Google Cloud Debuts New AI Chips

22 Apr 2026 · 44 min · 20 chapters

Ask about this episode

Ask anything about it. ChatGPT or Claude reads this page and answers with the times it was said.

Connect VO and ask about every podcast you hear, including the moments you saved. Add to ChatGPT · Add to Claude

In short

This Bloomberg Tech episode is a fast “news dump” focused on AI infrastructure, cybersecurity, and EV/robotics business updates. Topic 1: Alphabet/Google Cloud unveils new TPUs: TPU 8T (training) and an inference-specific TPU AI, aiming to speed and reduce the cost of AI workloads.

Key claims

Google’s custom silicon and scale make it the “lowest cost token provider,” widening cloud growth and margins versus rivals; Mythos may have been trained largely on TPUs, validating TPU performance.

Notable examples

partnerships with Oracle (natural-language access to Oracle data), NVIDIA (agentic/physical AI for AI factories), Salesforce, CrowdStrike, Broadcom, and Thinking Machines (Miramirati). Topic 2: Anthropic Mythos was accessed by unauthorized users; Bloomberg reports a contractor credential plus “URL sleuthing” enabled access, raising concerns about basic security. Topic 3: Rivian CEO RJ Scaring discusses R2 production ramp after tornado damage, pricing ($45k base to $57k top), and LiDAR + in-house silicon (800 TOPS per chip) coming end of year.

Guests

Mandeep Singh (Bloomberg Intelligence), Rachel Metz (Bloomberg AI reporter), RJ Scaring (Rivian CEO), Natasha Mascarenas (Bloomberg), Kelsey Griffiths (Bloomberg telecom reporter), Renan Halak (Vast Data CEO), Rahsaan Shias (KPMG AIQ lead), Cara Coulson (Bloomberg EV reporter).

Written by AI. May contain mistakes. Listen to the episode to check what was said.

Chapters

Tap a time to open that second in VO

Google Cloud's New TPU Announcement

2:29 to 4:36

Discussion on the unveiling of Google Cloud's latest Tensor Processing Units (TPUs).

“First, we check in on these markets, which manage to be shrugging off, yet further chaos that occurs in the Strait of Hormuz.”

Partnerships and Market Impact

4:36 to 6:04

Analysis of Google's expanding partnerships and its implications for cloud market share.

“Mandeep Singh for Bloomberg Intelligence is with us.”

TPU Performance and Competitiveness

6:04 to 7:42

Exploration of TPU performance compared to competitors and its advantage in AI workloads.

“Does it matter whether they're startups or whether they're legacy businesses?”

Rivian's R2 Production Launch

7:42 to 10:34

Interview with Rivian CEO about the R2's production start and company resilience.

“Mandeep, just briefly, how much do you think that this adds to the revenue run rate?”

R2 Vehicle Launch and Features

14:00 to 16:48

Explore the details of the R2 vehicle's features and pricing strategy.

“and we couple that with cameras, radar, and a LiDAR that sits at the top of the windshield.”

Autonomous Driving Capabilities

16:49 to 19:25

Learn about the future rollout of autonomous driving features in R2.

“even the advanced driver assistance capabilities of R2, they are predicated on that next-gen gear.”

Electric Vehicle Market Trends

19:26 to 20:26

Understand the acceleration of electric vehicle adoption in global markets.

“And it's no surprise that four of the top five battery makers in the world are located here.”

Unauthorized Access to Mythos AI Model

23:50 to 26:57

Discuss the implications of unauthorized access to the Mythos AI model.

“Copyright 2026, JPMorgan Chase and Company.”

SpaceX's AI Acquisition Strategy

26:58 to 28:00

Examine SpaceX’s potential acquisition of AI startup Cursor and its implications.

“state actors or something like that, that are using access, legit access seeming to technology and then getting access to mythos?”

Elon Musk's Potential Acquisition of Cursor

28:00 to 29:04

Discussion on Elon Musk's potential $60 billion deal to acquire AI startup Cursor.

“Right now, the US-listed shares of Alibaba up 1.3%.”
Show all 20 chapters

Impact of SpaceX's IPO on M&A Landscape

29:04 to 30:44

Analysis of how SpaceX's IPO plans could affect the M&A landscape and the valuation of Cursor.

“If not, as we exclusively reported yesterday, the$10 billion dollar fee is actually being looked at as a breakup fee.”

Telecom Industry Rumors and Potential Merger

30:44 to 31:53

Exploration of rumors surrounding a potential merger between Deutsche Telekom and T-Mobile.

“I mean, this is the same reason that we saw Windsurf, another competitor, a once a standalone competitor to Cursor last year seek compute partnerships and opportunities with bigger technology companies.”

Historical Context of Deutsche Telekom and T-Mobile

31:53 to 34:10

Insights into the historical relationship between Deutsche Telekom and T-Mobile and implications of a merger.

“I didn't have this on the bingo card, but there is a backstory, a shared history, particularly at the C-suite level.”

Vast Data's Growth and Valuation Insights

37:10 to 40:30

Discussion with Vast Data CEO on the company's growth, valuation, and future plans.

“more than tripling the company's valuation to$30 billion.”

Market Trends and Future of AI Investment

40:30 to 42:08

Exploration of market trends in AI investment and the role of companies like Vast Data.

“Definitely the last couple of months have been challenging, but if I look at five years out, ten years out, I still think the UAE can leapfrog over a lot of other countries.”

AI Investments and Business Strategies

42:08 to 42:44

Discover how companies are leveraging AI for data management and insights.

“Our system enables very, very large scale, very large capacity and ability to store vast amounts of information and give access to that information to the largest GPU clusters in the world.”

KPMG Report on AI Investment Trends

42:44 to 44:09

Learn about KPMG's findings on AI investments and their challenges.

“Look, while the average global firm plans to invest about$186 million in AI in that span, there's a significant value gap.”

The Great Skill Reset in AI

44:09 to 46:00

Understand the importance of upskilling in the evolving AI landscape.

“They are hiring some, but really it's about upskilling their existing resources because they have the depth of knowledge about their organizations to allow them to get the most from AI.”

