Inside Amazon’s Potential $50B OpenAI Investment, Nvidia’s Impressive Earnings & Stock Fall

16:46Cory Weinberg:And I think that's most obviously reflected in the top-line growth rate at the two. Like you mentioned, Nakash, Snowflake is growing around 30 % year-over-year. Salesforce is growing closer to 12 % as of this quarter, which was actually their fastest growth rate in the last several years.

17:01Anita Ramaswamy:I want to ask you about how AI is affecting both these businesses differently. You kind of hinted at it here, but I'm going back to some comments that Martin Pierce wrote in his briefing newsletter last night. He wrote that AgentForce's ARR, for instance, is just about 1.7 % of Salesforce's total fiscal 2027 projected revenue of about$46 billion. So, I mean, the story here, Agent Force is growing, but it's still a small part. Snowflake, like you said, it's an infrastructure layer. Maybe they're seeing a little bit more of a benefit. I mean, is the story different here?

17:39Cory Weinberg:I mean, there's some difference, but even with Snowflake, it's hard to tell exactly how much of their acceleration and revenue growth has been as a result of AI. They've disclosed the number of customers that they're seeing, and that seems to be growing really nicely. They also said this quarter at Snowflake that they signed their largest deal ever, which was around$400 million worth. And they attributed that at least partly to their AI suite. And so it seems like Snowflake is benefiting from selling AI products to customers. I thought one interesting point on Snowflake is that their free cash flow margin has actually gone from 43 % a year ago to 61 % in the latest quarter.

18:15Cory Weinberg:So that kind of shows that even as they invest in AI, they have been able to continue to generate cash at a faster rate. I think with both Snowflake and Salesforce, however, they're huge companies. And like you pointed out, the AI product, especially at Salesforce, is a pretty small portion of their overall revenue. So even though it's growing like over 100 % year over year, we still don't really know how much of a meaningful catalyst that's going to be to top line growth going forward. And I think another important point to note is that the acquisition of Informatica for Salesforce, which closed in November 2025, was a meaningful accelerant and driver of their growth.

18:51Cory Weinberg:So I think it remains to be seen, Akash, whether the organic growth for both of these companies can continue. The one negative point on Snowflake that some investors were considering is that they gave their revenue guidance for their fiscal 2027, which their fiscal year ends in January. And Snowflake essentially implied that their revenue growth would decelerate to around 27%. Now, that's still a really fast growth rate for their product revenue, which is the majority of their top line. But it's slower than it was in previous quarters. And so that could be a sign that there is something more fundamental about AI that is challenging both of these businesses in a meaningful way.

19:29Anita Ramaswamy:Great. Anita, I want to pivot and talk about a story that you published today on other big tech companies, bigger tech companies even, than the Salesforce's and the Snowflakes. You wrote about the hyperscalers, even the metas of the world. it's funny we don't call meta a hyperscaler and yet i still i still seem to conflate it with it i know it's people are going to say you know don't conflate the two and i'm not suggesting it but it's kind of hard to talk not about that in this conversation anyway you wrote about these big tech companies and the debt that they're taking on and how it affects their credit profile what were the

20:06Cory Weinberg:questions that you went into this story with so i went into this story wondering about how all of these companies have issued debt in recent months. We saw two debt raises in November. We saw a debt raise more recently than that. Specifically, I'm talking about Alphabet, Amazon, and Meta, which I guess one should argue if they really invest in Meta Compute, maybe they will be a hyperscaler someday. But that's another conversation for another day. The question that I was really trying to answer with this piece is, what will it take for these companies to actually see an increase in their cost of capital?

20:40Cory Weinberg:They're issuing more and more debt. It seems like they're continuing to plow more and more CapEx into their AI investments. And increasingly, that's being funded in the debt markets, not just the equity markets. And so I wanted to look at what the potential impact could be of, you know, could they get downgraded by the credit rating agencies? Are they even anywhere close to reaching that point? And are there any other factors that could meaningfully increase their cost of capital for the long term?

21:05Anita Ramaswamy:And so what did you ultimately find?

21:07Cory Weinberg:So the good news for investors in these big tech companies is that it looks like a credit downgrade is pretty far off. The threshold for these companies is around one times debt to EBITDA. And, you know, all of them by the end of 2026, at least when I talk to S &P analysts, are projected to be closer to like, you know, below 0.5 times, let's say, 0.2, 0.3 around there. And so they have a lot more capacity. I mean, each of these companies can take on about$200 billion in incremental debt as long as they generally meet the projected EBITDA that analysts expect them to make.

