker 4: they are gaining ground thanks to the adaptability and the
[4:33] Speaker 4: affordability of these open weight models at not only Z.AI,
[4:37] Speaker 4: but others that we have talked about on this program before.
[4:39] Speaker 1: Like Moonshot, DeepSeek, and Minimax.
[4:42] Speaker 4: And Reflection wants to get into the game here of
[4:45] Speaker 4: open weight in the U.S. We talk about companies like
[4:49] Speaker 4: OpenAI and Anthropic so much, but their systems are proprietary.
[4:52] Speaker 4: And we have seen Jensen Wang, the CEO of NVIDIA,
[4:56] Speaker 4: advocate for more build-out of an open weight system here
[4:59] Speaker 4: in the U.S. to make these offerings wider spread through
[5:04] Speaker 4: the economy here, but also to offer some security advantages
[5:07] Speaker 4: as well. They are looking to break in. The Wall
[5:09] Speaker 4: Street Journal has said that they have been in talks
[5:12] Speaker 4: to raise $ 2. 5 billion at a $ 25 billion valuation. The
[5:17] Speaker 4: company was founded by two former researchers from Google DeepMind,
[5:21] Speaker 4: and they are going to be one to watch as
[5:23] Speaker 4: their offerings develop. And they also happen to land a
[5:26] Speaker 4: sovereign AI deal through the U.S. government with the Republic
[5:30] Speaker 4: of South Korea recently. So we'll have to keep watching
[5:33] Speaker 4: where they take this Beam product in the future, Ed.
[5:37] Speaker 2: Bloomberg's Mike Shepard, thank you very much indeed. Let's take
[5:40] Speaker 2: a look at today's big number, $ 12 billion. That's how
[5:43] Speaker 2: much DeepSeek's looking to raise in a $ 0. 10- backed funding round,
[5:46] Speaker 2: according to sources, blowing past its original target as it
[5:50] Speaker 2: prepares for an IPO early next year, which could be
[5:52] Speaker 2: one of China's biggest AI debuts yet. Bloomberg Global Executive
[5:56] Speaker 2: Editor Peter Elstrom is with us. Let's start on this
[5:58] Speaker 2: DeepSeek piece then. Very, very similar to what we see
[6:01] Speaker 2: in other markets, right? Do a big private round as
[6:04] Speaker 2: a precursor to go in public.
[6:07] Speaker 5: Yeah, talk about investor demand. DeepSeek went into the market,
[6:10] Speaker 5: as we reported earlier, looking for a little bit less
[6:13] Speaker 5: than $ 10 billion. And then as we're reporting now, they
[6:17] Speaker 5: found strong investor demand to put money into the company.
[6:20] Speaker 5: They could now raise between $ 12 billion and $ 15 billion.
[6:24] Speaker 5: So a very big funding round for these Chinese companies.
[6:27] Speaker 5: And what does that tell us? I mean, first of all,
[6:29] Speaker 5: there's a lot of support from the Chinese government and
[6:32] Speaker 5: Chinese companies for this fundraising. Tencent, the big operator of WeChat,
[6:36] Speaker 5: of course, is going to be one of the lead investors,
[6:39] Speaker 5: as we've heard in this round. Also, CATL, the big
[6:43] Speaker 5: battery maker, is interested in investing in this round. So
[6:47] Speaker 5: that's one. Two, it probably tells you that DeepSeek, which
[6:50] Speaker 5: has largely been an AI lab at this point, coming
[6:53] Speaker 5: out with some very innovative models, but it seems like
[6:56] Speaker 5: they're delving deeper into building their infrastructure. They want to
[6:59] Speaker 5: be able to have the compute the kind of data
[7:01] Speaker 5: centers that we're seeing some of the big labs like
[7:03] Speaker 5: you were talking about with OpenAI and Anthropic. And then
[7:06] Speaker 5: I'd say number three is that there's a disconnect here
[7:11] Speaker 5: in the valuations. If DeepSeek raises this money, they're talking
[7:14] Speaker 5: about a valuation of $ 75 billion, as you were just
[7:17] Speaker 5: talking about with OpenAI. If they reach $ 1. 4 trillion, you're
[7:21] Speaker 5: talking about 20 times as much. I'm not sure that
[7:23] Speaker 5: that's sustainable. Certainly there's a discount in the China market.
[7:26] Speaker 5: There may be some inflation on the U.S. side too
[7:28] Speaker 5: because these companies do compete with each other.
[7:31] Speaker 2: A lot of that analogous with Moonshot AI, which has
[7:34] Speaker 2: closed a private market round we've been reporting on for
[7:37] Speaker 2: quite a long time. But now we have a better
[7:39] Speaker 2: sense of its plans to go to the public markets, too.
[7:43] Speaker 6: Right, right.
[7:44] Speaker 5: Moonshot, of course, surprised investors with this Kimi K3 model
[7:49] Speaker 5: that came out and was very competitive with some of
[7:51] Speaker 5: the models in Silicon Valley. They now are looking at
[7:53] Speaker 5: an IPO at the beginning of 2027. That's roughly going
[7:57] Speaker 5: to be around the time that DeepSeek is seeking to
[7:59] Speaker 5: raise money in the IPO market too. So you've got
[8:02] Speaker 5: two of the leading AI labs from China going into
[8:04] Speaker 5: the public markets aiming for next year. It may depend
[8:07] Speaker 5: a little bit on what we see in the U.S.
[8:10] Speaker 7: Offerings.
[8:10] Speaker 5: We'll get Anthropix almost certainly before then. We probably will
[8:13] Speaker 5: not get OpenAI at that point. But you're going to
[8:16] Speaker 5: have these two companies going into the market, testing the
[8:19] Speaker 5: appetite for investors for their kind of model. Of course,
[8:22] Speaker 5: the business model is very, very different here. They're the
[8:25] Speaker 5: open source models, as you were touching on earlier, much
[8:28] Speaker 5: less profitable at this point, much lower revenue, but they
[8:32] Speaker 5: feel like they have an opportunity to be able to
[8:34] Speaker 5: grow and expand globally.
[8:36] Speaker 2: And Peter, just to double check, so Moonshot AI looking
[8:39] Speaker 2: to Hong Kong for public markets, DeepSeek, it looks like
[8:42] Speaker 2: a mainland China listing.
[8:46] Speaker 5: That's probably the case at this point. I mean, they
[8:49] Speaker 5: need to do some internal restructurings to be able to
[8:52] Speaker 5: proceed with those IPOs, but that's the plan at this point.
[8:55] Speaker 5: And part of it has to do with their investor base.