Addressing AI Concerns in Business

46:00 to 47:40

Explore the responsible use of AI and its societal impacts.

“which is just one example, but they are at the forefront of at least marketing the virtues of an AI agent.”

Harnessing AI in Everyday Business

47:40 to 49:09

Gain insights on how individuals and businesses can adopt AI effectively.

“And so I think being responsible and addressing those within a governed framework allows enterprises to be able to move forward with confidence and hold themselves accountable to what they're doing.”
Hear the part that matters, and keep it.Open this episode in VO. Double tap your headphones to save a moment as you listen.
Get VO free

Transcript

Automatic transcript. May contain errors.

0:00The right technology can strengthen human judgment. That's why Deloitte brings together AI and data analytics with multidisciplinary teams who can help you connect the dots across your enterprise. From risk to operations to customer needs. So opportunities don't slip by and surprises don't spread. Because the smarter your systems, the sharper your instincts. That's how technology makes people better at what they do best. Deloitte. Together makes progress. Learn more at Deloitte.com slash Together Makes Progress.

0:58The Visibility Gap, a podcast presented by Cigna Healthcare. Everyone has been there. Your team's feedback is scattered across emails, chats, and sticky notes. It's a mess. But PDF Spaces and Adobe Acrobat gives you one collaborative workspace to streamline every file and comment. So, if you need six departments to finally agree on a proposal, do that with Acrobat. Need to turn a mountain of feedback into one plan of action? Do that with Acrobat. Want to stop searching for files and finally get everyone on the same page? Do that. Do that. Do that with Acrobat. Learn more at adobe.com slash do that with Acrobat.

1:43Bloomberg Audio Studios. Podcasts. Radio. News.

1:53Bloomberg Tech is live from coast to coast with Caroline Hyde in New York and Ed Ludlow in San Francisco.

2:02Caroline Hyde:This is Bloomberg Tech. Coming up, Alphabet's Google Cloud division unveils the latest generation of its Tensor Processing Unit, or TPU, and announces a slew of partnership deals will break it all down. Plus, a small group of unauthorized users has accessed Anthropik's new AI model, Mythos, which the company says can enable dangerous cyber attacks. And we'll catch up with Rivian CEO RJ Skaring to discuss a big milestone for the company's new R2. That's later this hour. First, we check in on these markets, which manage to be shrugging off, yet further chaos that occurs in the Strait of Hormuz. The oil market doesn't shrug it off, and we're currently seeing Brent up almost 3%.

2:40We're back above$100 a barrel. What's interesting is equities managed to push through that inflationary headwind and that geopolitical anxiety. In fact, we've got new record high on the NASDAQ 100, S &P 500, very close to that record. So, Ed, there is a lot of optimism. A lot of it's around earnings and other news.

2:58Caroline Hyde:Yeah, news is driving stocks. Google had a big jump in the pre-market, carried over into the main session and is sustained up 1.7%. There is a lot of Google news to wade through. First, let's talk about tensor processing units or TPUs. Alphabet's Google Cloud division unveiled the latest generation of its TPUs designed to make AI computing services faster and more efficient. The new lineup comes in two versions. The TPU 8T for creating AI software and the TPU AI for running AI services that have been created. In other words, Caro, training and inference specific. I mean, it sounds like Amazon in many ways with their chips they've been making, but Google has led the charge in terms of who designs their own chips and then vertically integrates.

3:43It's really led the path.

3:44Caroline Hyde:We talked earlier in the week that there would be an inference-specific TPU. We got both, a training and an inference-specific TPU. What else? I think it's important that Google had been reporting, Bloomberg had been reporting all of this, but also let's talk about more in the world of Google Cloud because the company has just expanded several partnerships. Oracle, to give joint customers a simpler way to interact with their Oracle data using natural language. But there's also with NVIDIA to help power agentic and physical AI in AI factories. There are also agreements with Salesforce, CrowdStrike, Broadcom, startups such as Miramirati's Thinking Machines.

4:17There was a whole wealth of headlines we had to navigate through. But maybe it's the fact that they're making their own chips is becoming more efficient. And that makes more people want to come in on the cloud relationship.

4:27Caroline Hyde:Let's be honest, this is a massive news dump from Google Cloud. And there are signs that internally more code is being written with AI, but they're doing deals. And cloud could be taking market share. Let's stay on Alphabet. Mandeep Singh for Bloomberg Intelligence is with us. I don't know if you were going to publish one React or a dozen React because there are so many pieces of new information. I suppose let's start on the TPU angle. That's a big focus for Google Cloud. Why? Why do we care so much about their custom silicon? I mean, look, right now it is about having the capacity to deploy AI workloads.

5:03That's what everyone is scrambling for. You have seen, you know, Anthropic really kind of change their models because they don't have the capacity to serve their best model. And look, Google has that advantage when it comes to owning that TPU supply and also deploying it at scale with their family of apps. And I think the best data point we are going to get in this earnings season is the cloud growth for Google segment. Already was close to 50 % last quarter when Azure grew 40%. So that gap most likely is going to increase. And not only are they growing faster than the other cloud players, their margins are expanding.

5:44And that's where people will realize they are the lowest cost token provider when you compare them to everyone else out there. We don't have to wait long to get that data. April 29th is when the earnings come. What do you make of all the expanded deals then? Does this just hint the more and more people are coming to them for the cloud? Does it matter whether they're startups or whether they're legacy businesses? Yeah, look, I think a lot of the deals involve vertical integration where they are providing the TPUs, they're doing it on their cloud, and it's using Gemini. So, you know, if you are using Google Cloud, you're getting that complete package, whereas with everyone else, you're using either NVIDIA chips or AWS cloud.

6:28And to me, the best illustration is, you know, the deal that Amazon has had with OpenAI and Anthropic, where they are giving Anthropic$5 billion money to spend on their cloud, whereas the deal that Anthropic had with Google, 3.5 gigawatts, Google is not giving Anthropic any money. So that is a sign that illustrates that Google probably is in a better position when it comes to negotiating with these frontier LLMs. And Ed, for our client base of a Bloomberg terminal, for example, you think about just the finance industry. Citadel's over there talking about how they're using Google. But there's a story that's really well read on the terminal talking about Vista striking a deal to speed up Google AI and its software portfolio.

