Tubi CEO on Streaming Landscape, Google’s AI Strategy, Starcloud’s Space GPUs | Nov 3, 2025

0:13Welcome, everyone, to the Informations TI TV. My name is Akash Pasricha. It is Monday, November 3rd. We have an exciting show lined up for you today. AWS and OpenAI are announcing a$38 billion deal, which will see OpenAI run workloads on AWS, while AWS will provide OpenAI with access to NVIDIA chips. We're going to break that down with Aaron Holmes, who covers Microsoft and all things enterprise software. We're also going to talk about a new story that he has out today about the challenges that companies are seeing with AI agents. Then I'm talking with the co-founder and CEO of StarCloud, the startup that just launched a satellite into outer space carrying an NVIDIA GPU.

0:54We'll also hear from a Google executive and the vice chairman of ServiceNow about each of those companies' AI playbooks. And we're going to wrap up the show with Tubi's CEO on how the streaming and media landscape is changing and on that platform reaching profitability. It is an exciting show. We've got a lot to get to, so let's get right on into it. AWS and OpenAI have signed a big multi-year deal together that will see OpenAI run some of its workloads on AWS infrastructure. OpenAI will get access to massive compute power via AWS's NVIDIA chips. Joining me now to break down the news is our Microsoft and Enterprise software reporter, Aaron Holmes.

1:31Aaron, it's good to see you. Happy Monday. Happy Monday. Happy to be here. So it's another week and another big deal that OpenAI is signing with a major cloud player. What did you make of the news this morning? Yeah, so I mean, this is definitely good news for AWS, which has been growing at a slower rate than Google Cloud and Microsoft Azure in the last few quarters. And, you know, there was a perception that AWS might have been missing out on some of this, you know, AI cloud spending bonanza. but we've seen OpenAI is really eager to sign basically any compute deal it can get its hands on. And this is definitely a smaller deal than what OpenAI has committed to spend on Oracle or on Microsoft in the coming years, but it does get AWS back into this conversation and basically will help boost their AI cloud business in the coming years.

2:21One of the things that I noticed in the announcement, there was no discussion of Tranium, which is the chip that AWS is working on. This, of course, is also the chip that last week executives couldn't stop talking about on the earnings call. That was the same call where AWS showed that they have actually been able to accelerate their growth rate. What did you make of the fact that Tranium was not in the conversation? I feel like maybe this comes up later on. They just weren't ready to announce it yet. I don't know what to make of it. Yeah, so, I mean, this announcement, you're right. It specifically said OpenAI will be using NVIDIA chips in AWS data centers.

2:56I think it basically just shows that NVIDIA is still state-of-the-art when it comes to GPUs for things like training and inference. And yeah, I mean, Amazon has been trying really hard to push its own in-house chips with, you know, mixed results so far. I think that the fact that OpenAI is opting to stick with NVIDIA chips, you know, maybe speaks to some of the challenges that AWS has seen convincing these big labs to use Tranium at scale. Right. Well, I want to turn to a story that you published this morning about how enterprises are dealing with AI agents. You looked at how companies are racing to integrate agents into workflows, but the reality is that many AI agents are underperforming in real life on the job.

3:37Why is that? What's going wrong here? Yeah, so we're in this really interesting place where, you know, AI is undeniably changing a lot of how work is done. And a lot of people are getting personal productivity benefits from things like chatbots or AI coding tools. But I've spoken to a lot of customers who have tried to deploy autonomous AI agents to take over things like customer service. And, you know, they're still a little bit concerned by what they're seeing. You know, for example, I spoke to somebody at Bosch Power Tools who wanted to build an AI agent to explain the user manual of different power tools, but they haven't been able to put that into production just because it kept hallucinating wrong information that could result in a customer getting hurt.

4:18So as a result, we're seeing AI labs actually devote more resources to helping these customers configure agents because I think it can be harder than they thought originally. And so this is the idea where the labs will actually give people talent, essentially saying, hey, take our people, let us show you how to configure these agents. I imagine that that has to take a hit on profitability for these companies, no? Yeah, I mean, I think we've seen OpenAI do this a lot over the past year, and now we're seeing Anthropic, but also, you know, cloud firms like AWS start to step up their efforts to help customers configure agents.

