Meta’s $200 AI Agent ‘Hatch’, Sam Altman-back Helion Valuation Hits $15.5B, Snowflake’s New AI Tools

0:13Welcome, everyone, to The Information's TI TV. My name is Akash Pasricha. It is Thursday, June 4th. We are watching chip shares today as Broadcom reported quarterly results that disappointed investors. Broadcom shares were down roughly 15 % early this morning. Micron and Intel shares are also down. We have more coverage of Broadcom results on our website. Today on the show, Meta is looking to charge up to$200 a month for its high-end AI agent. We're going to unpack the information's exclusive reporting shortly. We then have exclusive news that Helion, the nuclear fusion company backed by Sam Altman, has raised nearly half a billion dollars in funding.

0:53We'll talk to the company's founder and CEO, David Kirtley. We're also breaking down the latest quarterly results from cloud security company Netscope, which reported a deceleration in its business. Netscope CEO Sanjay Barry will join us shortly. We're talking with the chief data and AI officer at Snowflake as the company wraps up its big conference. We've also got our AI reporter coming on to tell us about her conversation with the co-founder of Databricks and Perplexity. It's going to be a busy show, but a fun one. So let's get right on into it. Meta is looking to charge up to 200 bucks a month for an agent tool that resembles OpenClaw.

1:32That is according to an exclusive report from my colleague Jyoti Mann at The Information. I want to bring on Jason Dean, our San Francisco Bureau Chief, to help us break it all down. Jason, welcome back to the show. It's great to have you here. Great to be here. Thanks, Josh. So what do we know about this agent tool that Jyoti reported on? Yes, as you said, they've been discussing internally charging as much as$199.99 a month. They're calling it Hatch. And the idea is for it to be a sort of consumer-friendly, user-friendly version of OpenClaw, which is, of course, the agent tool that has caught fire in the tech community, but is also notoriously been sort of difficult to use, not very layman-friendly.

2:17So this is – the idea here is for Meta to build something that can be used for consumers and for prosumers. To make their own agents, essentially. Right, right. This can be used. It can Vibecode according to these internal documents that Jyoti got to see. It can also do basic agentic tasks like put things on your calendar, send emails. Of course, that itself is interesting because Meta does not have an email program or calendar program that people use. So it's obviously going to have to be able to do things off Meta platforms. And do we know when this is going to be released at all? They had been looking at April.

3:03That's obviously come and gone. Other documents talked about trying to get it ready by July. I think the timing is fluid. This is obviously a big endeavor for meta, not the kind of thing that it already does. And so, you know, I think mid-summer was the goal, but I could easily imagine it going later than that. And we'll talk about the broader strategy here in a second, but do we know$200 a month, is there going to be a free version at all? Is that in the mix? They're talking internally about a tiered version. And I think there would be a free version. And the idea is that the premium version or premium versions would give you a lot higher usage limits, similar to what we've seen from other AI offerings.

3:50Right. Okay, so you touched on this. But, I mean, this seems like a bit of a leap, right? I mean, these are not the products that we know Meta to be involved with. We also know that it's trying to develop the AI models themselves. I mean, what's your take here? Does this feel like a bit of a moonshot bet that is likely to be scrapped at some point? What's your best guess here? Well, I don't know. I would predict now that it's going to be scrapped, but certainly its success is not guaranteed. Yeah, as you point out, as we were just talking about, this is not in their typical sort of wheelhouse.

4:29But at the same time, they're spending an enormous amount of money on AI infrastructure, on AI models. And that's shown benefit in their core business, which is 98 % advertising. It's not necessarily enough to justify them being at the same scale of investment in infrastructure that the hyperscalers are that are renting out data centers, Microsoft, Google, et cetera. So Mark Zuckerberg has been very, very open about the fact that they're looking for new revenue streams. They have a lot of ambition around agents. They've already been rolling out agentic functions on their existing platforms, and those are designed to do things like enable companies to do tailored messaging, interact with customers on WhatsApp, on Instagram now.

5:26But this idea of doing something that would range off platform, that would have much greater functionality than anything that they offer really at the moment is new territory. And I think, you know, they're going to have to build a lot of muscles that they don't have. Right. And one piece of the story that I think is important that Jyothi mentioned is that they're using Claude's models right now in the testing phase here, which is probably an indication in and of itself that they're using somebody else's models to test it. But I also do think, though, and maybe I'll run this by you, Apple didn't use their own.

