Meta to Become the Biggest Nuclear Buyer Among Hyperscalers

0:00Hello, I'm Stephen Carroll. I'm in Brussels where many of Europe's biggest decisions get made. And I'm Caroline Hepker in London with the hosts of the Bloomberg Daybreak Europe podcast. We're up early every weekday keeping an eye on what's happening across Europe and around the world. We do it early so the news is fresh, not recycled and so you know what actually matters as the day gets going. From Brussels, I'm following the politics, policy and the people shaping the European Union right now. And from London, I'm looking at what all that means for markets, money and the wider economy. We've got reporters across Europe and around the globe feeding in as stories break.

0:37So whether it's geopolitics, energy, tech or markets, you're hearing it while it happens. It's smart, calm and to the point. And it fits into your morning. You can find new episodes of the Bloomberg Daybreak Europe podcast by 7am in Dublin or 8am in Brussels, Berlin and Paris. On Apple, Spotify, YouTube or wherever you get your podcasts.

1:02Bloomberg Audio Studios. Podcasts. Radio. News. This is Bloomberg Tech. Coming up, Meta agrees to a series of electricity deals to power data centers, making it the biggest buyer of nuclear power among the hyperscalers. Plus, Minimax, one of China's largest generative AI startups, goes public in Hong Kong. What that means for China's AI ecosystem. And Snowflake plans to buy Observe, an AI-powered observability platform. We're going to discuss the move with Snowflake's CEO. Let's get right to it and what's happening in the markets. Later in the program, we're going to go big on Intel. Lit Bhutan, Intel CEO, has spent the last 24 hours in Washington, D.C.

1:44in the White House with the president and with Howard Lutnick, the Commerce Secretary, glowing remarks from President Trump about Lit Bhutan's leadership of Intel, but also the progress that the U.S. government's made in taking a stake in the chip maker. More to come. Then there's the big one, Meta investing in a multi-gigawatt deal with some big names in nuclear energy. Oclo up 14 % on that deal, of course, backed by Sam Altman. Vistra Corp up 13%. Meta up 0.8 % to 1%. We said at the top of the program, this puts them among the biggest energy buyers among the hyperscalers. And we remind ourselves, Meta is not technically a hyperscaler.

2:22Let's get to Bloomberg's Riley Griffin. Details here are really important. size of the deal, how many gigawatts, but also the structure, please. Yeah, so three different agreements, and they actually are different in form and function. So this is about supporting up to 6.6 gigawatts in nuclear energy, but it is both about ensuring that existing nuclear power plants continue to thrive and investing in future nuclear power. Meta's head of global energy told me last night that they heard that there's real concern about the amount of energy that's out there. As you know, Ed, there's an insatiable demand for energy, and this cements their low-carbon play.

3:04I find this fascinating because while Meta is not a hyperscaler, a hyperscaler being a cloud computing company that basically leases capacity, Meta is doing this on its own behalf, right? It has been aggressive in securing the supply of energy it needs for its own data centers, which principally are used for training and inference of its own activity in AI. What else has it done and where does this bring them to date? Well, I want to note that you made a really important point there. Prometheus and Hyperion, these are two of their biggest AI plays. Prometheus actually is in Ohio and these agreements are in Ohio and Pennsylvania.

3:43So some of this energy presumably is going to go to the Prometheus data center cluster, which is one of their biggest plays. But we're also seeing Meta turn with Hyperion to natural gas. At least three natural gas plants are being fired up just for that single facility. So it's not like nuclear is the only strategy that Meta is employing here. We're going to get to the sort of available energy sources around the world in a little bit. There is an interesting point. You co-reported this with Will Wade, who's just been really on top of the nuclear side of the story, there does seem to be some anxiety from the technology companies that the existing nuclear infrastructure that does exist in America, limited as it may be, is also at risk of being shut down.