21:42Anita Ramaswamy:And when people are calculating these credit ratings, is it only the amount of debt that they have on their balance sheet that they take into consideration? What are the other factors that they consider as they're coming up with these ratings?

21:57Cory Weinberg:Yeah, so this is actually something interesting that I learned as I was reporting out this piece was talking to these credit analysts. they're not only taking into account the credit metrics, like, you know, can the cash flow cover the debt and what that is going to look like going forward, but also some strategic considerations in terms of the business. So if you take all three of these companies, Meta actually has the lowest credit rating of all of them. And that is largely a function, I've been told by these credit analysts, of the fact that their business is really dependent on advertising revenue, whereas the other two companies are better diversified.

22:26Cory Weinberg:And so that sort of comes with a little bit more of incremental risk. In some of my conversations, these analysts were saying that they have this downgrade threshold of one times leverage, but that they could even potentially revisit that if these companies show depressed cash flows in the future. And so of all three of the companies, I would say Meta is probably, you know, looks the riskiest right now just based on their credit rating. And that's not just a function of their current cash flow and debt situation. It is very much a function of what the future of their business could look like and how analysts are perceiving that.

22:58Anita Ramaswamy:Now, the other piece that you wrote about that I found interesting is you talked to these analysts about how much demand there is for this debt and the fact that these companies are, I mean, these are blue chip tech companies. And I found myself thinking about, well, clearly the demand for the debt comes from the credit rating itself. Although I wondered if maybe there was a bit of a dynamic in terms of, you know, maybe it doesn't matter what the credit rating is if there's demand for this company at all, then the companies can just keep issuing it.

23:28Cory Weinberg:Yeah, I think the credit rating has some bearing on the demand for the debt, but it's more like correlated than directly causation linked. And that's because, you know, there are other factors, like investors are looking at the credit rating as one measure of the financial health of these businesses, but they're also looking at a number of other factors, you know, they're about, it's similar to like how you would evaluate an equity analyst report, perhaps on a company. And I think that supply and demand dynamics are a really big factor here. Like if all these companies continue to issue more and more debt and they eventually flood the market with supply, right now we're seeing that there's a lot of demand for this hyperscaler debt.

24:05Cory Weinberg:Like all these debt offerings have been oversubscribed. They've been able to raise the amount of money that they want to raise, but it's possible that if they flood the markets with so much debt that that picture changes. And there's also the possibility that investors end up seeing the underlying credit health of these businesses quite differently from how the credit ratings agencies are seeing them. So I think there's a lot of different ways that a potential cost of capital increase on the debt side could flow through to any of these three companies. But I think credit ratings are a really big, meaningful indicator that investors are going to be looking at.

24:36Great.

24:37Anita Ramaswamy:Anita, I want to thank you for coming on. That is Anita Ramaswamy, our financial analysis columnist here at The Information. Another defense company is raising another big funding round. Saronic is raising up to$1.5 billion at a$7.5 billion valuation in a funding round that Kleiner Perkins will lead. That is according to an exclusive report from my colleague Corey Weinberg, our Deputy Bureau Chief of Finance. I want to bring on Corey to talk more about this deal. Corey, welcome back to the show. It's great to have you here. Hey, Kyle. What is Saronic? I never heard of this company.

25:14Sri Muppidi:Yeah, Saronic, well, if you haven't spent a ton of time sort of in the trying to come onto the shores of the United States, then maybe you wouldn't have swimming in our waters here. No, I mean, Saronic is a naval ship company. They are quickly becoming one of the best funded startups in defense tech. and essentially what they do is that they build sort of warships and naval vessels and try to sell them to the U.S. Navy.

25:49Anita Ramaswamy:Okay, but you said in the story these are autonomous warships?

25:53Sri Muppidi:Yeah, they're unmanned generally. So Saronic is bundling sort of their autonomous software with the ships And, you know, that's sort of a big future priority for the U.S. Navy.

26:10Anita Ramaswamy:Okay. And so the U.S. government, are they the only customer for Cerronek right now or who are they selling to?