7:11This is very much financially focused in terms of the industry that it's changed.

7:14Caroline Hyde:So let's just call Vista a private equity firm. or an investor that has a portfolio of companies. This is a kind of tried and tested format where Vista makes a deal with Google Cloud, but the net result is the portfolio companies get access to either the compute capacity or something on custom silicon, whatever it might be. But Vista has the clout to do that on the portfolio company's behalf. That's a playbook we've seen before, but yeah, spend a bit of money on AI. And spending they are. Mandeep, just briefly, how much do you think that this adds to the revenue run rate? how much does all of this really matter to Alphabet's huge business model, which isn't oiled about.

7:51Caroline Hyde:Also, real quick, like, well, I'm not going to read you every single headline, but insert company name here, expanded pack with Google Cloud. Some reports in the press, like with Thinking Machines, it's a multi-billion dollar deal. That will show up, right? Surely it will show up. I mean, Google's share of infrastructure cloud where, you know, think of the triopoly. There was Amazon, Microsoft and Google. It was always third, wasn't it? Way third. Well, they will move up. I think with AI workload share, they will certainly have a higher share compared to what they had in the traditional cloud.

8:24And this will probably be the biggest driver of that, owning the TPUs. And then obviously their chips doing far better than other ASIC providers, including Amazon.

8:33Caroline Hyde:I just want to go back to the basics of the TPU. You and I have discussed that body of research that you did comparing the existing generation TPU performance against other ASICs and NVIDIA. What's the basic story of the TPU? What's the selling point of Google saying you should use this instead of using the latest NVIDIA gear? I mean, look, these are not confirmed reports, but Mythos may have been trained entirely on TPUs. Or majority of the cluster for that Mythos training was Google TPUs. So from that perspective, if the best model out there, whether it's their own Google Gemini or, you know, Anthropics models is trained on TPU, that is a pretty good validation of how good their chips are both for training.

9:22And then they deploy that scale across their family of apps, search being that biggest kind of surface where they deploy their own models at scale. And that's all inferencing. We want to go back to something that you said at the very top of the conversation. You hinted that mythos may be being unfolded as it is because of lack of compute, rather than perhaps its all-powerful nature. What gives you that feeling, and how will the token economics change that for some of these LLMs in the future? Yeah, so right now, if you look at the last data point that Google has shared, they're around 1 ,500 trillion tokens per month.

9:58That's about three to four times that of OpenAI and Entropic based on our work. So they're already operating at such a high scale. Why? Because they are the lowest cost token provider. They don't have to get those chips from NVIDIA to, you know, deploy it for inferencing. And that is the difference, I think, that will become more and more obvious as we go towards agentic AI, which is basically doing inferencing much longer than using it in a co-pilot or a chatbot type of inference format. All night, all day, all night. Mandeep Singh and Bloomberg Intelligence, great to have you break down what is a whole swath of news around Alphabet.

10:38Coming up, Rivian's long-promised R2 starts rolling off the factory line. We'll speak with the CEO about the next chapter for the electric vehicle maker. This is Bloomberg Tech.

10:56Caroline Hyde:Rivian's new R2 is rolling off the line in normal Illinois. A big milestone for the company. It's a smaller, more affordable SUV. And for many, this is the vehicle that will define Rivian's path to profitability. The company continues to burn through cash. Joining us now from the factory is CEO RJ Scarrange, who has driven quite a few R2s off the line already today. it's a big moment but RJ we should point out that on Friday the facility was hit by a tornado by a big storm there was damage to certain parts of the the factory what was the net result of that is everyone okay and has there been any material impact to the the start of production plan yeah I mean well of course you can't plan for these kinds of things and we didn't expect a tornado to hit Friday night, but, you know, the teams responded really well.

11:51Fortunately, everyone was safe. We followed our protocols as soon as we knew there were tornadoes in the area. But yeah, tornado went through the South end of the plant and ripped the roof off the building and knocked down some of the, some of the plants as well. And so the last 72 hours have been around the clock. You're getting all the water out of the facility and getting ready to rebuild the south end but you know it's like the ultimate example of resilience of the business where the teams have come together we still had vehicles come off the line this morning is the plan still

12:24Caroline Hyde:the same you can get through the same kind of ramp strategy that you had the same volumes in in that initial output yeah our ramp up this week and into next week we're not making changes to the plan we're obviously having to make some accommodations for the fact that a big it's the logistics area of the plant. And so the way that material was flowing into the plant through the dock doors does not have to come in a different way. But we're working through it. We're working through it. What's interesting is orders, demand, configuration. RJ, what do people want and how willingly are they putting in the orders?

13:03Yeah, I mean, you said I drove a bunch of R2s off the line today. I've been driving an R2 for a while and the car is just incredible uh it's the combination of the package the driving dynamics the the efficiency uh of course the design and just the aesthetic feel so we're we're really excited to get folks into it uh there's a tremendous amount of enthusiasm we've had a few events where we've hosted potential customers to experience it and you know we were in denver a couple of days ago and there's like a super long line just to sit in the car so we love to see the enthusiasm. And I can't wait to continue to ramp up production and start seeing these on the road.

13:43I mean, I loved Ed's enthusiasm when he came to have a look at your in-house AI chip. And I'm interested as to where, therefore, that vertical integration comes in. You're also including LIDAR. Look, there's a roadmap for these variants. When does that enter production, RJ? So our in-house silicon, which is 800 tops per chip, we have two of those chips in the vehicle. and we couple that with cameras, radar, and a LiDAR that sits at the top of the windshield. That comes in at the end of this year. And so our launch edition still has a very highly capable self-driving platform, which is essentially a further development of what's in our R1 vehicle now.