4:26Is this kind of future-proofing a little bit? Absolutely. That is what I'm hearing from Meta's head of global energy. They started back a December two years ago, and they were looking into what nuclear needed to keep going. And they heard we need investment now into plants that could potentially shutter. So this is a future proofing, as you say. Bloomberg's Riley Griffin leading our coverage of Meta with a must read. Thank you very much. Access to power has been top of mind in AI. We spoke about that with NVIDIA CEO Jensen Wong earlier this week. In order for a new industry to emerge, you need energy.

5:00And so I think it's safe to say that we wish we had more energy in the United States. Europe wish it had more energy. I think the world all wish we had more energy. And so we have to invest in all sorts of different forms of energy. Let's bring in Paul Meeks for more. He's managing director and head of technology research at Freedom Capital Markets. And when I was reading your research last night, Paul, a little bit relieved coming up this morning when I saw the meta headline, because you're very focused on the real terms footprint of data centers. That's what we're talking about here. You heard Riley's reporting on the specifics of what meta has done to secure future capacity.

5:39How much of a bottleneck is that to your mind right now for this sector? It is absolutely critical of all the bottlenecks. It is the most important. I like what Meta is doing here, but folks need to realize that if you ramp up nuclear capacity, we're not going to really see power generation until 2030 earliest, maybe even not until 2032. So what do we do for the next four to six years? So the immediate draw is probably not gas. But when I look at all of these companies and we cover the AI hyperscalers and the NeoClouds and everybody adjacent to them, we are looking for their availability for power, their capacity to find power, because it's not a demand issue.

6:35It is a supply issue focused on this particular metric. You know, Paul, I played you just some of the interview that we conducted with Jensen Wong earlier this week. And when I was sitting in front of him, you don't really get any sense of anxiety from him, right? Think about the mechanics of how this transaction works. NVIDIA brings GPUs. Actually, that's probably not fair. Increasingly, they bring more of the content of the server. But there is an acknowledgement that something needs to change principally in America. He did go on to talk about the differences between Europe, Asia, and the United States and where we're sourcing energy.

7:16Do you get a sense from your research that the companies you cover acknowledge the severity of that deficit right now? So when they are public facing, particularly somebody like Jensen Wong, who is essentially the spokesperson for the entire industry, they have to exude optimism. However, behind the scenes, particularly when I'm talking to companies about my financial models, which of course, the revenues and then the follow on cash flow and earnings come from? Do you have the power? Yes, they are anxious. And I think they're actually very anxious. There will be people watching Bloomberg Tech that work at Meta or in the AI industry.

8:00There will be Meta investors watching this program. And some of these people may never have heard of an not Clo. Have you had to change the remit and breadth of your coverage, if you say I cover AI data center, to include future sources of electricity supply? Yeah, it's really interesting, Ed, that years ago, nobody in my line of work covering the tech sector like I have for decades would be interested in any utility company. It's the antithesis of tech. They are boring, regulated, slow growth businesses. However, now they've come to the fore. And of course, these are not just your regular utilities.

8:40You know, they're your new era utilities. But yes, if you are a good analyst, you have to cover the entire supply chain, the entire ecosystem. And because demand is so robust, we're focusing on supply, absolutely critical. And it's not just the power. There are physical constraints here, like simply pouring the cement to build the next data center, which can be a bottleneck, even though it sounds so old school in a new school business. But yes, these are some factors that are very, very interesting that people didn't even conceive of a couple of years ago. Paul, is there an adequate federal level framework regulation to support the energy requirements of the AI industry in this country?

9:32You know, there is not. And here is my problem. We are still suffering from a lack of solid regulation for the Internet, because even today in the U.S., the Internet is operating under 1996 legislation when Bill Clinton was president. And so I'm hoping that we're not too restrictive, right? We need to embrace our entrepreneurial companies and give them room to roll. But we need to have some guardrails on everything AI, because what happened last time is we end up abdicating the legislation to the EU because we didn't have the will to do it ourselves. So we missed that opportunity in the internet.