[8:57] Speaker 5: If you have more of a global investor base, you'd
[8:59] Speaker 5: want to go public in Hong Kong. If you have
[9:01] Speaker 5: more of these Chinese companies investing, you want to go
[9:04] Speaker 5: public in the mainland.
[9:06] Speaker 1: Bloomberg's Peter Elstrom. Thank you very much.
[9:08] Speaker 2: Coming up, Google goes nuclear. The tech giant strikes a
[9:11] Speaker 2: major new power deal to fuel AI's boom.
[9:14] Speaker 1: We have the details of that next. This is Bloomberg Tech.
[9:32] Speaker 2: Google's making another big bet on nuclear power to fuel
[9:35] Speaker 2: the AI boom. The tech giant has struck a 20-year
[9:37] Speaker 2: deal with Constellation Energy that will help unlock nearly 900
[9:40] Speaker 2: megawatts of additional nuclear capacity and drive more than $ 4
[9:44] Speaker 2: billion in upgrades across 11 reactors. Bloomberg's Brody Ford broke
[9:48] Speaker 2: the story. It's just really interesting. A lot of emphasis
[9:50] Speaker 2: on it being new capacity. Constellation is fronting a lot
[9:54] Speaker 2: of the investment that's needed here.
[9:57] Speaker 1: Explain the deal.
[9:58] Speaker 8: There's a significant need.
[10:00] Speaker 1: For new power.
[10:01] Speaker 8: I mean, it feels like every single day we read
[10:02] Speaker 8: stories about data centers being unable to get the power
[10:05] Speaker 8: they need. And so we're seeing tech giants make these
[10:08] Speaker 8: increasingly large bets on power and Nuclear is one that
[10:12] Speaker 8: a lot of people see as very promising, right? It's
[10:14] Speaker 8: emissions free. It helps avoid a lot of the kind
[10:17] Speaker 8: of backlash we're seeing and potentially could come online quicker
[10:20] Speaker 8: than new inter new grid connection. So this is the
[10:24] Speaker 8: largest tech nuclear deal I've seen. And it just shows
[10:27] Speaker 8: that the data center power demand is insatiable.
[10:31] Speaker 2: We are familiar with Constellation in part because last week
[10:33] Speaker 2: I think they did a deal with Amazon that wasn't
[10:35] Speaker 2: quite as big, but also the deal that they did
[10:38] Speaker 2: with Microsoft, for example, Three Mile Island. In aggregate, we're
[10:43] Speaker 2: seeing the hyperscalers go right to the power generators as
[10:46] Speaker 2: opposed to the grid.
[10:47] Speaker 1: Is that kind of a fair summary?
[10:49] Speaker 8: Yeah, I think there's a lot of pressure on them
[10:51] Speaker 8: to add new capacity onto the grid, right? The big
[10:55] Speaker 8: kind of talking point against the tech giants right now
[10:58] Speaker 8: is that they're sucking up power that everybody else wanted.
[11:01] Speaker 8: And so saying that, look, we're adding almost a gigawatt
[11:04] Speaker 8: of new nuclear capacity onto America's largest grid, this is
[11:08] Speaker 8: important for them functionally, but also rhetorically.
[11:11] Speaker 2: You know, 900 megawatts does seem like a lot, but
[11:14] Speaker 2: what other deals has Google done, apart from the one
[11:16] Speaker 2: that's just been announced this morning?
[11:18] Speaker 8: They've done a handful of nuclear deals. I mean, you
[11:21] Speaker 8: also mentioned Microsoft's done some big ones, like Three Mile
[11:23] Speaker 8: Island's supposed to come on next year. And You know,
[11:26] Speaker 8: we're seeing a constant inflow of these power purchase agreements
[11:30] Speaker 8: of bringing on new capacity to the grid. But the
[11:33] Speaker 8: big backlog right now is speed. It's how quickly can
[11:36] Speaker 8: you get the power going. And so by upgrading some
[11:39] Speaker 8: existing reactors, some existing nuclear sites, it seems the bet
[11:43] Speaker 8: is quicker. than some other methods.
[11:46] Speaker 1: All right, just real quick.
[11:47] Speaker 2: So Constellation is going to spend more than $ 4 billion
[11:49] Speaker 2: on this.
[11:50] Speaker 1: What's Google going to spend? How does it work?
[11:52] Speaker 5: Right.
[11:53] Speaker 8: I mean, presumably more than that. Presumably Constellation wants a
[11:57] Speaker 8: profit margin. Our initial story said Google would spend over
[12:01] Speaker 8: a billion, I think, based on what we're seeing in
[12:04] Speaker 8: this announcement, probably quite.
[12:06] Speaker 1: A bit more than that. Bloomberg's Brady Ford, thank you
[12:08] Speaker 1: very much. Thank you.
[12:10] Speaker 2: Another news item, NASA is looking to lock in rockets
[12:13] Speaker 2: for its push to build a permanent base on the moon.
[12:16] Speaker 2: The agency's preparing bulk purchases of launches, potentially buying four
[12:21] Speaker 2: or five rockets at a time as it faces a
[12:23] Speaker 2: looming shortage and SpaceX begins pulling back Falcon 9 availability
[12:28] Speaker 2: beyond 2028. NASA needs heavier launches starting.
[12:31] Speaker 1: In 2029 to deliver equipment.
[12:34] Speaker 2: For its $ 30 billion lunar base effort. Okay, coming up,
[12:39] Speaker 2: we'll be joined by Emily Bender of the University of
[12:42] Speaker 2: Washington and longtime AI LLM critic to break down the
[12:47] Speaker 2: real risks associated with the AI race.
[12:51] Speaker 1: That's next. This is Bloomberg Tech.
[13:05] Speaker 2: OpenAI is apologizing to Australia after an AI agent accessed
[13:10] Speaker 2: a government website it wasn't supposed to, and the government
[13:13] Speaker 2: wasn't told for months. Now the breach is putting AI
[13:16] Speaker 2: oversight and accountability in the spotlight. Bloomberg's Paul Allen reports
[13:19] Speaker 2: from Sydney.
[13:21] Speaker 9: Representatives from Anthropic, Microsoft, Google, all coming here today for
[13:26] Speaker 9: the Select Committee on Artificial Intelligence. The star of the show, however,
[13:31] Speaker 9: was Jason Kwan. the chief strategy officer for OpenAI, and
[13:35] Speaker 9: he began his testimony to the committee with an apology
[13:39] Speaker 9: for a data breach that happened back in June.
[13:42] Speaker 10: I want to begin with an apology. During internal training
[13:46] Speaker 10: and evaluation, our models accessed Australian government websites in ways
[13:51] Speaker 10: they were not directed to. That should not have happened.