14:19But across the vehicle, you know, R2 just represents so many learnings for us as a business, whether it's the driveline, the power electronics, just the way it's being built in terms of cost and cost efficiency. and so there's there's like not a there's not a detail in the car that hasn't been thoughtfully evolved or developed relative to what we what we've done in r1 rj i want to go a bit more

14:42Caroline Hyde:specific on what caro's talking about what the story with r2 has changed a little bit i think you're probably thinking about an autonomous future a bit more than you were but the idea was to get close to that 45 000 mark a vehicle for the masses you start with a premium a more expensive variant but what are you seeing in the demand pockets for getting to a lower price point that is where you want to take on the established oems right yeah well the average price of a new car in the united states is about fifty thousand dollars and so really important else as we're developing r2 is recognizing we wanted to you know be right in that range and so having a variant that starts at 45 and then going up to our top spec of for the performance version with, you know, all the kit for the interior is at 57.

15:35It allows us to really nicely straddle that. And of course, you know, launching any product, you have the decision of what do you launch first? You launch the base spec, mid spec, top spec. Right. And it's a school to sort of make everyone happy with when you have a launch edition. So we decided to launch with the premium spec, but we'll be introducing like a mid-spec vehicle, and then the$45 ,000 version will come thereafter. But that all rolls out over roughly the next year. And there's so much enthusiasm for the product that we feel really confident with the way we've configured the specifications.

16:11And even the pricing, we feel quite strong that we've priced it appropriately, especially when you look at the capabilities and then the cross shop of what other vehicles are in the space.

16:22Caroline Hyde:I've experienced the R2 that does have the LiDAR and the in-house silicon. And it's so interesting. There's going to be a generation of R2s that don't have it. I know that you said they're very capable with the NVIDIA silicon and without LiDAR. But that's a big call to make, right? Is there any flexibility to accelerate the timing of where you can start production on the R2 line in Illinois and bring in that gear? because what you're pitching for the autonomous capabilities, even the advanced driver assistance capabilities of R2, they are predicated on that next-gen gear. Well, the launch edition still has a very high ceiling, and so this will get to, and our Gen 2 R1 will also do this.

17:08Later this year, we'll start to roll out point-to-point level 2s, so meaning hands off wheel, eyes on road, but you type the address in and the car completely drives you there. And then into next year, we'll start to roll out level three. So hands off, eyes off for specific domains. In this case, you know, starting with highways. And that will be true for Gen 2 of R1. It'll be true for the launch edition of R2. And of course, be true for the variant of the vehicle that has the higher compute stack and more capable perception. And what that more capable perception is going to allow for is a higher ceiling of what the vehicle can ultimately achieve.

17:46And so it'll achieve higher levels of capability. And importantly, it serves as a really valuable part of our data flywheel. And so think of the R2 fleet. It's a high volume product as being part of training what we call our large driving model. And the existence of a LiDAR plus the enhanced inference just allows us to do a better job of capturing lots and lots of driving miles to train our model. Look, other automakers are probably looking at that flywheel, RJ. I'm really interested in some of the hints we're getting, some of the coverage in the markets about discussions to license your software, your autonomy stack, your underlying technology to other automakers.

18:27Is that in the works? Well, I think in the fullness of time, when you look at what we're building as a business, there's really two revenue opportunities. One is we continue to grow by selling vehicles. And so, of course, R2 is the embodiment of that. That's what we're driving towards. But there's another element of our business, which is leveraging the technology we're developing. And the first example, that was what we did with Volkswagen. We did a$5.8 billion software licensing deal. And that allows us to deploy our, this is not our self-driving, but our compute stack and our software platform, our OS, across a wide range of electric vehicles within the Volkswagen portfolio.

19:10And so we're excited about the potential to do more deals like that with our operating system in the vehicle, with our electric vehicle. And certainly as we think about autonomy, the potential that's there for us to also license that in the long term. Rivian CEO RJ Skarange we'll let you get back to those R2s that are rolling off the shop floor we so appreciate your time today from Illinois meanwhile electric vehicle adoption it has been accelerating particularly in China and increasingly spilling into global markets driven by a wave of lower cost models and the country's lead in cheaper battery production it's the topic of tonight's episode of Primer take a listen you're not driving a car you're driving a battery and China right now is the leader in batteries China's EVs are already cheaper on average than similar gas cars.

19:58And it's no surprise that four of the top five battery makers in the world are located here. Also one of them, BYD, has been jockeying with Tesla for years for the title of the world's biggest EV maker. All that success means lots of competition at home, propelling China to look outward for expansion. And while some big markets like the US and the EU have resisted China's cheaper EVs, Many emerging markets have brought them in.

20:27Caroline Hyde:You can catch the full episode of Primer tonight on Bloomberg, 6 p.m. Eastern, and on Bloomberg Originals at 8 p.m. Terrific work from Caro and the team. Now, coming up, Mythos, the AI moth model that Anthropic warned was too big a cybersecurity risk to release widely, has been accessed by unauthorized users. That's according to a Bloomberg source. We have that report next. This is Bloomberg Tech.

20:55The right technology can strengthen human judgment. That's why Deloitte brings together AI and data analytics with multidisciplinary teams. People with deep industry experience who can challenge assumptions and help you connect the dots across your enterprise. From risk signals to operational pressure points to shifting customer needs, Deloitte helps you see what's coming sooner. So opportunities don't slip by and surprises don't spread. It's not just dashboards. It's real clarity in the moments your decisions are made. When models reveal patterns, people can ask better questions. When data and people are connected, leaders can move faster with confidence.

21:36And when your teams are aligned, smart choices can scale from the front line to the C-suite. Because the smarter your systems, the sharper your instincts. That's how technology makes people better at what they do best. Deloitte. Together makes progress. Learn more at Deloitte.com slash together makes progress. Support for the show comes from Public. Lately, it feels like there are two types of investing platforms. Some are traditional brokerages that haven't changed much in decades, and others feel less like investing and more like a game. Public is positioned differently. It's an investing platform for people who are serious about building their wealth.