6:04They went with Gemini, right? So I wonder to what extent you'd think it's a possibility Meta could just roll out some of these products not using their models. Do you think that's an option? They have put a lot of time and energy and effort over several years into building foundation models. They haven't been hugely successful by the standards of the other companies that are spending as much as they are on this. But I think there's definitely a goal there to have their own core technology under this, among other reasons, because it would be very expensive to roll out a sort of wrapper product built entirely on Anthropix models.

6:48But they're not there yet, clearly. They've announced these new models. They haven't built them into this system as of the sort of April-May timeframe that these documents are from that Jyoti was able to review. So I think that's a huge challenge. I'm sure that they want to build it on their models. We'll have to see whether they're able to succeed. Great. Well, Jason, I want to thank you for coming on. That is Jason Dean, our San Francisco Bureau Chief, here at The Information. Helion, the nuclear fusion company backed by Sam Altman, raised$465 million in funding from Thrive Capital and a whole host of investors at a$15.5 billion valuation.

7:33That is almost three times what it was valued at last year. I want to bring on David Kirtley, founder and CEO, to talk through this moment for the company. David, welcome back to the show. It's great to have you here. Thank you, Akash. It's good to see you again. Okay, so nearly$500 million in new funding. What are you going to use it for? Yeah, acceleration. So we are actually taking what we've already started, building Orion, the world's first fusion power plant, that allows us to take the infrastructure that we're already putting in the ground right now for that power plant and start building the generator and getting that installed.

8:07And then Omega, the world's largest capacitor and fusion pulse power manufacturing facility that we brought online last year. This allows us to pour more fuel on the fire, accelerate Omega manufacturing, bring jobs to the United States and get Fusion built and online. I want to talk about the manufacturing component in a second here. How many people is the company now? Yeah, we are 718 people as of as of this week, which is really exciting. Everyone's in Washington State or where are they? Almost everybody. We have a few people in D.C. and I'm really excited. Actually, we have just started our third hire out in Malaga, Washington, where we're building that power plant.

8:45And everybody else is in Everett where the main manufacturing is, where we build the components and then ship them to deploy them on site. Wow. You know, every time we have you on, you've raised a giant sum of money and the valuation keeps ticking higher and higher. You have become a bit of an expert fundraiser at this point. What have you learned about fundraising that you applied to this last funding round that you didn't know the first time you went down to pitch? Oh, that's a great question. A lot of it is is there's two parts to it in my mind. One, having that large scale vision. You got to be able to deploy.

9:21You have a technology that can deploy globally and make a real impact on the world and on the markets. And so that's been a big focus of Helion. on. And the other one, what's new, is ticking off all of those key business risks of deploying commercial power. So that means things like proving the technology, operating our seventh generation system, Polaris, setting world records for temperature, fuel, and operation. But then also on the regulatory risks, securing the first permits for environmental and building permits, and then commercial risk, the first PPA, power purchase agreement for power from Microsoft, and then continue to build that customer book.

9:57And that's really key to hit all those pieces. How about managing all, I mean, you've got so many names on your cap table now. I mean, what have you learned about, you know, managing these relationships, deciding who the lead investor is going to be? I mean, just reflect on those decisions for a minute. Yeah, so our$465 million fundraising round, our Series G was led by Thrive Capital. And for me, a lot of it, if you look at our cap table and how we have grown the company, it's about mission alignment. It's about finding those investors that are thinking about that scale, that they're really planning on how do you deploy at scale and make that huge impact on the business, on the market, and on the world.

10:40And so Thrive certainly is an alignment. You look at our past marquee investors and you see that too. And the new investors that we've been able to bring in on this round and our existing investors that have all joined in and invested in this round as well. Right. Okay. So everybody on that same mission. So TinyMerge is the new initiative that you're working on. This is a bit of a strategy change the way that I understand it. Walk us through what TinyMerge is and how you decided to adopt that approach. So this is actually the original and consistent Helion approach since we founded the company is that to build large-scale fusion systems and be successful and move quickly, iterating and building in parallel has always been the mantra.