[13:55] Speaker 10: We also should have handled our response better. We are sorry.
[14:00] Speaker 10: And we know we have work to do to rebuild
[14:02] Speaker 10: trust with the Australian people.
[14:03] Speaker 9: So it was back in June that an OpenAI agent
[14:06] Speaker 9: managed to access a non-public facing Medicare statistics site. It
[14:10] Speaker 9: wasn't until August that OpenAI found out about it, and
[14:13] Speaker 9: it wasn't until September 10 that it alerted the Australian
[14:16] Speaker 9: government and then did so via email. Jason Kwan apologized
[14:20] Speaker 9: for that delay as well, and he was asked under
[14:23] Speaker 9: questioning whether if OpenAI hadn't told the government, whether.
[14:26] Speaker 1: It still wouldn't know.
[14:28] Speaker 9: And he said that that was a reasonable proposition. Under
[14:30] Speaker 9: another question, he confirmed that Not even Sam Altman, the
[14:33] Speaker 9: OpenAI CEO, knew about the breach when he spoke with
[14:37] Speaker 9: the Australian Deputy Prime Minister Richard Miles earlier in September
[14:40] Speaker 9: as well. There was no personal information or data that
[14:44] Speaker 9: was leaked during that breach but it does underscore the
[14:47] Speaker 9: risks involved. So Jason Kwan and the other representatives of
[14:52] Speaker 9: the AI companies here that expressed a willingness to work
[14:55] Speaker 9: with the government on the legislative framework.
[14:58] Speaker 4: Mr.
[14:58] Speaker 9: Quam has also asked if there are any jurisdictions that
[15:01] Speaker 9: Australia could copy when it comes to doing this. He
[15:04] Speaker 9: pointed to the United States, saying that in the US,
[15:07] Speaker 9: the plan is to get these models into the hands
[15:09] Speaker 9: of trusted actors, into governments first, before they were released
[15:13] Speaker 9: more publicly. And he suggested that this was a model
[15:16] Speaker 9: Australia could follow as a US ally and a Five
[15:18] Speaker 9: Eyes partner as well. This hearing wraps up its work
[15:21] Speaker 9: at the end of November. There's likely to be some
[15:23] Speaker 9: sort of legislation in 2027, but the goal for Australia
[15:27] Speaker 9: here is to manage the risks of this technology without
[15:30] Speaker 9: crushing the opportunities that are provided by it as well.
[15:33] Speaker 1: Paul Allen, Bloomberg, Sydney.
[15:37] Speaker 2: That hearing highlights a bigger question. Are we focused on
[15:40] Speaker 2: the right AI risks? Our next guest says the debate
[15:43] Speaker 2: is too often dominated by doomsday scenarios and extremes, distracting
[15:48] Speaker 2: from harms that are already happening. Joining us now is
[15:51] Speaker 2: Emily Bender, Director of Computational Linguistics at the University of
[15:55] Speaker 2: Washington and author of the AI Con, How to Fight
[15:59] Speaker 2: Big Tech's Hype and Create the Future We Want.
[16:03] Speaker 1: What are the right risks, Professor Bender?
[16:06] Speaker 11: So there are so many risks, and many of us
[16:08] Speaker 11: have been talking about them and documenting them for many years.
[16:11] Speaker 11: I want to start off by pointing to environmental impact,
[16:14] Speaker 11: and that is both in the energy production, but also
[16:16] Speaker 11: water use and the production of the chips that go
[16:19] Speaker 11: in the data centers. I want to point to the
[16:21] Speaker 11: degradation of our information ecosystem. Now that it is cheap
[16:24] Speaker 11: for people to create fluent-looking text and plausible-looking images, that
[16:28] Speaker 11: has huge harms. And I want to point to the
[16:31] Speaker 11: gigification of the workplace. So many workers are being displaced
[16:34] Speaker 11: because of this myth that so-called AI can do their job.
[16:37] Speaker 11: And then they're displaced into jobs where they're basically providing
[16:40] Speaker 11: the backstop for that janky software. We're seeing concentration of power.
[16:44] Speaker 11: We're seeing the motivation and extension of surveillance systems. And
[16:47] Speaker 11: also on an individual level, we're seeing people becoming dependent
[16:51] Speaker 11: on chatbot companions, often leading to grievous self-harm.
[16:54] Speaker 7: And that's not all.
[16:55] Speaker 11: That's just what I could do off the top of
[16:57] Speaker 11: my head. Right.
[16:59] Speaker 2: Professor Ben, I just want to point out on the
[17:01] Speaker 2: water use, because this has come up in the program.
[17:04] Speaker 2: If you go back to June, NVIDIA was talking about
[17:06] Speaker 2: this data from the Manhattan Institute where data centers use 0.2%
[17:11] Speaker 2: of daily, 0.2% of daily water usage in the US
[17:15] Speaker 2: and that that's a number that's becoming dramatically lower. I
[17:21] Speaker 2: think we're having this conversation in the context of the
[17:24] Speaker 2: actions an AI could take and the risk presented if
[17:28] Speaker 2: those actions are outside of what we intended the AI
[17:31] Speaker 2: to do, right? But right now there are loads of
[17:34] Speaker 2: case studies like Australia we just listened to.
[17:37] Speaker 1: Mm-hmm, mm-hmm.
[17:39] Speaker 11: That is where the doomsday scenario people would like us
[17:43] Speaker 11: to look because they have this shared fiction of AI
[17:46] Speaker 11: going rogue and then somehow improving itself and then taking
[17:49] Speaker 11: over the world. What's actually happening here is that OpenAI
[17:52] Speaker 11: and others are creating and running malware.
[17:55] Speaker 7: And then when their malware.
[17:57] Speaker 11: Accesses things it's not supposed to access, they say, look
[18:00] Speaker 11: how clever and scary our software is. Instead of saying,
[18:05] Speaker 11: we did this, we're saying, they say the software is
[18:07] Speaker 11: doing it. And that was all throughout that report, that
[18:10] Speaker 11: anthropomorphizing language.
[18:11] Speaker 7: So what happened?
[18:13] Speaker 11: OpenAI designed a system to hack into websites and ran it,
[18:18] Speaker 11: and it.
[18:18] Speaker 7: Hacked into the websites.
[18:21] Speaker 11: That doesn't mean that an AI has gone rogue.
[18:23] Speaker 1: Yes.
[18:24] Speaker 2: What's interesting about this is the wide-ranging schools of thought
[18:28] Speaker 2: about what to do about it. And one of those
[18:30] Speaker 2: schools of thought is if an AI behaves in the
[18:34] Speaker 2: way you just described, it's behaving illegally. And the laws
[18:38] Speaker 2: of the land already provide for illegality for such a breach.