10:14Here is the next wave. We need to embrace it and at least come up with some guardrails that are going to guide all this because there's some important decisions on the federal level to be made. In your research, Paul, it was very interesting to see you approach data center from the perspective of, yes, energy supply, but also real estate, the REITs, right? And what I really want to ask you is, away from the GPUs, where are the frailties in that industry? Is it literally the folks building the tin can around the outside of the data center, the labor, the concrete? Yeah, we've seen some hit in short term revenues because a couple of the number or a few of the players have subcontractors.

11:02Now, this is not them. This is their subcontractors that are delayed what we call building a powered shell. So it's more than a closet hut that you might find the army in. These powered shells are a little bit more sophisticated, but one company I cover had a$150 million revenue deferral from 25 to 26 because at one of their 41 data centers, someone couldn't pour concrete because it was raining. And so, yes, all these things come into play. And it's interesting that they may not necessarily be technical bottlenecks, but old school physical, including construction bottlenecks in addition to the power.

11:45I just got back from Vegas with all the attention on artificial intelligence in the digital world. We have an industry that's getting hit by a rainstorm. Paul Meeks. Isn't that unbelievable? Unbelievable. It's great to have you on Bloomberg Tech. A lot more to come. And coming up, an exclusive interview with the co-founder and COO of Minimax after the company just went public in Hong Kong. And that's the debut trade. And this is Bloomberg Tech.

12:23Minimax, one of China's largest generative AI startups, backed by Alibaba and Abu Dhabi Sovereign Wealth Fund is now public. Shares surged in Hong Kong after an IPO that raised$619 million, closing up 109 % in its Friday debut trading. Bloomberg's Stephen Engel spoke exclusively with co-founder and COO Ye Ye Yun about growth and expansion plans as Minimax faces hot AI competition both locally and globally. We really focus on capital efficiency, cost efficiency. So we had only spent around 500 million USD in total, probably only one or two percent of the big gens spending. So we did all the optimization, made innovations on it.

13:12How do you compete though, if you're going to go global, how do you compete with an open AI or others that are throwing lots of capital at their business? is it viable and what is the technology gap looking like in the next few years? Actually, I don't think it's a competition. It's more about all the top talents across all the world. They bring the technology breakthrough to the society, to the end users. So you will see each model has pros and cons. So we focus more resources on building the models, building the best product experience to get more and more users from all the world to use it.

13:50We're hearing as well the Chinese government might approve NVIDIA's H200 for China in this quarter, perhaps. Is that something you need to look at, importing more advanced accelerators for your training and the like? I think in every industry with a really high growth, when the new technology comes here, so you will see every startup, all the companies, they face resources constraints. Probably cheap constraints, computing constraints is only one of the challenges. So in the past, you will see our model performers and our product performers. So our resources are stable, we can get access to all of them.

14:28So we focus more on our R &D and foundation models. So we just use whatever chips that make sense for us at that specific stage. Is your goal to stay indigenously with your chips, locally made chips, or would you buy NVIDIA chips as well? It doesn't matter what kind of chips it is. It's more about which chips can give us best ROI, which can help us to achieve our mission to make the best technology accessible to all of the users across the world. So when does the emphasis turn from efficiency to profitability and also getting that steady revenue growth that will then justify a higher stock price valuation?

15:11So we put lots of R &D resources and innovations on the efficiency part. You will see our API, we have an enterprise business, you will see the gross profit margin of our API is more than 55%, which is probably already one of the highest globally. So we focus more on making the best technology and the best user experience. People will love to pay for the best performers models for their use cases. You've also come out of a price war, essentially. A race to the bottom, some have called, because you're trying to keep costs down and you want to survive as well. There was a battle of survival and you have been one of the survivors.

15:54But at the same time, the startups are not the only ones. The big players, Alibaba, ByteDance, Tencent and others. How do you compete with them at a low cost basis? Actually, it's not, I don't think that's a press word. It's more about the performance competition. So you need to make your model best. And the people, I mean, end user or enterprise customers, they don't choose your model because it's cheap or expensive. It's more about its performance. So that's the key. As you said, I think it's not about competition between all these big gens. But it's more about some of them, they are like our cloud providers.

