Alibaba's AI Spending Spree, Concerns of Circular AI Financing

22:21You'll have everything you need to scale your go-to market efforts. Elevate your wins with Adio. Start your free trial at adio.com slash iHeart.

22:36Ed Ludlow:Cubanoid robots could take off commercially within just two years. That's according to the founder of China's Unitary Robotics. This comes as the World Robot Conference is in full swing in Beijing, highlighting the industry is moving beyond viral entertainment and focusing on real-world uses from assembly lines to household chores. Bloomberg's Alan Wan has the report. Robots that can sing, dance, and do kung fu. That's all passe at this year's World Robot Conference in Beijing, where thousands of companies are here to showcase their latest human robots. But this time, we'll focus on the practical uses, not just for households, but also for industry.

23:15I've seen robots that can do the work of assembly line workers, but with precision. I've seen also robots that can pick up laundry and put them in washing machines. But it's not all serious business. There are also robots that can play ping pong with you, as well as table hockey. One of the highlights that this year's forum is Unitree CEO and founder Wang Xingxing's keynote address. In the aftermath of the company's successful market debut in Shanghai, which made it China's first listed humanoid robot company, he said that humanoid robots can reach mass market popularity within a decade after manufacturers can resolve problems that hold back their usefulness.

23:50He also said that the industry can take off within two years once at least 80 % of the task can be completed by robots. This is Alan Wan in Beijing.

24:05Ed Ludlow:The lead that U.S. frontier labs have had in the global AI race is narrowing. Rival models from Chinese developers are catching up with the performance of those from OpenAI and Anthropic. On top of that, Chinese models' lower prices are attracting global adoption. Bloomberg AI reporter Rachel Metz is with us to discuss the big take. What's interesting about the big take is like we've tried to put data against it, quantify it, erase. It's animated. Explain the methodology there. Yeah, so the idea here was to look at a lot of different providers, both from model providers, both from China and from the U.S., and do some interesting things with the models.

24:45One of the things that was done for this story was simply taking some powerful models and saying, hey, make a website for a fictional coffee shop called Brewberg and seeing how the different models did it and also how much was it costing. So like a model from Anthropic, I think it was Fable 5 was by far the most expensive. And then you have an assortment of, but it also did a really, really good job. And then you have an assortment of others that were much cheaper that may not do quite as good a job, but they could certainly do a good enough job in a lot of cases.

25:20Ed Ludlow:Right. The case study and one of the biggest stories on the show in recent weeks was Kimi K3, right? and how Kimi K3 stacked up against frontier models from the US labs. Just tell me a bit more about that. Yeah, I mean, what we're starting to see, both from that model in particular and some other models as well that are coming out of Chinese labs, is that they're doing really well on specific kinds of benchmarks. These are measures generally of how well an AI model can or should do on a kind of task. And I think it's important, though, to note that this isn't real world use. I mean, that could be quite different in many ways, but these can give a sense for in math or in science or in coding how well a model could perform.

26:09Ed Ludlow:Token prices seems to be a big factor. How much is that coming up in your world at the moment? I mean, I think that what we're seeing more and more, and my colleague Shereen Ghafari has done some great reporting on this quite recently, is there's a bit of a shift from talking about token pricing to talking about cost per task. How much does it cost to do this thing or that thing? And I think there's also been a lot more cost consciousness on the part of companies, perhaps on individual users. If you're trying to do something with a powerful model, it gets quite expensive quite quickly. That's because it costs a lot of money to do computationally heavy tasks.

26:50So a lot of these Chinese companies, they recognize that their pricing can be quite a bit lower. And I think we'll see more and more companies saying, hey, this could be a really good bank for my buck.

27:03Ed Ludlow:Bloomberg's Rachel Metz with the big take. Thank you so much. Now coming up, sources told Bloomberg SpaceX approached AI coding startup Cognition about a potential acquisition. The companies responded saying they're not currently talking about a deal. We have the details next. This is Bloomberg Tech. this is what's happening in markets basically stocks down and a rally in bonds has fizzled out and remember this is all about the treasury planning to curb borrowing costs and the market basically interpreting it as a short-term fix the nasdaq 100 for what it's worth is down for a fifth straight session so kind of over a longer period in aggregate there's a little bit of pressure under technology shares i bring up again that next week is nvidia earnings and yes it's one company and one earnings report, but NVIDIA is NVIDIA, and that's a part of the treading of water we see in equity markets.