4:55And I mean, you know, Anthropic and OpenAI for now are not really concerned with profitability. So they can pour a lot of resources into making sure that these pilots actually get off the ground and work effectively. And I think that's sort of the strategy that we're seeing play out right now. What was the consensus from the people you talked to around whether or not these agents will make work completely obsolete, whether or not they will be completely autonomous in some ways, or whether or not they will still need to have a human in the loop? You know, most of the software executives or CEOs, founders I spoke to now feel like it's going to be at least a couple of years before AI agents can truly automate jobs, which is interesting because I think earlier this year, you know, you had folks like OpenAI's, you know, chief product officer saying it would be the year of agents.

5:43But I think, you know, companies are starting to run into the harsh reality that a lot of these agents are not quite ready to work completely autonomously without having a human essentially check their work. Right. Well, it's a story that is always evolving. And so, Aaron, I want to thank you for coming on. That is Aaron Holmes from our newsroom here at The Information. We've talked before on this show about the ambitions of some companies to build data centers in outer space. This weekend, one company at the center of that ambition, StarCloud, took early steps to making that dream a reality. On Sunday, the satellite company launched NVIDIA's H100 GPU onboard a SpaceX rocket and sent it into outer space.

6:21I want to bring on Philip Johnson, co-founder and CEO of StarCloud, to tell us more about the launch and what's next for the company. Philip, happy Monday to you. It's a great Monday for all things StarCloud, I imagine. Thanks so much for having me on. Yeah, no, it's been an awesome weekend. Awesome weekend. So tell us about the launch. Just walk us through it. What time did it happen? Yeah, so launch was 1 a.m. on November 2nd, so Sunday, very early morning. Okay. We had a bunch of us down at Cape Canaveral. I'm still down here at Cape Canaveral right now to watch the launch. So yeah, separated from the spacecraft about an hour after launch, and then we got the first telemetry back about 12 hours after that.

7:02First telemetry back. What would translate that for us, for those of us who don't know Space Speak? So that's essentially when we pass over a ground station, it allows us to form a radio link. And then that will tell us, okay, is the satellite healthy? Have the solar panels deployed? Is the battery charged? Are we receiving data as we should be? So yeah, it's a very nerve wracking moment up until that point. I think about half of all first time startup satellites don't ever make connect, like make this first contact. So it's a pretty nerve wracking moment. And the company is, what, less than two years old right now?

7:37Yeah, we started early last year, so 21 months. Okay. What is the GPU? First of all, how many GPUs did you launch? So it has five, but the two that are interesting is one H100 from NVIDIA and one A6000 or so from NVIDIA. Okay. And what is the chip doing up there now? Well, right now, we still haven't commissioned it, so it will take about a month before we lower altitude to the point where we're going to be commissioning the chip. But what it will be doing is running high-powered inference on imagery from other satellites, as well as being the first to do things like training a model in space, first to do fine-tuning of a model in space.

8:15We're going to run a version of Gemini from Google Cloud. So it's mainly just to prove, this is really a demonstrator to prove that you can run high-powered terrestrial data centigrade GPUs like H100s in space. So you're actually using it to power some of the AI workloads from the other satellites that are up there right now. How do you deal with some of these issues with having satellites in space at all? Stuff like space debris, temperature control is obviously an issue. You got to figure out how to keep the thing cold, even though it is outer space. People think of it as cold. It's actually a bit of a more complicated issue than that.

8:51Yeah, that's true. So for space debris and micrometeorites and things like this, you essentially, I mean, for us, we're going to be avoiding the most congested parts of space, which are where all of the other satellites flying between about 400 to 800 kilometers altitude so the first few are flying actually very low uh 380 kilometers it's kind of it's what's called very low earth orbit um you have a self-cleaning property of that altitude because they have low levels of drag from the upper atmosphere and that kind of de-orbits anything within a few months anyway so it's very very clean those orbits and then for the later versions of the satellite we'll be flying quite a bit higher around 1200 kilometers altitude and that's so that we're always in the sun and also not too too many people want to fly there.

9:30So, so, so not too much, but you always have the risk of micrometeorites coming in from outside orbit. And so for that, we need shielding on the most, most sensitive parts. So for example, the chips, and then we allow a certain degradation over time for other parts. So for example, the solar panels, you just allow parts to pass through them. Right. Right. Talk to me about the business. Sorry. Go ahead. There was one part on cooling. Cooling, right. Yeah, we need very large deployable radiators. So that's the core IP that we're developing at Star Cloud is these enormous low-cost and low-mass deployable radiators is to get rid of this heat.