Read the full transcript

22:13On Public, you can build a portfolio of stocks, options, bonds, crypto, without all the bugs or the confetti. Retirement accounts, yep. High-yield cash, yes again. They even have direct indexing. Public has modern design, powerful tools, and customer support that actually helps. Go to public.com slash market and earn an uncapped 1 % bonus when you transfer your portfolio. That's public.com slash market. Add paid for by Public Holdings. Brokered services by Public Investing, member FINRA SIPC. Advisory services by Public Advisors, SEC Registered Advisor. Crypto services by zero hash. All investing involves risk of loss.

22:51See complete disclosures at public.com slash disclosures. Small businesses are the pulse of every community. They bring people together, create opportunities, and drive growth. With a widespread presence in communities across the country, Chase for Business supports small business owners at a local level. That makes it possible for you to connect, learn from each other, and grow together. There's a real commitment to seeing small businesses succeed. The Chase for Business team has knowledge and expertise that span a wide range of financial areas. They can help you make more informed decisions as you navigate the complexities of running your business.

23:25They'll help your business grow with individual guidance and convenient digital tools all in one place. With that guidance and your determination, you can take your business farther and help build a brighter future for your community. Learn more at chase.com slash business. Chase for Business. Make more of what's yours. The Chase mobile app is available for select mobile devices. Message and data rates may apply. JPMorgan Chase Bank N.A., member FDIC. Copyright 2026, JPMorgan Chase and Company.

23:58A small group of unauthorized users have accessed Anthropik's new AI model, Mythos. It's according to a person familiar with the matter and documentation viewed by Bloomberg News. Now, as you probably remember, Anthropik has said Mythos can enable dangerous cyber attacks. And because of that, the company only gave access to select users to help them probe their own systems vulnerabilities. Let's get the latest of Bloomberg's AI reporter, Rachel Metz. This caused quite the stir when you reported it last night. And what's so extraordinary is who's done it and how they did it. Rachel, these aren't people wanting to wreak havoc.

24:31They want to play with the thing. Exactly. Yeah, these are people that enjoy playing with new AI models. and they had been trying to figure out the names of various AI models for Anthropic in particular through a variety of techniques such as having some familiarity with what the online location of Anthropic's old models looked like, what that URL would look like, and using that to guess as to what it might be in the future.

25:02Caroline Hyde:so being very specific about this Rachel one of the uh of the group which was uh a discord community right had a contractor relationship with anthropic so had some sort of credentialed access but the rest of it was sleuthing right thinking where is the the the source code for this likely to be explain literally the mechanisms by which they found mythos sure so I mean the key piece here is that one of these people has worked as a contractor for a third party, not for Anthropic directly, but for a third party that works with Anthropic. And through that, they had a authentication that they would need in order to access certain models.

25:47With that, and then with some sleuthing that they did, including details that had been involved in data breaches online, and just their past experience. They sort of put these pieces together of what the online address could be for the Mythos model. And all those things together led them to be able to access it on the same day, they say, as when Anthropic announced it. I can't help but feel that this is terribly basic cybersecurity practice 101 that was missed here over at Anthropic. And what can this therefore mean when they're meant to be very carefully through Project Glasswing allowing other people to use it.

26:30Yeah, I mean, I think that's one of the things that the person that I spoke with was really concerned about. They are part of a small group that is just playing around with this. From what they said, they're not doing anything that they shouldn't be doing with it. Obviously accessing it in an unauthorized way, but not doing anything bad with it. they think, well, if it was this uncomplicated for us to get access to it, what does that mean for other people? Are there other people working for third-party contractors who could be nation state actors or something like that, that are using access, legit access seeming to technology and then getting access to mythos?

27:10Caroline Hyde:Right. The most read story across Bloomberg today, unauthorized access to mythos. Bloomberg's Rachel Metch, top reporting. Thank you. Coming up, SpaceX and Cursor are teaming up to, quote, create the world's best coding and knowledge work AI. We have the report next. It's half time. And this is Bloomberg Tech.

27:38Caroline Hyde:Welcome back to Bloomberg Tech. If you're just joining us, there are so many news headlines that are driving individual names in the markets. There's a report from the information that DeepSeek, the China-based Frontier Lab, is thinking about doing a round at a$20 billion value, but that specifically Alibaba and Tencent, who are not yet investors in DeepSeek, are considering participation. Right now, the US-listed shares of Alibaba up 1.3%. It did shoot up in the pre-market after that report hit. Again, that's from the information. We have not yet verified it, but it would be an interesting development, Caro, for China's domestic AI ecosystem.

28:17It certainly would be. And it just thinks about the story that we continue to hear of M &A in ecosystem, in AI ecosystem. And let's discuss that a little bit more closer to home. Elon Musk is moving to catch his rivals in AI coding tools through a potentially$60 billion deal. Now, SpaceX has secured the right to acquire AI startup Cursor, but it's not pulling the trigger just yet ahead of its IPO. But he goes, Natasha Mascarenas joins us now. and Natasha. So after much sort of excitement about Cursor and other companies looking to potentially purchase it, it looks as though SpaceX has got an edge.

28:52It sounds like it. I mean, this is a very unique, a very Elon Musk deal. It's the right to acquire Cursor later this year for$60 billion, like you said. If not, as we exclusively reported yesterday, the$10 billion dollar fee is actually being looked at as a breakup fee. And so in the next six months, what have you, we expect to see some sort of update there. And Cursor no longer need to look at venture capitalists for its compute needs for the short term as well.

29:23Caroline Hyde:So on that point, we had reported that Cursor was in talks to raise funds at a$50 billion valuation. I'm just going to point out the obvious, that SpaceX, which owns XAI now, has filed confidentially for an IPO. And on paper, this is a big piece of M &A, which would materially change the parameters of that S1. Give us the backstory. This is not straightforward, Natasha. Not straightforward at all. I mean, listen, to your point, we're not going to see, we're not expecting to see SpaceX make an acquisition right before the IPO. It would require a huge upheaval. However, I mean, this is a huge change to the SpaceX IPO narrative, right?