27:50Ed Ludlow:I want to get to a story that broke in the last 24 hours. On Wednesday, Bloomberg reported citing sources that SpaceX had approached AI coding startup Cognition in recent months about a potential acquisition. Our reporting noted that Cognition didn't engage with the takeover approach. After the story came out, Cognition CEO Scott Wu responded on X, saying, this is not true. Huge respect for the SpaceX team, but Cognition is not for sale and we haven't been talking. Elon Musk, the founder and CEO of SpaceX, then responded to Wu's post saying, we haven't talked with Cognition about anything except making sure Grok works well for their needs.

28:29Ed Ludlow:Let's bring in Bloomberg to Rebecca Torrance who broke this story with me and the team. So that's the reporting. Let's recap it. At some point in the last couple of months, SpaceX approached Cognition. What else do we need to know? Absolutely. So essentially what we know from our sources is that SpaceX approached Cognition interested in making an acquisition. Cognition ultimately rebuffed them. There are no active deal talks currently happening. That said, according to our sources, the companies are continuing to discuss working together in various ways, including potentially for Cognition to use SpaceX's computing capacity.

29:03But no active deal talks at the moment. And again, this happened in recent months.

29:09Ed Ludlow:Yeah, what was made clear to me is that Cognition considers itself not for sale at any price, but that, you know, the technology at the technology layer, there's a lot of interest in Devin and how Grok could work with Devin. The context is that SpaceX has just closed the cursor acquisition. And we've just reported that Cognition's out in the markets raising money at a much higher valuation. Give some numbers behind that. Right. So Cognition is in early talks for a new fundraising round at at least a$40 billion valuation. That's up from$26 billion in May. That's what we reported last week. This would be, if SpaceX had attempted to acquire Cognition, this would be the second large takeover attempt for the company in recent months.

29:55The cursor deal was just completed last week at$60 billion. So pretty hefty price tags on these. Cognition has been quite adamant that they want to stay independent. And at that kind of valuation, It's really no surprise. This is an incredibly hot space, as you mentioned.

30:11Ed Ludlow:And again, after we published the report, Scott Wu, the CEO of Cognition, and Elon Musk, the founder and CEO of SpaceX, both posted on X, denying parts of the story, not addressing the main point, which is two months ago, SpaceX approached Cognition with the idea of acquiring them. Bloomberg's Rebecca Torrance with the reporting. Go and read the full report on Bloomberg platforms. I want to stay with private markets. It's defense tech startup Castellian has raised a billion dollars at a 13 billion dollar valuation. The round combines 800 million in equity and 250 million committed to a revolving credit facility.

30:46Ed Ludlow:The firm, which builds weapons and defensive systems, plans to use the funding to boost production of its blackbeard hypersonic missile system. Castellian's co-founder and CFO Andrew Kreitz joins us here in San Francisco. A big round. And let's start there. The participants in it and how it came together. Yeah, absolutely. Thank you for having me. Really excited to be here and really excited to be announcing our Series C. So we ran a competitive process. There was a tremendous amount of interest in this round, quite frankly, from a whole range of different investors. So one of the things that was exciting is it allowed us to be really deliberate, really thoughtful about who we selected as our partners.

31:23Ed Ludlow:What are the advantages of going with a traditional VC who's just been micro-focused on the national interests of this country and its allies versus one of the sort of older legacy Wall Street institutions that are also so interested in you? Well, that's sort of the excitement here is being able to combine the best of both worlds. So Andreessen Horowitz, which co-loved the round, has been a Castellian backer since our seed stage. They bring a huge amount of expertise in sort of early hardware development. They have a fantastic network, and they're very, very risk tolerant, especially early on.

31:54You would also fall within the American dynamism bucket. We absolutely would. Would you agree with that? Absolutely. Absolutely. And they've done incredible work across defense, of course, but really across reindustrialization broadly. What's exciting about bringing in J.P. Morgan, Carlyle, those are two firms which have track records that speak for themselves. Carlyle especially, of course, sort of the gold standard in aerospace and defense private equity. I think what's exciting about bringing them is two things. One, depth of experience and network in different places across capital markets, across government.