16:34and some of their products, you see other models. So it's more about the collaboration. All this company brings foundation models to all the world. Minimax co-founder and COO Ye Ye Yun speaking exclusively with Bloomberg's Stephen Engel. Public listings in Hong Kong are hot right now. Chinese industrial robot maker InnoVance is now also considering a second listing in Hong Kong, according to sources. Let's get out to Bloomberg executive tech editor Peter Elstrom to discuss this, but also like this growing trend of more of China's sort of post-chat GBTAI firms going public. The team that you lead out in Asia are busy right now.

17:13In Inovance's case, what are the details in our reporting that we need to know about? Yeah, there are a lot of these IPOs that we're seeing, and we're seeing a much more forward move into the public markets for a bunch of these companies. Inovance has already gone public in the mainland in Shenzhen. Now it's looking at an IPO in Hong Kong, too. that's partly a reflection of how hot demand is from investors for these kinds of companies. When you're talking about Minimax in particular, they went public. As you mentioned, their shares more than doubled in the debut. The founder of the company is very fascinating.

17:46These are, some people call them the AI tigers, some people the AI dragons. But these are startups that really have very, very different strategies for how they're approaching the market compared with some of the US competitors, like in OpenAI. They tend to be much lower cost models. They tend to be much more aggressive about implementing the technology with their customers, with corporate customers, and giving them very low cost, very efficient models that they can use and they can actually deploy more quickly than some of the U.S. counterparts. Peter, in the world of technology or in financial markets, like a term like an AI tiger or AI dragon, you know, it's kind of commonplace.

18:20We bandy it around. But for those watching Bloomberg Tech that are saying, what are you talking about? How do we define the AI tiger? What are we talking about here? Well, we're talking about a whole series of companies coming out of China. In Stephen's interview with the Minimax COO, they talked about the competition there. We've, of course, heard a lot about DeepSeek and how good their model is, how technically good it is, but also how low cost it is. Behind DeepSeek, you have a whole bunch of other companies like Minimax, like Jipu, which just went public on Thursday. So you had the IPO of Jipu on Thursday, Minimax on Friday.

18:53Investors showed a lot of demand for those shares. And what these companies have been able to show is that with a very low-cost model, they're out there in the market competing. It's not clear that they're going to be able to compete directly with OpenAI or Anthropic, but they have very innovative models. I mean, first of all, the U.S. models are not going to be let into the China market. But these companies are also going out very aggressively and competing in some other parts of the world. We've written about how much traction DeepSeek is getting in Africa in particular because their companies are sensitive to cost.

19:22They're sensitive to how much power, how much computing resources they need to be able to deploy these models. So they are having some successes. We're going to see a lot of competitive clashes in the months ahead. Bloomberg's executive editor for Tech, Peter Elstrom, great to catch up with you. Thank you so much. Now, coming up on the program, Elon Musk's Grok image generation tool faces backlash after reports that it's generated thousands of undressed women and children. Got more on that data next. This is Bloomberg Tech.

20:00Elon Musk's AI startup XAI is burning cash quickly. The company spent almost$8 billion in the first nine months of the year. That's according to internal docs that Bloomberg's viewed, though revenue has nearly doubled quarter over quarter to$107 million. XAI's losses have been mounting on costs to build data centers, recruit talent, develop software, and to build AI that is self-sufficient. And according to Bloomberg's reporting, the plan is to use that AI to eventually power Tesla's humanoid robots in the Optimus program. Right. Sticking with XAI, the company has limited access to Grok's image generation tool for most users following widespread criticism that the feature was producing thousands of explicit images of women and children.

20:50Bloomberg reporter Cecilia D 'Anastosio broke the story, has been tracking the developments. Cecilia, we just summarized the issue, right? I think let's start with the methodology, how we went about looking at the data, how we got the data, and we'll go from there. Thank you for having me. Bloomberg worked with a researcher who scraped thousands of images produced by Grok and published to the platform X between the period January 5th and January 6th. The researcher analyzed those images to determine what percent of them were sexualized and nudifying and found that every hour during that 24-hour period, X was publishing about 6 ,700 images identified as such.