30:05About two months ago, Elon Musk announced that XAI and SpaceX were merging. Now we have this deal coming out of the woodwork. It's a huge update. And listen, I mean, I think Cursor shareholders are extremely happy with the news. Around this time last year, Cursor was anticipating being valued at around a$10 billion valuation. Now$10 billion is the breakup fee. And so talk about a huge change in sentiment around Cursor. Natasha, just remind us of the space in Vibe coding or coding more broadly as well, because it's very competitive, but also it feels as though XAI and SpaceX want to catch up here.

30:43Absolutely. I mean, this is the same reason that we saw Windsurf, another competitor, a once a standalone competitor to Cursor last year seek compute partnerships and opportunities with bigger technology companies. Windsurf, of course, saw its co-founders leave to join Google and Cognition, another coding company, acquired up the remaining assets. All of these companies are looking for more ways to secure compute. And so it's a deeply competitive environment. When the deal was announced yesterday on X, understandably, we saw Cursor CEO Michael Truel announce that first and foremost, this is a win for Cursor's agentic coding model.

31:22And so this is very much a response to the competitive environment because they need more and more compute and this solves that number one bottleneck.

31:30Caroline Hyde:Bloomberg's Natasha Mascarenas, thank you very much. Shares of T-Mobile in the US session are down 4%. Deutsche Telekom, which just closed trading in Europe, down 4.65%. Deutsche Telekom is said to be considering a merger with its US arm, a deal that would create the world's largest telecom company. That's all according to sources. For more, Bloomberg's Telekom reporter Kelsey Griffiths is with us. I don't know. I didn't have this on the bingo card, but there is a backstory, a shared history, particularly at the C-suite level. What are we reporting? What do we need to know? I can say I didn't have this on my bingo card either, Ed, but I do think that this is something that has been rumored in the telecom industry for a while.

32:14I had some sources tell me this was an open secret. This is something the T-Mobile parent company had thought about for a while. It comes and goes every few years and we're in one of those cycles. It's really interesting to see what could potentially happen here. According to our great colleagues reporting, there could be a holding company that would essentially bring in both the parent company, Deutsche Telekom, and T-Mobile and unite them under one sort of umbrella. What's wild? Tim Hurtges, who's over at Deutsche Telekom, has been frustrated for a while probably with the fact that Deutsche Telekom is worth less than T-Mobile, even though they own a little bit more than a majority of the U.S.

32:57arm. It sort of comes full circle. How is this going back in history in some ways? Right. We've seen a relationship, a back and forth relationship between the Deutsche Telekom parent company and T-Mobile U.S. The current T-Mobile US CEO, Srinya Gopalan, is actually a former T-Mobile or rather Deutsche Telekom CEO in Germany. So he has a close relationship with the current Deutsche Telekom CEO. They are really aligned on strategy. They're aligned on trying to increase profits, increase market share. And I've been told the two men are very close. So there is an alignment there. What we don't know is the reaction from T-Mobile domestically.

33:44It does seem like the alleged deal or potential deal would allow the parent company to kind of increase their control over this very profitable U.S. branch of the company. There's a lot of question marks, regulatory, about approvals, about whether it would spark more M &A in the U.S. And analysts are trying to get a grip on it as well as investors. Blueberry's Kelsey Griffiths, great to have you on. Thank you very much indeed. Coming up, an exclusive conversation with Vast Data CEO, Brennan Hallack, as the company hits a$30 billion valuation after raising another billion. This is Blueberry Tech.

34:22Support for the show comes from Public. Public is an investing platform that offers access to stocks, options, bonds, and crypto. And they've also integrated AI with tools that can assist investors in building customized portfolios. One of these tools is called Generated Assets. It allows you to turn your ideas into investable indexes. So let's say you're interested in something specific like biotech companies with high R &D spend, small cap stocks with improving operating margins, or the S &P 500 minus high debt companies. Chances are there isn't an ETF that fits your exact criteria. But on public, you just type in a prompt and their AI screens thousands of stocks and build a one-of-a-kind index.

35:02You can even backtest it against the S &P 500. Then you can invest in a few clicks. Go to public.com slash market and earn an uncapped 1 % bonus when you transfer your portfolio. That's public.com slash market. Ad paid for by Public Holdings. Brokered services by Public Investing, member FINRA SIPC. Advisory services by Public Advisors, SEC Registered Advisor. Crypto services by ZeroHash. Sample prompts are for illustrative purposes only, not investment advice. All investing involves risk of loss. See complete disclosures at public.com slash disclosures. Being a small business owner isn't just a career, it's a calling.

35:38Chase for Business knows how much heart and effort go into building something of your own. That's why they make business growth their priority. The Chase team takes the time to understand your mission, where you are now, and where you want to go. Their broad range of solutions is designed with you in mind so you can bring your ideas to life. From banking to payment acceptance to credit cards, you can conveniently manage all your business finances all in one place with their digital tools. Looking for tips and advice? Their online resources are always available to give you the solutions you need to help your business thrive.

36:12See how your business can get stronger and go farther with Chase for Business. Learn more at chase.com slash business. Chase for Business. Make more of what's yours. The Chase mobile app is available for select mobile devices. Message and data rates may apply. JPMorgan Chase Bank N.A. Member FDIC. Copyright 2026. JPMorgan Chase and Company. We buy insurance for peace of mind, but every year millions of claims are denied. Not because people did anything wrong, but because their policies quietly excluded what happened. Insurers know every detail. Policyholders rarely do. That's why My Policy Advocate exists.

36:48For just 27 cents a day, their platform reads your policies and explains where you are vulnerable. They don't sell insurance, they deliver transparency. Before you trust your policy to protect you, let My Policy Advocate tell you what it really says. Go to MyPolicyAdvocate.com.

37:09Caroline Hyde:Software maker Vast Data has announced it's raised a$1 billion Series F round, more than tripling the company's valuation to$30 billion. Here to discuss is Vast Data CEO, Renan Halak. Some big numbers. Congratulations. You must be pleased. But Jensen Wong talks about this five-layer cake of AI. I think we're trying to understand what vast data does, what it is. You know, you describe yourself as an AI OS, but let's start there. What do you do? Yeah, the five-layer cake, Jensen made this analogy of power, chips, infrastructure, models, and applications. We're that middle layer. we try to fill in that software infrastructure layer of the stack.