32:26And then number two, I think that they provide a really strong signal of our business's maturity and our path to cash flow generation for public markets.

32:33Ed Ludlow:The maturity from a financial standpoint is so interesting. Can we talk a bit about from a technology standpoint as well? So when I read about Kasellian, I kind of see three layers. There's Blackbeard in isolation, a longer range strike system, and then the defensive systems I mentioned earlier. Where are those in their life cycle? as products, as things out in the real world for the defense of this country. And again, I think allies we can talk about too. Yes, absolutely. So that's exactly right. Our first product is Blackbeard. Blackbeard is very, very well into development. And what this round really helps us do is go maximum speed on scaling up production.

33:07So we're already going as fast as possible to close out development, get Blackbeard into production next year. But what's really going to matter is increasing the rate at which we are able to produce them after that first article is delivered. The second product you talked about, much larger, much longer-ranged system. We think of it as being complementary to Blackbeard. You'll go on different platforms. It's kind of a different class of weapon. But this is really, to be honest, kind of the North Star product that we formed Castellian to go build. We've been working on this product since 2023.

33:35This lets us supercharge development of it. And then lastly, Interception. It's been top of mind. It's certainly been in the news because of events in the Middle East. There's a dire need there to get to scale, to get to low cost. And that's sort of the third area that we're driving with the Series C fundraise.

33:53Ed Ludlow:Andrew, I'm listening to the very specific words you're using. Build comes up quite a lot. Yes. You know, you make hypersonic weapon systems, right? But there seems to be a real emphasis on your competence in the manufacturing, the production process. Is that a fair interpretation? Absolutely. What we always say, what we've said since the company started is nothing matters until we are manufacturing at scale. This is, sure, part of it is a technology problem, but the real engineering challenge here, the real difficulty is not how do you make one unit. It's how do you make thousands and thousands and thousands at low cost.

34:24That is what we are relentlessly focused on. And since the beginning, everything we have done is backwards planned towards getting black beer into production as soon as possible at scale.

34:33Ed Ludlow:The outset of the war in Iran, we talked a lot about munition supply and supply chain, actually. So that was interesting. Your backstory also interesting. You know, you were in the finance org at SpaceX. Your job was to cost things out, essentially. You know, your customer in the Castellian context is the U.S. government, essentially, to this point. Reflect on that experience at SpaceX and how you've applied it here. Well, honestly, it's part of the core thesis and sort of the core insight that leads to us starting Castellian. We were at SpaceX. we saw different ways of doing engineering, different ways of verticalizing your supply chain and focusing on getting costs out.

35:13And it kind of opens up your mind to the idea that, oh, this stuff doesn't have to be as inherently expensive as it currently is. Yes. The pressure on that comes from the U.S. government or you put it on your service? I would say that in a lot of ways, it's the incentives that have been set that industry goes and meets. If your incentives are oriented around really exquisite capability and nobody really thinks you're going to need these in large quantities, that's what you build to. If the requirement is mass scale, low cost, still high capability, but focus on that scale and that quantity aspect, that's what we can drive towards.

35:51Ed Ludlow:That's what we're focused on. You have$500 million in existing contracts with the U.S. government and the different departments associated with that. You've just raised a billion dollars. So clearly your fundraising is ahead of current deals. But the opportunity is also outside the U.S. You can do business with allies. Where's the TAM? I would say the TAM here is enormous. If you think about U.S. allies and partners in Europe with the Ukraine war, I think that's an obvious situation a lot of people have their eyes on. But you look across the Western Pacific as well. We unfortunately live in a very dangerous world.

36:27And I think the reality is in the post-Cold War era, a lot of our allies and partners also sort of reduce their focus on defense production and reduce their focus on scale. What we are going to deliver for the U.S. and for allies and partners is that scale. And there is a lot of demand for it worldwide. Andrew, just very quickly to end,

36:47Ed Ludlow:Castellian, where does the name come from? What's the backstory? We had a different name early on. And we decided that we would make up a word that no one else could possibly take umbrage with. But it's meant to evoke a castellan, which is the Latin term for a keeper of a castle. Andrew Kreitz, CFO of Castellan, thank you so much for being here in person in NSF and taking us through the round. Coming up, we're going to speak with Mayfields and Naveen Chadha on the next phase of AI investing away from hardware and defense and some other domains. This is Bloomberg Tech.