21:31And just for a comparison, websites that are dedicated to publishing deepfaked images, AI-generated images of women who are sexualized, nudified, those top five websites out there published together 80 images per hour. So Grok is one of the biggest deepfake producers on the Internet today. Where are these images showing up? You know, to the uninitiated, people may not know the sort of interconnect between XAI, the AI company, and X, the social media platform, for example. But they are now kind of one entity as they combined last year. Well, that connection is exactly why X has become one of the biggest deepfake websites on the Internet.

22:14that anybody can, well, not anybody now, now you have to pay to do this, but over the last couple of days, anybody could just at Grok on the X platform and say, put this woman in a bikini under an image that a woman had posted on the internet. And a lot of women, women primarily are the victims of this, took to complaining to Grok directly and arguing with Grok in their comments. Grok would apologize for posting these images, but then continue posting them and not take down many of the images that the women felt were violating. Cecilia, did X or XAI respond to our reporting, engage with us on it?

22:49X has not responded to our numerous requests for comment over the last couple of days. Elon Musk himself, in a reply to a post on X, said anybody using Grok to make illegal content will suffer the same consequences as if they upload illegal content. And Musk there was referring to allegations, we believe, of child sexual abuse material that's been posted to X using the GROC app. Okay. Bloomberg's Cecilia D 'Anastasio. Important reporting. Read the story on the Bloomberg terminal on Bloomberg.com. Thank you. Now, coming up on the program, we're going to speak with Snowflake CEO Sridhar Ramaswamy.

23:23As the company enters into an agreement to buy AI-powered platform, Observe will get his observations on that piece of M &A. that's coming up next. You're halfway through the program and this is Bloomberg Tech.

23:48Welcome back to Bloomberg Tech. I've been using the power of my Bloomberg terminal. It's been a long week with CES in Las Vegas. The power of my brain is dwindling, but it tells me that Apple is down for an eighth straight session, which matches the run of declines that Apple saw in May of 2025. Now, using more of the power of the Bloomberg terminal, if Apple was full for a ninth day, we think that would be the longest run of declines since the early 90s. And obviously, we're only nine days into 2026, but so far year to date, this is a stock that's down five and a half percent. You saw the letter from Tim Cook to shareholders talking up the company's progress of late and last year.

24:32But right now, that stock heading down and will continue to track it. We had some M &A in the last 24 hours, but it's actually in the AI and software space. Snowflake has agreed to buy Observe, an observability platform, an AI-powered observability platform. Observability is something that's come up a lot recently, particularly in the context of agentic AI, when I've been around the table, particularly on the engineering side. Why this piece of M &A? Why now? Let's talk to Snowflake's CEO, Sridhar Ramaswamy. Those are the reasonable questions, right, Sridhar? I do want to get to defining observability and why it's important, but it's an interesting deal.

25:12Why did you do it? Hey, Ed, great to see you. Snowflake's acquisition of Observe is a game changer for our customers. As you pointed out, observability is basically looking at applications, websites, AI agents, to make sure that they are functioning properly. It's a diverse and messy problem. It's a big data problem. And because Observe is built right on top of Snowflake, they are able to use our very efficient storage as well as compute platform to help customers find problems 10 times faster, often at 3 to 4x less cost. Take a company like Topgolf. It's a customer of Observe. I go to Topgolf all the time with my team.

25:57I'm not much of a golfer, so I'm playing an Angry Birds game or something like that. Topgolf used to struggle when these games went down because they collected everything in a central place but could not identify that a particular booth had problems. They're able to do that with Observe in a matter of a few seconds, send out a tech team to go fix that problem and have happy customers. That's the power of this acquisition. Increasingly, capabilities like observability are going to be a core part of a platform like Snowflake. And we have great customers, folks like Barclays, Capital One, Commonwealth Bank of America.