37:59And we started from storage, building very large data storage systems. And then we added structure data with a database and a data engine to manage compute. And as we progress, more and more parts of that stack need to be filled out for AI to become widespread.

38:19Caroline Hyde:So here are some of the companies and groups that you work with, right? And there is all kinds of AI infrastructure, but NASA's on there as well. Pick anyone. Give us a case study of the work that you might do with one of those key customers. So, for example, with CoreWeave, they're the biggest AI cloud out there, and they're the way for us to get to these new AI startups, companies like Cursor and a lot of others, as well as the large AI companies out there. They're serving everybody, and we're helping them, again, provide those cloud services by having that software infrastructure layer in place.

39:02The money, Renan. What is it that you need the money for, and why, potentially, do you need even more money? You're considering potentially an IPO by the end of this year. So we don't need the money, is the honest answer. in parallel to tripling the business roughly year over year. We've been very efficient in the way that we build the company. We've been generating quite a bit of cash. In fact, on the rule of 40, we have a rule of 228 as of last year. The funding round is partially primary, partially secondary. And we expect to continue to grow the business with a very, very strong balance sheet to give our customers confidence that they can make very large bets on us.

39:48A lot of American-focused customers that we just saw, but you have been going global. And it's interesting that you've taken funds from UAE, from Singapore. How are you seeing the global expansion of fast data? So wherever AI is, that's where we see very fast growth. It's been surprising at times. The Middle East, Singapore are places that are moving faster. than some of your more traditional markets. Still, still the Middle East is running at this moment where people perhaps question how much they can focus in on investing on AI, particularly in the UAE, when we're concerned about geopolitics and war.

40:27I think that's a temporary thing. Hopefully we come out of the other end stronger and safer. Definitely the last couple of months have been challenging, but if I look at five years out, ten years out, I still think the UAE can leapfrog over a lot of other countries.

40:43Caroline Hyde:Dear Bloomberg Tech audience member, the rule of 40 is something that software companies do, where they think their revenue growth and their profit margin should combined be above 40%. But you know, Renan, the other type of company that talks about the rule of 40 is one that thinks about going public. Is that in your plans? We don't have a timeline. We've definitely been preparing. Amy, our CFO, she came about 18 months ago from Shopify, and she's been working very diligently to make us ready to be a public company. My best guess is that sometime later in this year, we will be ready for IPO, and then we'll need to decide if and when to pull that trigger.

41:28Caroline Hyde:In your world generally, particularly the data layer of what's happening in AI, or Cursor, that was a nut story, SpaceX, the right to acquire Cursor. What do you think it is that the market's assigning value to? What is it that they see in companies like yours and say, wow, yeah, this is worth the premium? So SpaceX, Cursor, amazing examples, XAI, all of which have been longstanding customers of ours. We love to see this type of consolidation happen when AI companies and AI teams work together to build even faster and even bigger things. And that's what we enable. Our system enables very, very large scale, very large capacity and ability to store vast amounts of information and give access to that information to the largest GPU clusters in the world.

42:22It allows our customers to organize their data, to make sense of it, to generate insight from it, and to build these new workloads in a way that could not be built before. Renan Halleck of Vast Data, raising a billion at a$30 billion valuation and mulling an IPO by the end of the year. We really appreciate you coming on. Come back soon. Have a feeling you might. Meanwhile, a new global report from KPMG finds that nearly three quarters of business leaders say, look, AI will be a top investment priority, even if recession occurs over the next year. Look, while the average global firm plans to invest about$186 million in AI in that span, there's a significant value gap.

43:01Nearly all have an AI strategy, but only 8 % are seeing a clear return on that investment. Rahsaan Shias, KPMG principal and AIQ program lead, joined us now to discuss this data. And you're going around surveying big companies with big revenue flows. And no matter what, they're all still all in on the AI investment, even without an ROI. Absolutely. We see spend, if we look at the survey results, are double what we saw in Q1 2025. And in the U.S., that's$207 million. If we look out in Asia-Pac, it's over$240 million. And so there's no turning back. Everyone is all in. And what you just heard earlier today on the show is there are new capabilities that are coming out that allow our clients and companies to expand from pilots who are across the enterprise.

43:50And that requires this level of spend. This level of spend needs to be met with capability and with training. And, Rahsaan, when people are spending that amount, what are they actually spending it on? Yeah, we are seeing them do everything from take agents to reimagine workflows across the enterprise to upskilling their professionals, right? They are hiring some, but really it's about upskilling their existing resources because they have the depth of knowledge about their organizations to allow them to get the most from AI. And so we're calling this the great skill reset because there's so much investment and so much focus on the human element to optimize AI in this day.

44:32Additionally, there's lots of focus on governance and making sure that organizations are addressing the risks associated with the expansion of AI across the enterprise, from data privacy to security to the ability to actually deliver their products to their clients in a way that their clients will trust.

44:54Caroline Hyde:a conclusion of the survey which is of 2100 c-suite names and and the report is that ai agent adoption is accelerating but what i found interesting within the data is to me it seems like from a low base you're saying 32 percent of those surveys are actually deploying agents does that come across to you as a low base it's not a low base because what we saw in earlier results was that everyone was in this pilot mode. And actually, we see a dip of those companies that are piloting only from 61 % down to 30%. And so what you see in this is this shift. And so we're excited for next quarter's results because we expect with the introduction of new capabilities, the increase in available compute, that we are going to see this number continue to grow at a very rapid clip.

45:46And that's what our clients are coming to us and saying they need to enable in their organizations in order to actually, which you mentioned earlier, get to that value and see that value at scale across their enterprise.

45:59Caroline Hyde:Because a company like Salesforce, for example, which is just one example, but they are at the forefront of at least marketing the virtues of an AI agent. They get criticized for actually converting that into sales and then have customers do something meaningful with it. So the other data point is that 27 % of those you surveyed have multiple agents. It's more than one. Where are you seeing the sort of economic benefit being generated? People making revenues from the investment they've made. Yeah, it's absolutely from either enabling new services. So we're seeing lots of that where they're able to enter new markets.