[18:44] Speaker 2: In your field of research, do you see that to
[18:46] Speaker 2: be the case, that if an AI were to conduct
[18:49] Speaker 2: an action that could be defined as illegal, there would
[18:52] Speaker 2: be sufficient safeguard in place to act on that?
[18:56] Speaker 1: So there is.
[18:57] Speaker 11: Sufficient safeguard, but it's not that an AI, which doesn't
[19:01] Speaker 11: refer to something, you know, it's not that a piece
[19:03] Speaker 11: of software is doing something illegal. It's that companies are
[19:07] Speaker 11: doing things that are legal. Open AI hacked into Hugging
[19:10] Speaker 11: Face and the Australian government and all these other things,
[19:13] Speaker 11: not some independent, unaccountable piece of software. And if you
[19:17] Speaker 11: look back just to the previous administration, we had the
[19:19] Speaker 11: FTC under Lina Khan saying exactly that, saying there is
[19:22] Speaker 11: no exemption for so-called AI in the laws. The FTC
[19:27] Speaker 11: regulates the activities of businesses. It doesn't matter what automation
[19:30] Speaker 11: they're using to do those activities.
[19:33] Speaker 2: I do want to talk about those extremes. whether or
[19:36] Speaker 2: not AI presents an existential threat to human life. And
[19:41] Speaker 2: there was the former anthropic researcher that put a figure.
[19:44] Speaker 1: On it, 10%.
[19:47] Speaker 2: Sam Altman's point when he testified before the UN Security
[19:50] Speaker 2: Council was it doesn't matter what the chance or the
[19:52] Speaker 2: percentage risk is. It's too great either way. Is it
[19:56] Speaker 2: important to acknowledge that, that risk is real?
[20:00] Speaker 11: No, the risk is actually fake. And this whole debate
[20:03] Speaker 11: about percentages is very funny. I have yet to see
[20:05] Speaker 11: the reporter who asked the anthropic lever, whistleblower, where he
[20:10] Speaker 11: got that number. I think the number just gives us
[20:12] Speaker 11: this veneer of mathiness. And then Sam Alton is trying
[20:15] Speaker 11: to back off of that with his remarks. But this
[20:17] Speaker 11: is all based on science fiction and based on thought
[20:20] Speaker 11: experiments that come out of effective altruism that start from
[20:24] Speaker 11: the premise that we have an all-powerful so-called artificial intelligence
[20:29] Speaker 11: and then say what could happen then. But the premise
[20:32] Speaker 11: is false and the premise is imaginary.
[20:34] Speaker 1: So it doesn't matter.
[20:35] Speaker 11: And meanwhile, we do have real problems that are being
[20:39] Speaker 11: swept under the rug. So Sam Altman suggesting that Australia
[20:42] Speaker 11: should look to the US for how to regulate this
[20:44] Speaker 11: stuff is bananas. We are not doing a good job
[20:47] Speaker 11: right now because we are letting all of the companies
[20:49] Speaker 11: off the hook.
[20:51] Speaker 2: Professor Bender, what benefits and net positives do you see
[20:55] Speaker 2: from AI in our society?
[21:00] Speaker 11: So benefits for synthetic media, basically none. The benefits Killer
[21:05] Speaker 11: app for synthetic text, for synthetic images is fraud. If
[21:08] Speaker 11: you want to look at all of the different technologies
[21:11] Speaker 11: that have been sold as AI, some of them are
[21:13] Speaker 11: legitimate kinds of automation. Like I work in computational linguistics.
[21:17] Speaker 11: Language technology can be quite useful. I'm talking about automatic transcription,
[21:21] Speaker 11: machine translation, spell check, right? But I think that we
[21:25] Speaker 11: do a worse job making decisions about this automation when
[21:28] Speaker 11: we call it artificial intelligence instead of getting specific about
[21:31] Speaker 11: the technology.
[21:35] Speaker 1: What do you think needs to happen here?
[21:37] Speaker 2: At the end of the day, all of your arguments
[21:40] Speaker 2: are still rooted in the question, do we regulate or not?
[21:43] Speaker 6: Yes.
[21:44] Speaker 11: So I do want to see existing regulation applied. I
[21:47] Speaker 11: am very concerned at recent moves in the EPA to
[21:50] Speaker 11: exempt power producers from environmental regulations, for example. I would
[21:56] Speaker 11: love to see any company that sets up a system
[22:00] Speaker 11: where people can go get synthetic text actually held accountable
[22:04] Speaker 11: for all of the text that comes out of that system.
[22:06] Speaker 11: I would love to see existing regulations about who can be,
[22:11] Speaker 11: for example, a doctor or a psychotherapist actually applied so
[22:15] Speaker 11: that we don't get people setting up systems where chatbots
[22:17] Speaker 11: are supposedly filling those roles. And I would like to
[22:20] Speaker 11: see stronger protections for labor organizing because a lot of
[22:24] Speaker 11: the issues around gigification of the workplace come down to
[22:27] Speaker 11: tech companies saying, we can do your job and then
[22:31] Speaker 11: workplace getting worse.
[22:33] Speaker 2: Professor Emily Bender of the University of Washington. Let's go
[22:35] Speaker 2: to private markets. Anthropic is heading toward one of the
[22:39] Speaker 2: most anticipated IPOs in tech. T. Rowe Price is already
[22:42] Speaker 2: a major investor. Emma Norchette is a partner on T.
[22:45] Speaker 2: Rowe Price's Private Opportunities Fund Investment Committee, helping oversee investments
[22:50] Speaker 2: across some of the biggest private companies in AI and tech.
[22:54] Speaker 2: And I'm delighted to say Emma joins us now. Welcome
[22:56] Speaker 2: to Bloomberg Tech.
[22:57] Speaker 6: Thank you so much for having me.
[22:59] Speaker 2: Let's just go to the thesis on Anthropic. What you
[23:02] Speaker 2: and T-Rose saw as the opportunity.
[23:05] Speaker 6: Yeah, so we've been a big shareholder for Anthropic. We've
[23:09] Speaker 6: been across the company for almost three rounds right now.
[23:12] Speaker 6: I think we got really, really excited when we did
[23:15] Speaker 6: the first round last year.
[23:16] Speaker 1: When was that?
[23:18] Speaker 6: Last year, we participated in a $ 170 billion round. That
[23:22] Speaker 6: was our first entry point. It was over the summer.
[23:25] Speaker 6: And we got really excited about the product market fit
[23:28] Speaker 6: that they had demonstrated with coding. It was probably one
[23:32] Speaker 6: of the largest markets.
[23:33] Speaker 7: They were very focused.