37:27The people who seem to get more done than everyone else, they're not working longer hours or running on more caffeine. They've just stopped wasting time on the stuff that doesn't move work forward. Switching apps, re-explaining context, hunting for files. Those aren't small inefficiencies. They're hours wasted every week. Superhuman Go gives you those hours back. From the makers of Grammarly, Go is an AI assistant that sits inside every tab and tool you already use, always available and already aware of what you're working on. Ask it to draft something, summarize a long thread, pull up a file, or prep you for a meeting.

38:08Go handles it without you ever leaving the page you're on. This is what it looks like when AI actually fits into your work, instead of adding to it. It's like having a teammate whose only job is to help you be better at yours. Go keeps up so you can move forward. With Go working with you, you can show off what you do best. Superhuman Go. Find out more at superhuman.com. That's superhuman.com. This is the Bloomberg Tech Minute brought to you by ChatGPT. Now with ChatGPT Work. I'm Carol Masser. Walmart has become a major player in the burgeoning business of quickly fueling electric vehicles, speeding past Costco as well as more established charging companies.

38:52Bloomberg's Kyle Stock writes, as of June, Walmart has opened about 46 high-speed public charging stations with 380 cords. There are now EV chargers at about 326 of its U.S. stores, including adding this year four high-speed charging stations in Bentonville, Arkansas, even though the state has been a laggard in transitioning to electric vehicles. Walmart has hosted chargers from other companies since 2018. The big difference now? Walmart is building its own network, and quickly. As of March of last year, every EV station at a Walmart is under the company brand. Now, Walmart is still a blip on the U.S.

39:32charging map, and yet it was second only to Tesla among charging networks expanding in the second quarter. That's the Bloomberg Tech Minute brought to you by ChachiPT. Put ChachiPT to work on your most ambitious ideas and projects. Get started at ChachiPT.com today by selecting Work Mode, available on Plus and Pro plans. Everyone's talking about how AI is transforming work, especially in sales. While the landscape shifts, one thing remains the same, the thrill of closing a deal. Whether it's a gong or a confetti machine, every team has its celebration rituals. Adio is designed for that moment.

40:09It's the agentic CRM that turns customer signals into actionable insights, helping you close deals faster with revenue agents and automations working around the clock. You'll have everything you need to scale your go-to market efforts. Elevate your wins with Adio. Start your free trial at adio.com slash iHeart.

40:33Ed Ludlow:VC firm Mayfield says AI is making it easier than ever to launch a startup while raising the bar for building a winner. The firm's invested more than$3 billion in AI companies, much of it at the earliest stages. Joining us now is Mayfield managing partner Naveen Chada. And I mean, it's great to have you here with us. Absolutely. It's a delight to be here today. So in a funny way, I want to get straight to returns. If you've put$3 billion into early AI startups. How's that gone for you in the startup curve? It's going extremely well. Having been in business since 1969, our core business has been investing in not only early stage companies, but inception stage companies still, where entrepreneurs have paper and pencil ideas.

41:19So 70 % of the investments we do are at that stage. In our history, we have had 120 IPOs and 225 acquisition. And the results have been phenomenal. And with AI, they're only getting back better and better by a factor of 2 to 3x. Exciting times. It's a 100x opportunity. And we believe early stage is the place to play.

41:44Ed Ludlow:Early stage is broad. Like inception, from inception means something very specific, right? You know, there's so much focus on San Francisco and Silicon Valley right now. But the history is this is the college dropout. or someone that never even went to college. Where are you finding the inception startup? Yeah, so the inception startups depends upon where you play in the stack. So if you're building a semiconductor company, those are not college dropouts. Those are people who have been around for 10, 15, 20 years. And we know 70 % of the value is accreting into the semiconductor layer. You move up to models.