46:36The second is allowing for vertical integration of services with other providers. to be able to deliver things in a different way. And then lastly, looking at the key economic levers of any organization and how they can take not only costs out of that, but actually increase speed. Right now, the currency is speed to execute. And so it's either to get new services out or to deliver what you do today in new and different ways. It's what we're doing as an organization. It's what we see our clients doing that is driving the ability to get to that value. Rahsaan, this is a B2B conversation, and rightly so.

47:13I want to talk to the disconnect that's happening at the moment. You talk about the skilling of the workforce. I just think about what's happening online, Reese Witherspoon daring to dip her toe into the idea that the women should be using AI in their everyday lives, and the hate, the disgust coming from many around it. What do you see in business? Is there a pushback? Is there a concern about environmental impact, about royalties, about the impact of AI writ large on jobs? Yeah, absolutely. And most companies that are leading the path, those that are at the front, they are addressing these challenges head on, talking about what they're doing to be responsible with their AI use and how they are holding their vendors responsible to do the same and really addressing the known things and the things that are unknown, particularly around sustainability and what the impacts are to the environment, as well as the impact to the human experience as it relates to work.

48:12And so I think being responsible and addressing those within a governed framework allows enterprises to be able to move forward with confidence and hold themselves accountable to what they're doing. The other thing I would tell you is, in this day, the access to AI capability is very advanced for consumers. And, you know, the pushback, Therese, she's absolutely right. Everyone should be out there trying to build their AI capability, their AI skills. And no matter how you start, that's okay. If it's helping you pick out an outfit, if it's helping you manage your schedule, if it's creating a new capability that the world has never seen, whatever you do, the key is start.

49:00Because the gap between those who are leading and those who are just joining is widening. And so there is no better time to make sure that you're upscaling yourself.

49:09Caroline Hyde:Rahsaan Shears of KPMG, thank you very much for joining us on VTech. Coming up, we're going to look ahead to Tesla's earnings, which are after the closing bell. That's next. This is Bloomberg Tech.

49:24Caroline Hyde:Tesla reports earnings after market closed today, but a blowout might not be enough for the stock, which is down nearly 13 % so far this year. The EV maker doesn't really trade on the serve cars anymore anyway. Investors focused on AI and robotics. Bloomberg's EV reporter, Cara Coulson. So it doesn't really matter what they say, right? RoboTaxi, Optimus, keeping everyone looking to the horizon? What do we need to know? Yeah, so there's, as you mentioned, a lot of people are very much looking ahead. They're looking at RoboTaxi, which is still kind of getting off the ground for the most part for Tesla.

49:57They're looking way ahead to Optimus, the humanoid robot. But there's also a lot of near-term concerns that people are looking for, including, you know, cash burn and whether Tesla can keep up, you know, keep putting revenue out to kind of counter that a bit. And no one's expecting huge Cybertruck sales, but where are we thinking about RoboCab or about things getting into customers' hands in the near term in terms of supply chain? Yeah, so Tesla is, you know, expected to put out their start doing volume production of their Cybertruck, which will primarily be used within their own RoboTaxi. There's not a lot of near term catalysts as far as their current lineup.

50:40They really have the Model Y, the Model 3, Cybertruck. So sales are becoming increasingly less relevant to a lot of analysts and a lot of investors in the near term. All the questions about humanoids and optimists. Bloomberg's Cara Carlson, it's great to have you. You'll be busy tonight, as will Ed. That does it for this edition of Bloomberg Tech. But remember, all the earnings after the bell. Tesla, IBM, we've got Texas Instruments.

51:04Caroline Hyde:Yeah, it really starts in earnest, and we will get a good global lens of the health of that industry. Don't forget, it was a huge news dump this morning. Recap all of it on the podcast. You know where to find it. Apple, Spotify, iHeart, and of course, on all of our Bloomberg platforms. From New York City, this is Bloomberg Tech.

51:25When you're running a business, the best days are the ones where priorities stay on track. For midsize and large companies, risk can affect multiple parts of the organization at once, from property and liability to cyber and regulatory challenges. At that level, managing risk becomes an ongoing discipline. At the Hartford, the focus is on helping businesses manage risk before it turns into something more disruptive. And when losses do happen, that work is paired with insurance coverage shaped by years of underwriting, risk engineering, and claims experience. Learn more at thehartford.com slash risk mitigation.

52:00Policies provided by Hartford Fire Insurance Company and its property and casualty affiliates, Hartford, Connecticut. Find home wherever you roam at Sonesta ES and Simply Suites, where longer stays feel comfortable, flexible, and easy. Stretch out and enjoy spacious accommodations and home-like amenities designed to help you settle in and stay productive or relaxed for however long you need. And when you're a Sonesta Travel Pass member, staying at Sonesta ES and Simply Suites means earning points toward free nights, upgrades, and more with every eligible stay. Go to Sonesta.com to book your stay and unlock the best rates with Sonesta Travel Pass.

52:36Here today, roam tomorrow. Join now at Sonesta.com. Terms and conditions apply. These days, it seems like AI agents are just about everywhere you turn, every field and every function. But without identity, you can't trust they'll serve your business instead of jeopardizing it. Fortunately, Okta helps you get identity right by securing your AI agent's identities, giving you a single layer of control, a single standard of trust. So whether an AI agent supports a single user or your entire enterprise with Okta, you'll turn risk into opportunity. Secure every agent, secure any agent. Okta secures AI.

From the publisher

Bloomberg’s Caroline Hyde and Ed Ludlow discuss Google Cloud's latest generation of tensor processing units, or TPU, and the Alphabet division’s new partnerships. Plus, a small group of unauthorized users has accessed Anthropic's new AI model, Mythos, according to a source and documents reviewed by Bloomberg. And Rivian CEO RJ Scaringe joins as his company’s new R2 model, its smaller, cheaper SUV, begins rolling off the manufacturing floor in Normal, Illinois.

See omnystudio.com/listener for privacy information.

More from Bloomberg Tech

All 343 episodes
Google Cloud Debuts New AI ChipsBloomberg Tech · 44 min
Listen in VO