10:09Right. Talk to me about the business model here. Are you guys selling compute to customers? Obviously, this is a very small workload up there right now, but do you have companies to you saying, hey, we will buy the compute workload when you get it to outer space? what does the business model look like for star cloud as a company who's paying you yeah for sure and so for the second satellite launching in october next year that's about a hundred times the power generation of the first one ten times the compute and we are going to be selling mainly to dod we are selling my dod um customers so you might have come across this golden dome that trump is um orchestrating it's basically a missile defense um system so um i can't speak too much about it, but for example, you can imagine it would be useful to have high-powered compute, apologies, high-powered inference on orbit in order to be able to do things like running imagery, running inference on the imagery you're collecting.

11:08For example, if you were to want to know, has a Mishab been launched in this location? You don't want to have to wait for a ground station, which is what they currently do. You want to ship that data immediately in space to somebody like us, and we can then provide an insight in real time. I've got to tell you, Phil, we've had hundreds of guests on the show. We're still a new show. We've had hundreds of guests on. We've never had anyone sneeze. Sneezing on air is a friend. Even I haven't done it. I'm sneezing all the time. So bless you, I should say. Bless you. Love it. You had it here fast. Okay.

11:42So you're working with DOD. And what do these contracts look like? Are these, you know, I don't even know how pricing would work for something like this. So in the end state, we'll be pricing ourselves a bit similar, you know, very similar to how cloud providers would for, you know, people like Core, even Lambda, how they would sell compute to the hyperscalers. So it's basically dollars per minute on GPU time. That's the end state. The DoD has a lot of programs to fund R &D. And so we'll certainly be going up to some of those opportunities. And that's really more like a milestone-based opportunity.

12:21Right. Well, it's a big accomplishment, and it's certainly one of the more fascinating stories we've had on the show. Philip, I want to thank you for coming on. That is Philip Johnston, CEO of StarCloud. Last week, Google Cloud put up some impressive numbers as Alphabet reported quarterly results. The division accelerated its growth rate to 34%, and operating profit jumped dramatically. Of course, Alphabet is also upping its cap-back significantly to stay competitive with its AI footprint. I want to bring on Oliver Parker, Vice President, Global Generative AI Go-To-Market at Google Cloud to talk about this moment for his group.

12:56Oliver, welcome to the show. It's great to have you. Good morning, Cash. Good to meet you. So this is kind of an exciting week for Google Cloud here. You're coming off some big earnings, obviously. Just can you give us a little bit of an overview here? How do you structure the portfolio of work that you oversee? Because you've got a really long title, and sometimes we just want to translate for people what you even oversee. Sure. So I oversee really what our AI go-to-market is as it relates to the cloud division. So obviously a lot of partnership across other parts of the company and really what we're doing directly with our models, really the platforms for developers, as well as new areas that we're focused on.

13:33We made a launch a couple of weeks ago with Gemini Enterprise. So in the full stack, which has obviously been a really big differentiator for us and where many of us, many of the customers are actually partnering with us. So yeah, that's my responsibility. really what we're doing in the cloud division around AI. And so we're going to come back to the company strategy in a minute, but we saw the news this morning, OpenAI and AWS signing a big deal together. Do you see that deal as an opportunity for Google Cloud or a threat to Google Cloud? Well, look, I think I probably wouldn't want to comment on what AWS and OpenAI are doing, obviously, as you can imagine, but I think we've seen sort of more broadly in the market and some announcements that OpenAI have been making with ourselves as well, just in terms of the compute capacity that they're seeking.

14:14So, again, I think this is just part of a broader industry requirement around compute and power and really people trying to figure out distribution of their models. And obviously, we have a unique position there where we obviously support distribution of many other people's models as well as obviously our own models and obviously being heavily invested for a long period of time in our own silicon with the TPU platform, too. And how does the deal today compare to the work that you guys are doing with OpenAI? We didn't share the financials on our partnership with OpenAI. Obviously, we did share financials, and you've probably seen that around some of the work that we're doing with Anthropic a couple of weeks ago.

14:50But I think if you sort of look broadly at the industry, what OpenAI are doing, just in terms of sort of the compute requirements that they need, I think that's been pretty open in the industry right now. And I think there's lots of stuff that Satya and Sam have been talking about too. But maybe it'd be good to talk a little bit more about what we're doing, and hopefully that's of interest to you. So I want to talk about that. So TPUs were a big point of discussion on the earnings results last week. And one of the questions I have for you is this sort of sets up Google nicely for sort of this end-to-end playbook as it relates to AI and selling workloads to enterprises.

15:25You know, when enterprises are coming to you and to Google Cloud broadly, are they asking you specifically saying, hey, we want this proportion of our workload to be run on TPUs, not on GPUs? Is that something that you have to sort of convince them on? What does the dynamic of that discussion look like? I think it really depends on the kind of customer that you're talking about, right? So you have obviously some of the very progressive digital companies, especially some of the startups that are looking to build lower down the stack. And then you have more traditional enterprises that are really looking for us to be able to manage a lot of that complexity.