42:21That's where you need researchers. You need professors. Again, they have the main expertise, but they're not dropouts. They are like PhDs. They are like research scholars. They are postdocs. But if you move up the stack to agents and agentic applications, that's where you see the young entrepreneurs. And I think as a VC firm investing at the inception stage, you need to have a prepared mind and invest in areas before they're hot and find entrepreneurs before they've started a company. So it's a lot of hard work.

42:57Ed Ludlow:You know, on the show of late and outside of the show, I've spent a lot of time down at Sequoia, talked to the IVP about this. It's almost as if the focus is resisting the urge to invest sometimes. You know, saying, okay, I'm committed to finding X number of companies per year of that high bar that you outlined at the beginning. Absolutely, right? Venture returns are driven by power law. It's not about the quantity of investments you do. It's about the quality. Our job is at Mayfield, we do selective investments, 8 to 10 per year. And our job is how do we find one to two unicorns per year and one decacon per year.

43:39So it's all about quality. It's not about quantity.

43:42Ed Ludlow:Let's talk about the exits. Absolutely. Particularly with the focus of the$3 billion that went into AI. But what have you learned from the IPOs and the acquisitions, say, over the last five years? Absolutely. So the scale required to go IPO has increased a lot. We have seen trillion-dollar IPOs. More are coming. So the market today is going to be dominated by those. But then the market will move towards mid-cap IPOs, which will raise a billion dollars, five to ten billion in market cap next year. but there will be a lot of strategic acquisitions which will happen 20 30 40 50 billion but also between 1 to 10 billion so there's going to be a lot of liquidity that is going to get created over the next few years you know both things seem to be true i've already said that exact sentence this week companies are staying private for longer because they can absolutely if you're sitting on the board of a company that could be a mid-cap ipo would you encourage them to do that Absolutely.

44:46Ed Ludlow:You have a fiduciary duty to your LPs to make a return at some point. I think what we have learned over the history is it's much harder to go from 100 million to 1 billion in enterprise value. But once you get scale, once you get repeatability, once you get some kind of network effect, the 1 to 10 billion and 10 billion to 100 billion, if you have all the right metrics, it's possible. But again, in the power law, this is going to be less than 1 % of the companies. If you find them, it's the job of VCs and entrepreneurs to keep going long and create iconic companies. In the environment right now, how much of the value is accruing or is associated with the founder themselves?

45:31Our belief is at this stage, we invest. It's all about the people because people build companies, people build products. So we tend to make a bet on the jockey and not the racetrack. And in our history, with so many IPOs and so many acquisitions, half the companies didn't execute on what their final idea was. They pivot. So our belief is, bet on A plus people. If they enter a B or a C market, they'll pivot to find the right market. So it's a patience game. It's running a marathon. It's not a sprint. So we index heavily at the stage we invest at on the entrepreneur, the founder. That's why we are known as a founder first firm.

46:15Ed Ludlow:Naveen, we just have 30 seconds, but what pressure is on you at the moment to raise more funds, to have more capital to deploy at higher scale? No pressure. Basically, we're going to stay focused on keeping our fund sizes small. 70 % of the dollars will go at inception stage. We're going to buck the tent away of raising growth funds. Stay nimble. Stay disciplined. Keep doing what we are doing and create alpha. So we are wealth creators. Take small boxes of money for our LPs. Make them bigger. We're not in the AUM business. That was my question. Is a growth fund coming? I appreciate your time, Naveen Chhada of Mayfield.

46:54Ed Ludlow:It's great to have you on Bloomberg Tech. Thank you very much. Let's get out to New York where Bloomberg's Yahaira Anand is standing by. Hi, Yahaira. Hi, Ed. It's time now for Talking Tech. First up, CoreWeave and Hudson River Trading signed a multi-year deal worth billions of dollars to build new trading research and models using CoreWeave's AI computing services. Hudson River will also become one of the first CoreWeave customers to access NVIDIA's newest Vera Rubin AI chips. Plus, after historically relying on products from companies like NVIDIA and AMD, Google's Waymo has built a specialized chip for its robo-taxis.

47:32The application-specific integrated circuits, or ASIC chip for short, will improve the robo-taxis reflexes and navigational skills in complex urban environments. And credit sites introduced a new way to measure the risk of overbuilding data centers. The metric tracks future contracted revenue versus contracted lease payments. Microsoft scores among the worst due to its high amount of future lease payments. Ed?