11:23So we have just started up a new system that we're actually building right now to additionally do fusion to be able to move quickly. But in parallel to that, we have large-scale magnet test facilities, large-scale electronics test facilities, and all the manufacturing pieces. Because it's not just about building a machine, deploying it and turning it on, writing a bunch of papers, like that's great. But really the key here is deploy commercial fusion clean, safe power is that you're moving and iterating and building all in parallel and all in speed. And so you're testing the engineering, the physics, the manufacturing and the regulatory parts all in parallel at velocity.

12:03And our new program, Tiny Merge, is one of those components. But it's a smaller, am I not correct? It's a smaller sort of approach to Fusion, right? Yes, and it allows us to iterate quickly on materials and electronics for Orion for the power plant, and then also train operators. It's something to get the world's first Fusion power plant operators who are going to run this power plant for the first time up to speed. And so there's a whole variety of things that we can work to optimize Orion and that first commercial plant by having these iterative subscale test beds. And you're going to hear about more later this year on all the other systems that we're building too.

12:43So you basically decided to go with a bit of a smaller sandbox to prove out and test things out before applying it to the larger Orion facility. Am I understanding you correctly? Absolutely. And that's part of the engineering process of as you're going forward with these large scale programs in parallel, small scale test beds to learn, iterate, and advance the technology further. And that allows us to move faster. And that, I think, is the key. So the manufacturing component to your business is really interesting, too, because you're not only looking to create the Fusion facility, but also create all the components and vertically integrate everything to make this facility.

13:21What is the single most difficult thing to manufacture now in the supply chain for Fusion? We look at it from the goal. And the goal is global scale deployed power and how fast can we get there? And so then you look at all the components of what do I need to build myself? What do I need to vertically integrate to get there faster? And so we focus on a couple of the core technologies, capacitors, the way we store energy, store energy, we deliver it to the fusion process and we directly recover it at high efficiency from the fusion process. And so that's one of the key components that we can manufacture ourselves.

13:57And that's a tool that it's stored. And so you want to generate the energy, you have to be able to store it somewhere. This is a tool to be able to hold it there, to start with it, to deliver it to fusion, recover it and then hold it. And we do all of those on our systems today. And so that's that's a really key part to it, as well as the power electronics. When we look at the cost of a fusion power plant using our approach and what we've been able to build and demonstrate, about half of the cost of the power plant is in the electronics themselves. And so we want to be able to attack that cost, lower that cost, increase our ability to manufacture that so we can go faster.

14:33And so that's really the key, because at the end of the day, we have to go deploy these in the world as fast as possible and as low as cost to the consumers we can get to. Last question for you, David. I mean, you've been working on fusion pretty much your whole life. If you weren't working on fusion, what would you be working on? Yeah, you look at the challenges in the world, and energy right now is one of those biggest challenges. So it would be solving the energy problem one way or another. And so I'd be focused on it. I think fusion is the best way to do that. It's the way the universe makes energy, and we should be doing that here on Earth.

15:06Right. Okay. Well, David, I want to thank you for coming on. Congrats on the funding round. That is David Kirtley, CEO and founder of Helion Energy here on TI TV. Cybersecurity company Netscope reported quarterly results. Revenue grew 28%, a deceleration from last quarter. The company is expecting up to 26 % growth in the current quarter. Shares were down about 20 % early this morning, although the company is one of a few early names to work with Anthropics' mythos model. I want to bring on founder and CEO Sanjay Barry to talk through this moment for the company. Sanjay, welcome back to the show.

15:41It's great to have you here. Yeah, great to be here. Okay, so help me understand this. You know, cybersecurity companies are talking a lot about the opportunity that AI is for its business. And yet, it looks like the top line decelerated this quarter for you guys. Walk us through the results. Yeah, so first of all, we're one of the fastest growing public cyber companies out there. We grew ARR by 29%. We grew new logos by close to 60 % year over year. And we have record gross retention, right? So our customers are staying and growing with us. And for us, what we look at ultimately is, wait, what is that impact we're having on customers?

16:19And we've never been more core to what they need to do in the world of AI. I can't get into a conversation without them talking about AI. And our AI security pipeline is growing at the fastest pace I've ever seen any product that we've ever had. And so we're obviously very energized about where the company's going. So, I mean, I'm just asking, why wouldn't Topline be accelerating then if AI is such a tailwind here? Yeah, look, the reality is for us, we're in the midst of a very large sales expansion. We win over 80 % of the time when we get to POC, and our whole focus is more feet on the street and awareness.