26:17Sri Muppidi:They're the only significant customer, to my knowledge. They, I think, aspire to also sell like a lot of sort of defense tech companies to U.S. allies. Sometimes that's easier said than done. You know, countries tend to want to sell or excuse me, to buy technology from homegrown companies. So that can be difficult, but they aim to sell to U.S. allies as well. And they also, you know, could have a commercial, you know, sort of could sell their software commercially at some point down the road. And they might need to. I mean, they have a big valuation now. It's potentially up to nine billion dollars.

27:00Anita Ramaswamy:And what do we know about the revenue profile for the company?

27:03Sri Muppidi:We reported in the story that they generated just over$200 million in revenue last year. So, you know, that's a, it's fair to say investors are expecting a hefty amount of growth in the coming years. And if they don't get it, then Saronic will look wildly overvalued. because, you know, that's a pretty significant multiple, at least looking at the previous year. But look, these are really big ships. They're going to sell them for – they're trying to sell them for a lot of money, and investors are betting that the Navy is going to buy them.

27:40Anita Ramaswamy:Okay. So, you know, one of the interesting parts that you had in your story is that Kleiner is leading this deal, and is Kleiner a big defense tech investor? Is this a practice that they're building out? I mean, is this a trend that we're seeing that VCs are starting to build out these practices? Because I can only imagine how much technical expertise must be required for an investor to understand any of this.

28:05Sri Muppidi:Yeah, it's really interesting, Akash. Kleiner Perkins hasn't done a major defense tech deal before. This will be out of their growth fund. You know, so it's designed for later stage companies. Kleiner Perkins has invested in flock safety, which obviously sells sort of license plate scanning cameras to police departments, but slightly different than defense tech. No, I was surprised. We had been hearing about this round for at least for weeks that Sironic was out there trying to drum up interest. And I personally expected either one of the VC firms that has really, you know, sort of put defense tech front and center of their whole message to kind of go out and try to lead this round, you know, like an Andreessen Horowitz or Founders Fund or ABC or something like that.

29:01Sri Muppidi:So I was a bit surprised when Kleiner sort of was the one that emerged as leading it. But, you know, look, I think from what I hear from VCs that are doing these types of big deals is like, why would I invest in, you know, the third AI, you know, coding startup right now, you know, when they might not exist tomorrow? you know sort of and I think folks are looking at companies like SpaceX that have built you know monopoly-esque businesses over a long period of time you know they see a hardware mode they see hey it's like expensive to build these companies but once you do it's pretty sticky so I think Zeronic is you know it makes sense why a firm like this would would bet on them but it certainly you know, I wouldn't be that surprised if in five years I'm back on the show talking about the, you know, sort of cautionary tale of Saronic as well.

30:10Anita Ramaswamy:But I mean, let's go back to sort of expertise. I am curious, the people you talk to, the people at these venture firms, I mean, are these, is it the same group of investors that were investing in, I'm not calling them low tech, but, you know, software companies of, five, six years ago, have they sort of expanded their focus area? Or do you have people who actually have backgrounds in the military or in building warships that are now coming on board to these investment companies to offer their expertise?

30:42Sri Muppidi:You know, everyone will say they've done really strong diligence, Akash, on the companies no matter what. Now, look, I think that's one of my major questions as more capital comes into this space is like do you as an investor do you know what you're buying um do you know how revenue is actually going to flow and whether this company is going to be able to sell their product you know across enough customers that it's going to to turn into a meaningful business um i mean look i think vc firms are increasingly tapping advisors who have been high up in in branches of the military or the government And those linkages are probably stronger than ever.

31:21Sri Muppidi:But yeah, I think the investors, particularly that have been doing Defendstack for a while, will look at some of these deals done more by generalists and say, I don't know about this one. So, I mean, that's something to watch going forward. It's not necessarily, you know, sort of explicit or directed just at Sironic. I think people have opinions on this company and some people are really bullish. Some people think, you know, the valuation is ahead of the traction. But, you know, look, I think VC firms are used to, you know, sort of following the herd to some degree and, you know, also accustomed to making bets that they can sit in front of their investment committee and defend.

32:10Sri Muppidi:And so, you know, I think this one might be like right on that edge.