[23:35] Speaker 6: And they had really clearly demonstrated their ability to onboard
[23:39] Speaker 6: all the major enterprise customers. Now, fast forward to today,
[23:44] Speaker 6: I think this was act one of Entropic. And I
[23:49] Speaker 6: think what's really exciting now is moving forward, I think
[23:52] Speaker 6: we are really thinking for large language models, about Act Two.
[23:56] Speaker 6: And I would tease this internally is that it moves
[23:59] Speaker 6: now to personal and enterprise operating system. And what these
[24:04] Speaker 6: companies are now going to be able to do is
[24:08] Speaker 6: not only summarize information, address questions, but really take action.
[24:12] Speaker 6: And I think if we were looking into this, they
[24:15] Speaker 6: don't need to replace every single application. What they need
[24:19] Speaker 6: to do, they need to be the layer where every
[24:22] Speaker 6: single type of workflow gets done.
[24:24] Speaker 1: Okay, this is interesting. I want to keep going into
[24:27] Speaker 1: Act 2.
[24:28] Speaker 2: On the enterprise side, it's easier to understand because Anthropic
[24:30] Speaker 2: got a lot of credit for their early move in
[24:33] Speaker 2: the enterprise. A more recent example, for example, has been Salesforce.
[24:36] Speaker 2: So through Claude, you could run CRM through Claude.
[24:42] Speaker 1: But you said personal.
[24:43] Speaker 4: Yes.
[24:43] Speaker 1: We're in the age of the AI agent, right? Grok, Bot, Muse.
[24:47] Speaker 1: Does Anthropic have something there?
[24:49] Speaker 2: What's the economic opportunity with that?
[24:51] Speaker 6: Yeah, so I think we will continue to see kind
[24:54] Speaker 6: of blurring between enterprise and personal. I think what is
[24:58] Speaker 6: probably most recent and that would probably work talking about
[25:02] Speaker 6: is talking a little bit more about OpenAI.
[25:05] Speaker 7: And Dev Day that happened.
[25:06] Speaker 1: Yeah, how was that?
[25:07] Speaker 6: Last week. And I think one of the most anticipated
[25:11] Speaker 6: or most exciting release from Dev Day was really Dots.
[25:15] Speaker 4: Yep.
[25:16] Speaker 6: And I think this is where we are starting to
[25:18] Speaker 6: see personal agents that we've all kind of discovered over
[25:21] Speaker 6: the last two to three weeks.
[25:23] Speaker 1: So where's Entropic?
[25:24] Speaker 6: Really entering the enterprise layer. I think Entropic today, if
[25:30] Speaker 6: we're looking into it, it's a tier where we have
[25:33] Speaker 6: a major partnership with them on the enterprise side. And
[25:36] Speaker 6: I do think that we will soon be able to
[25:39] Speaker 6: use Entropic to take action for enterprise as well.
[25:43] Speaker 2: Emma, you're a really interesting case study, not necessarily a
[25:46] Speaker 2: unique case study of an investor who has positions in
[25:50] Speaker 2: OpenAI and Anthropic. So how do you distinguish between them?
[25:55] Speaker 6: I think we don't need to distinguish. I think the
[25:58] Speaker 6: bet that really has been taken by Tiro is that
[26:01] Speaker 6: there's going to be a major value capture that's going
[26:04] Speaker 6: to go to the models that can stay on the frontier.
[26:07] Speaker 6: And today, if we're looking into it, I think both
[26:10] Speaker 6: OpenAI and Anthropic have been very well positioned. I think
[26:15] Speaker 6: we are very, very excited about this act too, because
[26:20] Speaker 6: when we are looking into it, I think despite, for example,
[26:24] Speaker 6: OpenAI's 1.2 billion weekly active users, there's probably only a
[26:31] Speaker 6: few percentage of the action that these users are taking
[26:36] Speaker 6: that leads to action.
[26:37] Speaker 7: And then I think.
[26:38] Speaker 1: Monetizable action, you mean, or just.
[26:41] Speaker 6: Monetizable or just simply action where we can really see
[26:46] Speaker 6: the ROI on what is happening with efficiency, but also
[26:52] Speaker 6: completed task. And then when we're looking a little bit
[26:56] Speaker 6: more into Entropic, I think what is also super exciting
[27:01] Speaker 6: there is as every single verticals that they can tackle
[27:05] Speaker 6: going forward. And I think one of the chapter that
[27:08] Speaker 6: Entropic has been public about, and I think for us
[27:13] Speaker 6: is crucial and we're seeing as a major chapter coming
[27:16] Speaker 6: up is really AI for science. And I think we're
[27:20] Speaker 6: going to see a lot more happening in that space.
[27:22] Speaker 2: But on your value capture thesis, you do see room
[27:26] Speaker 2: for two frontier labs that if they stay at the frontier,
[27:29] Speaker 2: Then economic benefit accrues to both.
[27:31] Speaker 6: Absolutely. I think for us, we don't see as winner
[27:35] Speaker 6: takes all. I think we will see a few players
[27:38] Speaker 6: that can dominate. And I think we have seen so
[27:41] Speaker 6: far the position for that with Anthropic and OpenAI.
[27:45] Speaker 2: Is the exposure from T-Road to Anthropic and OpenAI comparable?
[27:49] Speaker 6: So for a boat company, we have multi-billions. It's among
[27:53] Speaker 6: some of our largest position that we have on the
[27:56] Speaker 6: private markets. The other one that's kind of out there
[27:59] Speaker 6: in similar sizes is Databricks.
[28:01] Speaker 1: That's right.
[28:02] Speaker 6: Those are kind of three big bets that we've been
[28:04] Speaker 6: doing in kind of data, infra, and AI.
[28:07] Speaker 2: So the three of those things that those companies have
[28:10] Speaker 2: in common is they're waiting to go public in different
[28:13] Speaker 2: stages of that. I think one reason I was really
[28:15] Speaker 2: interested to have you on the show is like, The
[28:17] Speaker 2: game's kind of changed from when I first arrived to
[28:20] Speaker 2: Silicon Valley.
[28:21] Speaker 1: You see T.
[28:23] Speaker 2: Rowe much more often on the press release, co-leading rounds,
[28:28] Speaker 2: earlier into the cap table.
[28:31] Speaker 1: What does that reflect?
[28:32] Speaker 6: Yeah, so T. Rowe has been doing private investing for
[28:37] Speaker 6: almost two decades. But I think we are really doubling
[28:41] Speaker 6: down on our private investing practices. And I think as
[28:44] Speaker 6: we are looking into the landscape today, a lot of
[28:49] Speaker 6: what is happening at the frontier of AI is still
[28:51] Speaker 6: happening on kind of private markets. We've also seen other
[28:55] Speaker 6: noticeable trend, for example, defense tech and a lot of that.