16:00And I think that's honestly the uniqueness of what we offer here. which is really we have partnerships in all areas of our portfolio. And again, depending on the company, whether they want to really work closer to the silicon and to the hardware and to our accelerator platforms, right the way through to developers that are looking to leverage, you know, platforms like Vertex for choosing our models as well as anyone else's models, including companies like Anthropic, right the way through to a launch that we made a few weeks ago, which Sundar referenced actually on our earnings last week, really around things like Gemini Enterprise, which we sort of see as sort of the front door for AI at a business level.

16:32And again, really depending on where the clients are at and what they're building and how they want to have a partnership, we feel very good about being able to offer them choice and optionality at all areas of that stack. And as it relates to the TPU and having that as part of the offering, how do you make the case to customers to use that as opposed to something like NVIDIA's GPUs? Well, I think if you sort of look at what we've been very public about as it relates to TPUs and again, the announcement with Anthropic that was shared a couple of weeks ago in terms of them sort of having access up to a million TPUs, also look at what we do with Gemini from a training as well as an inference standpoint.

17:10So a lot of the serving and inference for Gemini is running through our TPU platform and sort of outside of sort of that optionality or that option of choice. I think what's interesting is, is that it puts us in ability to be able to provide great performance, great costs, and great latency requirements around those models. So we're seeing a significant acceleration of our Gemini family. And a big part of that is really due to that actually running from an inference standpoint, as well as being trained on our TPU platform, as well as obviously GPUs. When you think about all of the AI work that is happening at Google Cloud right now, how much of that is dependent on or based on usage-based pricing right now?

17:49Well, we have different monetization, different commercial packages for people that are using tokens right the way through to people that are using products that consume tokens, you know, to really alleviate budgets and how they sort of operate. So we're providing flexibility actually across multiple different products, really depending on what clients want. And that's a space that I think we see continually evolving. And, you know, we're going to look to sort of make sure that we are leading in that space to be able to find optionality and flexibility to many customers. And when you say flexibility, how do you see that shift changing over time, that mix between outcome-based pricing, usage-based pricing, and then more fixed pricing models?

18:25Well, I think, you know, obviously the cost of tokens has been a sort of a well-tried and tested approach, you know, with our APIs, as have many of the other frontier providers through sort of other hyperscalers. But we also now have products that are really sort of very IaaS-centric, which are sort of looking more at the per-user-per-month sort of kind of approach that I think, you know, the traditional SaaS industries have for a while. But we also then have certain outcome-based capabilities around things like call deflection when you start thinking about AI for customer service. So I think we're going to continue to experiment and actually sort of make sure that we lead in that category so that really the value of AI becomes very apparent in terms of how people buy.

19:00And those conversations are ongoing, I would say, and we've actually done really well here because I think we've proven a lot of our AI is actually now starting to make a difference across many of the enterprises that we're working with. Right. Well, Oliver, I want to thank you for joining us. That was Oliver Parker from Google Cloud. Okay, ServiceNow has in many ways been seen as a bit of an AI darling in the enterprise software world. The company has consistently grown its top line in or around 20 % every quarter for the past few years. It also boasts some very strong free cash flow margins, and yet it is one of the biggest SaaS companies in town, which means it is also having to adapt quickly to the AI era.

19:38I want to bring on Nick Zitzen, Vice Chairman at ServiceNow, to talk about where his team is taking the company. Nick, welcome to the show. It's great to have you. Well, Akash, great to see you. Frequent viewer, first-time caller, as they say. Well, we look forward to having you on more and more as the years go on here. I want to talk about the general vibe right now in enterprise software. You guys obviously reported earnings last week, and I think we saw the results there. But broadly speaking, I just want to take a step back here. I mean, look, the NASDAQ is up this year. ServiceNow stock is down for the year.

20:11All enterprise software stocks are down. And so I just want to take your temperature a bit. What do you think investors are reacting to here? Well, I think it's perfectly normal that investors try to figure out exactly what the shape of the industry is going to be moving forward. There's a lot of noise. And frankly, Akash, you talk to customers and they'll tell you the same thing. Everybody has an AI story. So I think it's pretty natural that investors are trying to take a bit more of a wait-and-see approach. What I will say for ServiceNow, and Bill McDermott has said, the enterprise AI neighborhood is a different and new neighborhood and much more advantaged for a company like ServiceNow than the traditional SaaS neighborhood.