32:15Anita Ramaswamy:well i i i love your your cautious uh analysis of a business like this cory because uh like you said it's uh it seems to be very much a sign of the times but at the end of the day who knows really

32:29Sri Muppidi:it's interesting i like writing about these companies more than you know the

32:33Anita Ramaswamy:the 10th coding startup well i i maybe i'm just sort of get your temperature on this i mean like where do you think this sits on the fence are you sort of looking at this being like you know i don't know about this one guys i mean i don't i mean i don't know and i haven't seen

32:47Sri Muppidi:you know the the the ships in action or like you know i haven't i have haven't talked to a uh uh naval customer yet or anything like that um it's unique i think andrel is kind of a company paving the way here in terms of, you know, companies that are capital intensive, burn a ton, and are trying to sell big, you know, sort of weapons and war machines that are quite expensive to manufacture and expensive to buy. So we'll see. I think it just like might take a while. I think with all these companies, I think if it's going to succeed, it's usually on a slightly longer time horizon than investors are used to.

33:34Great.

33:34Anita Ramaswamy:Well, Corey, I want to thank you for coming on. That is Corey Weinberg, our Deputy Bureau Chief of Finance, here at The Information. Robot data startup Encore is raising$60 million in new funding. That is according to a new exclusive story in the Information's AI Agenda newsletter. The problem of training robots with more high-quality data is one that is becoming increasingly important. I want to bring on the co-founders of Encord, Ulrich Stig Hansen and Eric Lando. Ulrich and Eric, welcome to the show. It's great to have you here. Thank you. Thanks for having us. You guys can have your own TV show.

34:09Anita Ramaswamy:Ulrich and Eric, it has a nice ring to it. What do you think? That's why we got together. For the TV show or for the robot day? Just for the names.

34:17Sri Muppidi:Just for the names. It's good name running.

34:20Anita Ramaswamy:Great. Okay. Well, I'm going to start with you, Ulrich. What is Encord? Tell us about the company.

34:27Matt Bryson:Well, NCAR is building AI native data infrastructure. So we work with more than 300 of the world's top AI teams that are building different types of AI applications, everything from autonomous cars to delivery drones to surveillance cameras and everything else in between. We generally work with companies that are building, again, AI systems for the physical world. And of course, a lot of the robotics companies that are now coming out in the Valley, which I think is a super exciting place to be.

34:56Anita Ramaswamy:and eric you raised 60 million dollars what's the valuation of the company now

35:02Sri Muppidi:we are uh over half a unicorn i think we can say that so well the over half the first first half

35:08Matt Bryson:of the unicorn yes okay it's missing it's missing a tail it's missing the tail yeah okay all right

35:15Anita Ramaswamy:well hopefully you find it next round we'll we'll put the tail on the unicorn so uh no worries

35:22Sri Muppidi:there.

35:23Anita Ramaswamy:Okay. So, you know, Eric, what I'm sort of hoping to get your clarity on here is what is good data and what is bad data in robotics? You know, because I think when we've been on the show, we've talked about teleoperators, I think they call them, right? We've talked about people going out with cameras themselves and, you know, kind of like Google Maps, collecting as much data as they can. You know, what distinguishes what good and bad is in this arena?

35:53Sri Muppidi:It's a good question. And it's actually very use case and context dependent. And it's something where we're actually discovering what the frontiers of quality data is. And a lot of the labs are trying to do this research to understand what makes their data actually important for their models. I think there are a few principles that you can follow. One is that you need diversity of data. And one of the issues that people have with some of the data collection that they're doing is that they're getting a lot of redundant, repetitive data. So it's not mining the different failure modes and edge cases that these models can see.

36:28Sri Muppidi:And that is like one of the big bottlenecks that a lot of the labs have at the moment.

36:33Anita Ramaswamy:And when you say incomplete data, what do you mean by that?

36:36Sri Muppidi:So you might have a lot of data of you in good lighting in your room with a robot making your bed. But let's say that it's stormy outside or you move your bed position, then the robot has never seen that particular context of data. So while a human can generalize and we know how to kind of operate under multiple conditions, these AI systems, they don't have the intuition. And so when you don't have the right data conditions in the training environment, then the robots will fail in things, in areas that are relatively trivial from a human perspective.

37:18Anita Ramaswamy:Basically, what I hear you saying is, you know, if this is a matter of someone taking a video as training data of a certain environment, I mean, you have to get every angle, every corner. It can't just be the two or three shots and hope that they get an idea. You need full coverage. Ulrich, what sorts of robots are we talking? Who are the customers that you're selling to?