48:01Ed Ludlow:Okay, thank you, Jairo. Coming up, Massachusetts is working on what could be the nation's strictest AI safeguards yet, and it's dividing the sector. We're going to have the details next. This is Bloomberg Tech.

48:15Ed Ludlow:Massachusetts lawmakers are seeking to establish the nation's most stringent state-level safeguards for AI, A proposal that has divided the very industry the governor's office is seeking to attract, with Anthropik supporting the proposal, while OpenAI is warning it will slow the release of cybersecurity models. Bloomberg's Greg Ryan joins us from Boston. Let's start with the proposal that's on the table. Outline it for us. Sure. So this proposal would require AI labs to publish their safety plans, to file regular risk reports, and would establish whistleblower protections for AI lab employees.

48:50All those elements are covered in laws that have been passed in the past year in states like New York and California. What really sets the Massachusetts proposal apart is this requirement that frontier models undergo independent reviews at least every 180 days to gauge the risk that will cause a catastrophe like providing advice for making a biological or nuclear weapon or a massive cyber attack.

49:15Ed Ludlow:So we opened the conversation with saying that the governor's office is also trying to attract the industry, right? What's that part of their strategy? Yeah, so, you know, Massachusetts universities like Harvard and MIT, they're throwing out all sorts of students who are creating startups, but they're going to Silicon Valley. They're creating successful AI startups, but a lot of them are headed to Silicon Valley to do so. So the governor has made a real push this year to retain more of that talent, to attract more AI companies to Massachusetts. And to that end, she's really put on a push for Anthropic and OpenAI.

49:52The state's partnered with OpenAI. They've made a chat GPT assistant available to all state employees. And I filed a public records request that showed that state officials and Anthropic executives have been meeting this year, been in conversations about potential partnerships. And so while on the one hand, the legislature is considering imposing these strict safeguards, the governor is seeking to bring these companies into Massachusetts.

50:16Ed Ludlow:Right. Greg, very quickly, where is this on the legislative path? How far is it from being real? So the state Senate passed this proposal last month. Now it's up to the House to decide whether they agree to these safeguards. And then, of course, the governor will have to sign it. So it's close to the finish line, but not quite there yet. Bloomberg's Greg Ryan with a really top story in the world of AI and regulation. Thank you very much indeed. That does it for this edition of Bloomberg Tech. Recap of what markets look like right now. And again, it's things outside of technology that are kind of the factors here.

50:51Ed Ludlow:We see US 30-year yield back at 5.24%, 25%. Remember, the story is treasury intervention in the bond market. And we've had some clarity around that. And that's that 100 down for a fifth straight session. Recap the show on the podcast. You know where to find it on the Bloomberg platforms and online on Apple, Spotify, and iHeart. Four days down, one to go. This is Bloomberg Tech.

51:18Healthcare doesn't always work great. If you've ever waited on a refill or couldn't schedule an appointment, you get it. That's the kind of stuff Optum is changing. They're using data and technology to integrate patient care, pharmacy, and everything else. So healthcare is connected, not complicated. What's that look like? Cheaper prescriptions that are easier to get and care that looks at the whole person. How you need it. Optum is helping make healthcare work as one for everyone. Learn more at business.optum.com. Everyone's talking about how AI is transforming work, especially in sales. While the landscape shifts, one thing remains the same.

51:54The thrill of closing a deal. Whether it's a gong or a confetti machine, every team has its celebration rituals. Adio is designed for that moment. It's the agentic CRM that turns customer signals into actionable insights, helping you close deals faster. With revenue agents and automations working around the clock, you'll have everything you need to scale your go-to market efforts. Elevate your wins with Adio. Start your free trial at adio.com slash iHeart. This is Jacob Goldstein from What's Your Problem? Running a business is hard enough. Don't make it harder with a dozen apps that don't talk to each other.

52:27One for sales, another for inventory, a separate one for accounting. That's software overload. Odoo is the all-in-one platform that replaces them all. CRM, accounting, inventory, e-commerce, HR. Fully integrated, easy to use, and built to grow with your business. Thousands have already made the switch. Why not you? Try Odoo for free at odoo.com. That's odoo.com.