20:54We're very optimistic that we can continue to curate the kinds of overperformance that you saw in our recent earnings. Now, you put up some big numbers,$500 million in ACV annual contract value by the end of this year and$1 billion by the end of next year. What are the challenges that you're seeing to enterprise adoption right now? Because even despite those numbers, we've been writing here at The Information about it is a slow and steady process for enterprises. What are the biggest hurdles enterprises have to overcome right now with AI? I think it's what I described earlier, Akash. You have every technology leader in every enterprise under a siege of incoming because companies, frankly, that aren't sort of natural AI players or don't have a soundbite driven, who do have a soundbite driven AI strategy and not a content driven AI strategy.

21:47So I think people are trying to figure it out. I think they're trying to figure out if I'm going to build a reference architecture for the next five to 10 years of my organization, and it's going to look different than the reference architecture from the last 10 years. How is it different? And what players belong in the new one that perhaps were in the old one and which ones should be replaced? So these are complex conversations. I think for a company like ServiceNow that has been at the core of the technology estate, helping to govern all the different systems and helping to integrate these enterprises, those exact challenges are only going to be worse in the AI era.

22:25So it's, again, a good sign for us and our strategic relevance for most of our customers, if not for all. What about the people issue to all this? Because we had Vinod Khosla on the show a couple weeks ago, and he talked about the idea that the software is one side of the problem. If you don't have the right people on the ground, which, in his opinion, was the case for a lot of enterprises right now, they don't know how to use the software. They don't know how to make use of it. That's one perspective. The other perspective we've been hearing from people is, well, I can't just replace my entire IT team of hundreds of people.

22:56Where do you stand on that spectrum of where the issue lies, and how are you approaching that right now? Yeah, I think it's more of a mindset challenge than, frankly, a skills challenge. It's a little bit of both, right? There's this ongoing debate, is AI revolutionary? It's going to change everything, how organizations run, or is it evolutionary, meaning you're going to see incremental improvements in some of these sort of legacy processes? I would tell you that inflection point, is it revolutionary or evolutionary? depends in large part on the mindset of people. What do we want it to do? And look, if you have a change-oriented environment and a culture that says, we're gonna embrace the possibility that this is a new business model, a new way for our business to serve our customers, a new way for us to make money, it's gonna be revolutionary.

23:46And there are a lot of organizations that are taking that approach. I would tell you, you know, the notion that you can go out shopping for a new team that's better and more AI native than the one you have. I don't subscribe to that. I think we at ServiceNow believe that you're going to see an increase in requirements for technical workers, for people who understand software. So I don't think this is a scenario where we can let AI become a code word for layoffs because we do that. I don't know how we expect people to embrace the unbelievable potential of what enterprise AI should really be. let's talk about the the public sector business that service now has you of course have a background in government too so i i know that this is a customer set that is very near and dear to your heart i i want to take your temperature here on on doge because earlier in the year doge made a lot of noise it was kind of the headline that was dominating the the the tech news altogether was you know what softwares are is the government cutting back on what software is the government using to replace point solutions, for example, bringing them all under one hood.

24:54ServiceNow, I recall at the start of the year, the discussion was, hey, they're not cutting back on us. We're actually seeing it as an opportunity for us. We've started to hear less about Doge altogether. And of course, it's a very different conversation now that we're talking in the middle of a government shutdown. But if we just put the shutdown aside for a minute, what is the temperature that you're seeing from the government in terms of their software spend? And what's going on with doge altogether. We haven't heard about it in a long time. Yeah, I mean, it's increasingly hard to take politics out of these kinds of conversations.

25:25But if you take the politics out of the conversation, I don't think anybody would deny that the U.S. government needs to adapt and needs to modernize. It's frankly unsustainable for government to run the way that it has. We don't have unlimited power to just continue to raise taxes or to cut programs. So we have to find more efficient ways for these agencies to run. I don't think that's a controversial statement. And I think many of the people at Kashi who work in government are perfectly up for that conversation. They want to be part of a change. They want to be part of a change that's positive, that provides a better citizen service.

26:01So when you look at a platform like ServiceNow, which is only in the early days of its deployment in government, I think the government customers that we talk to believe that we are 100 % part of the AI future. They believe that, you know, But lack of integration is a big challenge for them. I think they believe homegrown solutions that maybe were built 10, 20 years ago that people aren't using. I think that's part of the problem they have. So whether you call it Doge or whether you call it modernization or whatever you call it, like government wants to change. And for us, we see that as a massive tailwind for what our platform can do.