16:56And so about half, you know, roughly half of our sales reps, they're being ramped right now. And so - You're training them. We're training them. We're bringing them on. We're hiring them. I just came out of my new go-to-market, new hire session. I walked in and I went, okay, this auditorium is not big enough. So, and so for us, we know the opportunity ahead of us. We got a TAM that's in the hundred billions of dollars. We're the market leader in the analyst reports. We do have one of the highest gross retention rates. And so we know this is our time. And so we're investing in sales and marketing to go after it.

17:29Traditionally, you may or may not know, but nearly half the company was R &D. And so for us, get more at bats. We have an amazing batting average. So when you talk about selling the AI-infused products, and I did want to ask you about the sales cycle and your conversations with customers, and I take your point, you're training your sales team, so those conversations are still about to happen, if not happening already. But what is the net impact of AI on the sales cycle? Is it making it longer because companies are taking more time to understand what it is they have to buy? Or is it making the sales cycle shorter because companies really just want to take anything?

18:17Yeah, it's a great question. The reality is, look, you can't get into a conversation without them asking about, hey, how can you help me enable secure AI? Right. But they ask about it, but what about buying? Yeah, exactly. But here's the thing. They don't have a textbook. They don't have a, here's how I do it. So they're learning. Enterprises are in the stage of realizing that AI moved farther than security did. So security is catching up. And so the way I look at it is they are, we baited four new products last quarter and we just released them. Our beta customers converted, right? They became AI security customers of ours.

18:52But what I do foresee more is in the second half of the year is where customers will really start realizing, okay, this is how I do it. They'll start buying, implementing, and so on. And so I think it's just a lag that always happens with security of new technology adoption. And so I really do see it happening in the second half. Although the fiscal year guidance that you gave, if I'm not mistaken, that too was not an acceleration. That remained a deceleration, right? Yeah, we beat our guidance and we did raise our guidance as well, not just by our beat, but we raised it by more than our beat.

19:27And that showed confidence for us in what we see in our pipeline and in the AI security pipe as well. I want to ask you about Mythos. So you are among the companies who have had early access to Mythos. We've seen Project Glasswing expand. How are you using it? What are your initial reactions to it? Yeah, first of all, when you think about Mythos or you think about Daybreak, right, with OpenAI and beyond, they're really amazing frontier models that can help you find vulnerabilities. And that's how we use it. We have a harness where we take the different frontier models, as well as, to be blunt, the open weight models.

20:04And as part of our development cycle, we leverage them to find vulnerabilities, right, pre-release of any products. And so that's just a normal part of our pipeline of development now. Now, on the other hand, you also know that whether it's an open weight model or other models, attackers will get these models and they will leverage them. And the reality is what I tell customers is you have to assume vulnerability. You operate an enterprise, a company, a healthcare, a financial, manufacturing, retail, assume vulnerability. That's just the world we live in. And what you need to do is protect yourself in real time when those agents talk, when people get access to your data, protect your data.

20:44And that's the world we live in. And I think that's - Have you found a number of vulnerabilities in your own code base using Mythos? Yes. I mean, leveraging these products, you can all the way from development to pre-production. Yes, absolutely. They're great for finding vulnerabilities. And my view is if you're not, you're probably not building a broad enough harness to test your products. What about your own budget for these tools? We've reported the information that using Mythos is very expensive. Have you burned through your allotted budget for using Mythos? Has it been more expensive than you thought?

21:23lot? I mean, we overall have really focused on becoming more across the board, not just from R &D, but everywhere AI native. So we leverage what I call our smart efficient tokens. I'm not measuring my people on token maxing. I'm measuring them on outcomes and outputs, PRs for engineers, output for marketing. And so have I exceeded what I thought I would have a year ago? Of course, most people didn't budget for what they thought and what they're seeing now in terms of their token spend. And so you have to adjust. The reality is that most companies, including us, probably have less open headcount, right, for things like R &D.