[28:59] Speaker 1: You're in Anduril, for example.
[29:01] Speaker 6: Yeah, Anduril is one that we are very excited about.
[29:05] Speaker 6: And then Ramp on the financial side is also super exciting.
[29:11] Speaker 6: But I think as we are looking into these companies,
[29:13] Speaker 6: all of this growth is all private. So if we
[29:16] Speaker 6: want to access it, we need to be earlier in
[29:18] Speaker 6: those companies. So for us entering these companies at large
[29:23] Speaker 6: series B, series C, building conviction, getting to know management team,
[29:28] Speaker 6: helping them through that transition and getting them on the
[29:31] Speaker 6: other side of the IPO and being a bit more
[29:35] Speaker 6: long-term is really, I think, where we can distinguish and
[29:39] Speaker 6: provide value to these companies.
[29:42] Speaker 2: Look, OpenAI and Anthropic have filed confidentially. Databricks, everyone's wondering about.
[29:47] Speaker 2: I won't ask you to go into specifics. But generally,
[29:50] Speaker 2: you think about those case studies, how do you know
[29:53] Speaker 2: when those companies are ready to go public?
[29:56] Speaker 1: And what's T.
[29:56] Speaker 2: Rowe's role in advising, right, when you're a private investor
[30:00] Speaker 2: on the cap table?
[30:02] Speaker 6: Yeah, I think there is a misconception sometimes that if
[30:05] Speaker 6: you take money from T. Rowe, you're going to be
[30:07] Speaker 6: pushed to kind of go public at one point in time.
[30:10] Speaker 6: I think for us, we are patient capital. And I
[30:13] Speaker 6: think what we want to be doing is backing generational companies.
[30:18] Speaker 6: and ensuring that they are in a position where they
[30:21] Speaker 6: can choose their own timeline as to when they are
[30:24] Speaker 6: ready for public markets. And then I think if we're
[30:28] Speaker 6: looking into a lot of these companies, many of them
[30:31] Speaker 6: are ready today. So it's more of a decision of
[30:35] Speaker 6: the management team as to when they think is the
[30:37] Speaker 6: right timing. What we can say on our side is
[30:40] Speaker 6: that there will be demand for these kind of companies
[30:44] Speaker 6: and appetite from the public markets. And we are able
[30:48] Speaker 6: to kind of prepare these companies several quarters or years
[30:53] Speaker 6: ahead of the crossing by telling them what are going
[30:57] Speaker 6: to be the questions and things people will be focusing
[30:59] Speaker 6: on on the other side.
[31:00] Speaker 1: Emma Norchette of T.
[31:01] Speaker 2: Rowe Price, your first time with us on Bloomberg Tech,
[31:04] Speaker 2: but I'm looking forward to many conversations to come. Thank
[31:06] Speaker 2: you very much indeed. Coming up on the show, Olivia
[31:09] Speaker 2: Moore from A16Z joins to talk about the top consumer
[31:12] Speaker 2: AI apps. We may continue some of what we just
[31:15] Speaker 2: discussed with Emma. That conversation next. This is Bloomberg Tech.
[31:32] Speaker 2: Andreessen Horowitz has released its list of the top 100
[31:35] Speaker 2: generative AI consumer apps. The latest edition features familiar names
[31:39] Speaker 2: from the previous editions, but also finds there is room
[31:42] Speaker 2: for companies to deepen their.
[31:44] Speaker 1: Engagement with users.
[31:46] Speaker 2: Olivia Moore, A6NZ partner and the author of That report
[31:49] Speaker 2: is with me here in San Francisco. So this is interesting.
[31:53] Speaker 2: There's different ways you slice it up. You do the
[31:55] Speaker 2: top 50 consumer AI apps by monthly revenue or by
[31:57] Speaker 2: monthly active users. And those are different measures, obviously. But
[32:01] Speaker 2: I want to go to this idea straight away of
[32:02] Speaker 2: deepening engagement. What is that a measure of?
[32:06] Speaker 7: It's a great question. So this is actually one of
[32:07] Speaker 7: the big reasons that we wanted to introduce revenue data
[32:10] Speaker 7: this time. So typically, we've done this list just by
[32:13] Speaker 7: visits to websites and monthly active users.
[32:15] Speaker 1: We just showed the unique visits, right?
[32:17] Speaker 7: Yes, which is great. But it doesn't really capture very
[32:20] Speaker 7: deep AI usage, which is happening in products that often
[32:23] Speaker 7: live on the desktop, in things that aren't captured by
[32:27] Speaker 7: that data and that people are paying for. And so
[32:29] Speaker 7: there were some... Very surprising conclusions from that. We finally
[32:34] Speaker 7: got data from Yipit Data that looks across hundreds of
[32:37] Speaker 7: thousands of U.S. consumers. Less than 5% of them are
[32:41] Speaker 7: paying for even one AI app. Only about 13% of
[32:46] Speaker 7: those are paying for more than one. And what we
[32:49] Speaker 7: found was that the top 1% of spenders are driving AI.
[32:52] Speaker 7: most of the revenue.
[32:53] Speaker 2: That's why this table is so interesting though, like OpenAI
[32:55] Speaker 2: number one, Anthropic number two.
[32:57] Speaker 1: Anyone surprised by that?
[32:59] Speaker 7: That is not surprising at all.
[33:00] Speaker 1: Is that just first.
[33:01] Speaker 2: Mover advantage because they were first to productize the tech?
[33:05] Speaker 1: Yes.
[33:05] Speaker 7: I think two things are happening there. When you look
[33:07] Speaker 7: at both OpenAI and Anthropic on these lists, one, they
[33:10] Speaker 7: have the broadest base of consumers. Everyone knows them. They're
[33:13] Speaker 7: general LLMs. You can use them for anything. Two, the
[33:16] Speaker 7: reason why they're topping these monthly revenue charts is is
[33:19] Speaker 7: they really benefit from these power users who are using
[33:21] Speaker 7: these models mostly for coding, mostly for product building.
[33:25] Speaker 1: Token spend. Exactly. Let's go.
[33:27] Speaker 7: That top 1% of spenders that we found are spending
[33:32] Speaker 7: on personal credit cards almost $ 1, 000 a month.
[33:35] Speaker 1: Just to reiterate, again, we're talking about consumer AI apps.
[33:38] Speaker 4: Yes.
[33:39] Speaker 7: As consumers, they are spending completely independent of their work spend,
[33:42] Speaker 7: which is probably much higher, almost $ 1, 000, whereas the average consumer.