37:40Matt Bryson:You know, we've seen a huge proliferation in different types of robotics applications over the past couple of years. I think the one that stands out the most has been the humanoid form factor. So there's been a huge number of companies that have started in the last couple of years with building robots that will go into your house. They will go into the warehouse. They will go farming the fields. And they're building these humanoid form factors because they kind of fit the existing human workloads. So they kind of slot well into the current set of roles and jobs that we're doing as humans. So that is most of where we're seeing the growth today.

38:19Matt Bryson:We also do work with companies that are building different types of warehouse robots that are doing things like stacking packages in warehouses or delivery trucks. It's really any robotic application that we see a human or previous industrial robot taking on.

38:38Anita Ramaswamy:And so when you sell your offering to one of these companies, whether it's in an industrial setting or a humanoid setting, what exactly is the service there entailing on your end? Are you going out and hiring a group of people to collect this data? Are you simulating it? How are you doing it?

39:01Matt Bryson:Yeah, that's a great question. So I think it's good to ground ourselves in where robotics is today. So with robotics, we don't have the luxury of having an internet that we can do pre-training on, like we saw for the language model a couple of years ago. So a huge part of what our customers are doing and also what we are helping our customers do is collect the right data for the model. So that goes into the pre-training step. And of course, the models are trained and they start to learn how the world actually works and build a representation of what that looks like. Then once the data is collected and the model has been pre-trained, then you go into the post-training step.

39:36Matt Bryson:So for the post-training step, that's what you generally look at things like annotation and alignment. You kind of look at the model's output and start to grade it. Our software helps our customers with building that complete flywheel basically from day one. So they go and they get all the software and all the services and everything else that they need to build out their robotics foundation model, their robotics humanoid or whatever else it might be. We're also starting to see our customers shift into actual deployments, which is very exciting. I think these things are rapidly advancing and coming to market.

40:11Matt Bryson:And we, of course, also are helping our customers complete the flywheel by supporting them with the post-deployment steps. So once the robot is going in and it might be emptying your dishwasher or whatnot, we have a team that helps with the teleoperation of the robot So effectively steering the robot as it gets things wrong, we kind of call that exception handling. It's also a big part of the software that we offer. And that completes the whole pre-training to post-training to deployment flywheel. And we, again, help our customers at every step of that journey. Right.

40:48Anita Ramaswamy:Eric, I mean, help me understand here. So you have the software, but do you not have a team of humans that you've contracted to physically collect that data for the customers too?

41:00Sri Muppidi:Yeah, it's a combination of both. And actually to get the right answer for this, you need to both have the operations and the software platform. Because what we discovered is that just having one is not really sufficient to cover the full data stack for physical AI.

41:15Anita Ramaswamy:Got it. But tell me, you know, I'm curious about how you seek to differentiate yourself. I, Rocket Drew, the reporter who wrote the exclusive story about you guys today, he mentioned in the newsletter, you're building a giant warehouse with the money where you're going to collect all this data. You know, I imagine other companies are doing the same thing. What's different about you?

41:41Sri Muppidi:Well, I can comment a bit. One is that we spent the last several years first building the software platform. So to do that, you have to build scalable infrastructure to handle petabytes of data. And that's what we spent the last five years doing. So we spent a lot of time doing our own architecture, working with multimodal data. And now we have a bunch of customers that use our system at scale, which will be a huge advantage once we get the warehouse completely fully up and ramped. Great.

42:17Anita Ramaswamy:Ulrich and Eric, I want to thank you both for coming on. And I look forward to tuning in to whenever this TV show of yours debuts. I'll be your first viewer, that's for sure. Next time we'll send our robot avatars. There you go. You have to come up with a name for the robot as well. All right. Well, thank you to you both. That is Ulrich and Eric, the co-founders and co-CEOs of Encord here on TI TV. Thank you. That does it for today's show. A reminder, we are on this stream Monday through Friday at 10 a.m. Pacific, 1 p.m. Eastern. I want to thank you all for tuning in. We really do appreciate your viewership.

42:51Anita Ramaswamy:Make sure to subscribe to the information on YouTube and follow us on X, Instagram, TikTok, and check us out wherever you get your podcasts. podcasts i'm already excited for our next show tomorrow have a great rest of your thursday bye-bye for now