22:01And they're building these smaller agile teams and leveraging AI and tokens, frontier models like Mythos, but also for development more. Do you think that Mythos should be released to the public at some point? it's it's a great question um and um i go back and forth on this i do think over time that a broad swath of all companies should have access to it uh versus um you know a select set that's just my belief right i believe why why um i believe in the you know over time the democratization of uh these tools like you have some companies who will have access to them some who won't right and um the determination of you know what is critical infrastructure um will be different right depending on the could be the nation could be the eyes of the beholder and so on and so i do over time uh get see that having said that also some of these models you don't want to get them into the hands of the wrong people and so it's a tough you know trade-off and it's definitely a tough situation i think anthropics been managing it very well our collaboration has been amazing they're super responsive um and uh so we're really kind of impressed with how does how does mythos compared to open ai's daybreak program we haven't talked too much about daybreak on the show yet is that open ai's uh answer to mythos what is that program yeah gbt 5.5 um the daybreak program uh you can think of it as that that's that's the summary right it is and the way we think about it is you want to leverage as many of the models as you can uh just like in security defense in depth, right?

23:42You don't rely on one platform for all security and networking. Same thing. You don't want to rely necessarily on one model as you find vulnerabilities. And so which of the two models have you found to be more effective in spotting vulnerabilities in your code base? The reality is that we have found both to be very effective. Which is more? which is more that's a good question i'd have to i mean the reality is look there's a new frontier model every four to six weeks that's the reality and so what you have i think is this you know constant like one gets ahead another tries to catch up and and so um at any one point um you know one can be and the other can be at another point i i just want to ask you a little bit more about mythos here.

24:30So look, I don't know if and when it gets released to the public. It's the big question, as you say, if it gets into the bad people's hands, then that's a problem. But do you anticipate that, what do you think the business impact will be on the release of mythos broadly to your business? Is it that you'll just see more customers saying we need more protection, or Or do you see a way to incorporate Mythos into your own products and have a business impact there? Yeah, it's a good question. So if you look at what we do, we're not a code scanner, right, or a vulnerability scanner. We live on the opposite side of the spectrum of security.

25:13We basically live in the world where you assume there's vulnerabilities. And what we do is we are the highway for all your traffic. So when an agent talks to a cloud app or gets data, makes a tool call, when a user prompts something, right, we look at every prompt. We look at every tool call. We make sure, is that agent supposed to do that? Are they supposed to have access to that data? Is that user's prompt really appropriate? And does this response have malware? And so for us, we're the real-time governance. And as a result, with Mythos and with the explosion of AI agents, what we're seeing is we're processing trillions and trillions of AI transactions.

25:53And we're guardrailing them, protecting them, and ultimately enabling AI to be used safely without people shutting it down. And so that's kind of the way we see it. And Mythos for us obviously helps internally. I think customers view it as, wait a minute, I need to drive real-time protections because we have to assume attackers, even if they don't have Mythos, have open weight models, which can find vulnerabilities faster than they could have ever have found before. And so attackers do have new weapons as well. Great. Well, Sanjay, I want to thank you for coming on. That is Sanjay Barry, the CEO and founder of Netscope here on TI-TV.

26:36Snowflake is hosting its annual summit this week in San Francisco. It is putting a particular focus on its coding agent and co-work tools. I want to bring on Anahita Tafizi, Chief Data and AI Officer at Snowflake to share more with us about the company's strategy. Anahita, welcome to TI-TV. It's great to have you here. Good morning, Akash. Thank you for having me. I'm excited to join you here live from the Snowflake Summit. This is day four of our summit. We've had more than 20 ,000 people joining us across our customers, partners. I've had a lot of customer conversation. There's a lot of excitement about the announcements we've done this week, particularly, as you said, around our AI capabilities with Snowflake Cocoa and Snowflake Cowork.

27:17Right. Well, tell us what those products actually are. Yes, absolutely. So we announced a few new capabilities in Snowflake Cowork, our new personal assistant, personal agent, if you will, for knowledge worker. We have new capabilities like deep research, like artifacts, skill catalog, which really moves from reactive Q &A to truly your personal agent, if you will, for knowledge workers to get work done. We also announced new capabilities in our coding agents, Snowflake, Cocoa. We have new automation capabilities, cloud agents, ability to connect, fit the coding agents wherever you are across desktop, across mobile, even within Excel or cloud code.