[33:47] Speaker 7: Is spending maybe $ 20 a month on AI across one product,
[33:51] Speaker 7: probably something like.
[33:52] Speaker 2: A- That data is fascinating to me. because I think
[33:55] Speaker 2: you and I have discussed this in the past. We
[33:56] Speaker 2: were talking about it before we came on air. It's
[33:59] Speaker 2: like at home, the audit that my wife and I do,
[34:01] Speaker 2: it used to be like streaming, right?
[34:04] Speaker 1: Like I can't have all of these streaming subscriptions.
[34:06] Speaker 2: Now it's like, well, Bloomberg provides me with enterprise version
[34:09] Speaker 2: of ChatGPT. I also have my own personal one. Like,
[34:12] Speaker 2: should I have both? Do I need Gemini, Chord? Like
[34:16] Speaker 2: year to year or edition to edition, how is that
[34:18] Speaker 2: reflected in the changes on the table?
[34:20] Speaker 1: Yeah.
[34:21] Speaker 7: So I would say actually the biggest thing that we're
[34:23] Speaker 7: looking for to make change on the next few tables
[34:26] Speaker 7: is business models beyond subscription. So almost all of consumer
[34:30] Speaker 7: AI revenue has come from subscription and extra credits and
[34:34] Speaker 7: usage tokens purchased for that. So probably like 86% of
[34:38] Speaker 7: companies are offering that. That is completely unnatural in the
[34:42] Speaker 7: history of the consumer internet. The consumer internet has run...
[34:45] Speaker 7: on ads and on transactions. If you look at Meta,
[34:49] Speaker 7: it's like 97% of their revenue is ads. Google, I
[34:51] Speaker 7: think it's like 75%.
[34:52] Speaker 2: But Meta came with the subscription-based AI offering.
[34:57] Speaker 1: I know.
[34:57] Speaker 7: So I think more companies are expanding into subscription. Honestly,
[35:01] Speaker 7: I would expect for Meta, that's probably because they have
[35:04] Speaker 7: a lot of those whales spending an enormous amount or
[35:07] Speaker 7: costing an enormous amount in tokens and they need some
[35:09] Speaker 7: way to monetize them. But I think part of the reason,
[35:12] Speaker 7: back to your earlier question, why haven't we seen deeper
[35:14] Speaker 7: consumer AI usage? It's because almost all of these products,
[35:18] Speaker 7: it costs to use them and it's not cheap. And
[35:21] Speaker 7: most people either can't or don't want to pay for software,
[35:24] Speaker 7: especially more than one product. And so I think this
[35:27] Speaker 7: is going to be really interesting with the rise of
[35:29] Speaker 7: consumer agents like Instinct and Muse that are free. Maybe
[35:32] Speaker 7: they'll monetize via ads. Maybe they'll take a cut of
[35:35] Speaker 7: a transaction like a typical marketplace. I think that has
[35:38] Speaker 7: a much better chance of deepening consumer engagement.
[35:41] Speaker 2: I'm going to ask the control room to be nimble here,
[35:43] Speaker 2: and you tell me which table is best to answer
[35:45] Speaker 2: this question. But these apps, they're all different things. Some
[35:51] Speaker 2: of them are just generative AI tools, text. Some of
[35:54] Speaker 2: them are video models. Some of them are design tools.
[35:57] Speaker 2: I saw Figma is listed on there. It's hard to
[36:01] Speaker 2: compare them, right, if they serve different end uses.
[36:04] Speaker 5: Yes.
[36:05] Speaker 7: They serve different end uses, and yet the power.
[36:08] Speaker 1: Which table? Which table is best to answer this question?
[36:10] Speaker 7: Oh, that's a good question. I think we have one
[36:12] Speaker 7: probably about power user spend over time.
[36:16] Speaker 1: We have monthly active users up now.
[36:17] Speaker 7: Okay, let's keep this up.
[36:18] Speaker 1: Let's go with it.
[36:18] Speaker 3: Okay.
[36:19] Speaker 1: Yeah.
[36:19] Speaker 7: So I think what you'll see among the power users
[36:23] Speaker 7: is they're using basically three buckets of products. And we're
[36:27] Speaker 7: missing a lot of the rest of that in the
[36:29] Speaker 7: consumer AI world so far. So what we're seeing a
[36:32] Speaker 7: lot of, and probably either the revenue list or the
[36:34] Speaker 7: web list is a better one to show this.
[36:35] Speaker 1: All right, guys, let's try to pull up revenue.
[36:37] Speaker 7: Mobile has a lot of their own kind of mostly
[36:40] Speaker 7: photo-based use cases. And you'll see AI avatars, homework help. Yeah,
[36:43] Speaker 7: this is perfect. So you'll see a lot of it
[36:46] Speaker 7: concentrates in three things. So one would be... building products,
[36:50] Speaker 7: so that would be things like Lovable, Fall, Replit, N8N, Suno.
[36:55] Speaker 7: Those are pretty much developer tools, but we're seeing heavy
[36:59] Speaker 7: consumer spend on them to make things. Then we have
[37:02] Speaker 7: things to market products, so you're seeing things on there
[37:05] Speaker 7: like Runway, like Kling. Higgs field is a really big one.
[37:09] Speaker 7: And then the last category is just general productivity. How
[37:13] Speaker 7: do you organize and distribute all that work that you're
[37:16] Speaker 7: doing with AI now? So things like GenSpark, Manus, Granola.
[37:19] Speaker 1: Right.
[37:20] Speaker 7: And so you'll notice what's missing here is everything else
[37:24] Speaker 7: in consumer, like social apps, dating apps, marketplaces, fintech, health.
[37:31] Speaker 7: None of these things.
[37:32] Speaker 1: Your point is those are ripe. for opportunity.
[37:34] Speaker 7: They're right for startups. Yes, because all of the consumer
[37:38] Speaker 7: products we've seen thus far have had to make money
[37:40] Speaker 7: because model costs are so high. And so they're products
[37:43] Speaker 7: that people are using for work, even in their stance
[37:47] Speaker 7: as a consumer.
[37:47] Speaker 2: So Olivia, you are a partner at Andreessen Horowitz, right?
[37:50] Speaker 2: There must be a reason the firm produces this.
[37:54] Speaker 1: How does it inform where you invest next?
[37:57] Speaker 7: Yeah, it gives us a really good sense of what's
[37:59] Speaker 7: happening now, which is About half of Americans are using AI, 25%
[38:03] Speaker 7: of them are using it daily, but still sub 5%
[38:06] Speaker 7: are paying. So then we kind of asked ourselves why,
[38:09] Speaker 7: what needs to change for that number to go up?
[38:12] Speaker 7: And I think the real thing for me is a
[38:15] Speaker 7: lot of these consumer AI products are oriented around helping
[38:19] Speaker 7: consumers save time. and build things. And in reality, most
[38:23] Speaker 7: consumers are looking for better ways to spend their time,
[38:26] Speaker 7: social apps, entertainment, other things with network effects. There's been
[38:30] Speaker 7: almost no multiplayer AI app so far, which is pretty surprising.
[38:34] Speaker 7: And so doing this list and seeing what's not on
[38:36] Speaker 7: this list gives us a lot of excitement and inspiration
[38:39] Speaker 7: for where startups are going.
[38:40] Speaker 1: Looking for it.
[38:40] Speaker 2: Olivia Moore, partner at ASICCZ. It's great to have you
[38:42] Speaker 2: back on the show.
[38:43] Speaker 7: Thanks for having me.
[38:44] Speaker 1: Thank you for being here, Anessa. The debate over the.
[38:46] Speaker 2: Data center build-out is not isolated to the U.S., though
[38:49] Speaker 2: it is poised to be a key factor in the
[38:51] Speaker 2: upcoming midterm elections. JPMorgan Chase Chairman and CEO Jamie Dimon
[38:55] Speaker 2: spoke with Bloomberg's Tom McKenzie about the issue from the
[38:58] Speaker 2: company's textiles conference in London.
[39:01] Speaker 12: Data centers, you build them where they're wanted. you know,
[39:05] Speaker 12: and half of them, half the new ones in America,
[39:07] Speaker 12: Texas and Virginia. And, you know, people don't want them there.
[39:10] Speaker 12: You're not going to win.
[39:11] Speaker 7: You get lawsuits.
[39:12] Speaker 12: But there are a lot of states who would take
[39:14] Speaker 12: it that have access to power. And of course, you know,
[39:17] Speaker 12: any big company has to be a good community citizen.
[39:19] Speaker 12: So I think a lot of these people are doing it.
[39:21] Speaker 12: They're going into town. They're being part of the community.
[39:24] Speaker 12: They're helping the community get the rights. They make commitments
[39:26] Speaker 12: that won't raise consumer electricity costs. It's just things like that.
[39:30] Speaker 12: And that'll be worked through. And AI has these huge
[39:33] Speaker 12: positive effects. I mean, I do think it'll cure a
[39:35] Speaker 12: lot of diseases. And it'll save accidents in the roads
[39:39] Speaker 12: and composite materials. It'll make life unbelievable for people. The
[39:44] Speaker 12: downside is obviously what you read about with the agents
[39:49] Speaker 12: and mythos and all these things that can cause trouble.
[39:51] Speaker 12: And that's a legitimate concern. It's a real thing. I'm
[39:54] Speaker 12: not going to get hysterical over is it existential or not.
[39:57] Speaker 12: What we're doing is rolling up our sleeves and going
[40:00] Speaker 12: to work to fix it. So the banks have always
[40:02] Speaker 12: had a system in place where we help each other.
[40:05] Speaker 12: We also started a thing called the Alliance for Critical Infrastructure.
[40:08] Speaker 12: It's starting with 50 companies. It's starting in the U.S.
[40:11] Speaker 12: It's across six industries. you know, tech, financial services, water, transport.
[40:17] Speaker 12: These are industries that the government pointed out as lifeline,
[40:20] Speaker 12: but eventually it should probably expand to healthcare, pharma, agricultural.
[40:24] Speaker 12: I mean, every industry is kind of critical, and it
[40:27] Speaker 12: should also, we need to help the smaller companies too,
[40:29] Speaker 12: because they have different vendors and different needs, and so
[40:32] Speaker 12: it's a lot of work. We simply have to do it.
[40:34] Speaker 12: AI created vulnerabilities that we didn't know about, and we
[40:38] Speaker 12: always worried about cyber before these things.
[40:40] Speaker 1: And you've thought about cyber risks for a long time.
[40:42] Speaker 12: Yeah.
[40:42] Speaker 1: I know that.
[40:43] Speaker 12: It's our biggest risk.
[40:44] Speaker 1: It's your biggest risk.
[40:45] Speaker 9: Does AI make cyber systemic risks more likely?
[40:50] Speaker 1: Oh, yeah.
[40:50] Speaker 12: That's what I'm saying. That mythos, whatever you thought it
[40:53] Speaker 12: was before. And when I posed my chairman's letter, I said,
[40:56] Speaker 12: I said, cyber is our biggest risk. And they didn't
[40:58] Speaker 12: publish mythos yet. And I would say that went up
[41:00] Speaker 12: tenfold after mythos.
[41:03] Speaker 2: JP Morgan CEO Jamie Dimon speaking with Bloomberg's Tom McKenzie.
[41:07] Speaker 1: One more thing before you go.
[41:08] Speaker 2: Paramount has completed its $ 110 billion purchase of Warner Brothers.
[41:13] Speaker 2: A new company is called Skydance. David Ellison will continue
[41:17] Speaker 2: to serve as CEO and chairman. The former head of Mattel,
[41:20] Speaker 2: Enon Crease, joins him.
[41:22] Speaker 1: As the co-CEO.
[41:23] Speaker 2: Some other familiar names from the merging companies will remain
[41:26] Speaker 2: in leadership roles. but cuts throughout the ranks for Skydance
[41:30] Speaker 2: staff may be coming. Ellison agrees, saying, quote, integrating two
[41:34] Speaker 2: companies will bring change, including difficult decisions that affect our workforce.
[41:38] Speaker 2: That's according to a company memo sent to staff obtained
[41:42] Speaker 2: by Variety.
[41:43] Speaker 1: It's not the only change coming.
[41:44] Speaker 2: Recall, to get the deal over the finish line, it
[41:47] Speaker 2: committed to getting 30 films in theaters a year, investing
[41:50] Speaker 2: $ 1. 5 billion in domestic film production over five years, and
[41:53] Speaker 2: creating independent news oversight board.
[41:56] Speaker 1: Now that it controls CBS News and CNN.
[41:57] Speaker 2: Class B shares opened higher, began trading today on the
[42:01] Speaker 2: MRC under ticker SKYD. We're off now by 4.7%. That
[42:06] Speaker 2: does it for this edition of Bloomberg Tech. Recap on
[42:09] Speaker 2: the podcast. You can find it on the terminal as
[42:11] Speaker 2: well as online, Apple, Spotify, iHeart.
[42:13] Speaker 1: This is Bloomberg Tech.

Shows are listed from Apple Podcasts' charts and directory, and Podcast Index where it is used. Every show and episode belongs to its publisher and plays from the publisher's own feed; this site hosts none of them.