Nvidia GTC 2026 Takeaways, OpenAI-AWS Pentagon Deal, Asana CEO on AI Agent Tool Launch

0:13Welcome everyone to The Information's TI TV. My name is Akash Pasricha. It is Tuesday, March 17th. First up today, NVIDIA is making headlines at its flagship conference, including a trillion dollar revenue projection. will discuss the biggest news out of GTC, including NVIDIA's new Grok chip. The information also has exclusive reporting that OpenAI has signed a new contract with Amazon Web Services to sell its AI to government agencies. Shreem Upidi, one of the reporters behind that story, will join the show to share with us what she knows. MasterCard said it will acquire stablecoin infrastructure platform company BVNK for$1.8 billion.

0:53dollars. We'll bring on our finance editor to talk about what that means for the crypto sector. And we'll wrap the show with a conversation with the CEO of Asana as it pursues its own AI strategy amidst the SaaSpocalypse. It's going to be a fun show, so let's get right on into it. NVIDIA made a string of big headlines at its flagship GTC conference that kicked off yesterday. At the center of the event was Jensen Huang's keynote. He projected the company will generate $1 trillion in revenue between 2025 and 2027 from its Blackwell and Ruman chips. The company also unveiled its inference-focused chip based on Grok's technology.

1:30It is a paradigm shift for NVIDIA. I want to bring on Nick Patience, VP and Practice Lead for AI at the Futurum Group, and Aaron Ginn, co-founder and CEO at Hydra Host, to help us break it all down. Gentlemen, welcome to the show. It's great to have you back. Hi, Cash. Thanks for having me. Nick, I want to start with you. So this trillion dollar projection, what did you make of it? Well, obviously, it was certainly a bold. Jensen likes bold numbers and bold messages. So I think it was certainly that. I think the average kind of estimate for that period of revenues was something like 835. So it's certainly a big tick up on that.

2:08I guess, yeah, it matched to, though, kind of a broader narrative about AI infrastructure, I think. the you know the key point is that you know he sees inference as such a such a big part of this is no longer just about training um and it obviously speaks to you know the the issue he doesn't think there's any there's any going to be any slowdown in in demand here okay and so speaking of inference nick i want to stay with you the grok chip did it meet all of your expectations did it did it over impress what was your reaction it's pretty impressive um considering they only acquired the company just before Christmas.

2:44So that's very fast by any standard. So I think the big story around it is not only the chip they're talking about, but what it says architecturally about NVIDIA. So pairing GPUs with Grok's LPUs in a kind of heterogeneous inference stack, which is quite a departure from NVIDIA's GPU only or GPU primarily model. And if that 35 times tokens per watt claim holds up, and I have no doubt that it will, that kind of changes the cost curve for running agentic workloads at scale. I think it's NVIDIA acknowledging that inference is fundamentally a different engineering problem from training. Aaron, you're on the ground there at GTC.

3:28Take us inside your conversations with folks there and their reactions to the Grok chip specifically. I think that the theme that I'm feeling, this has been coming out here for the last half a decade, is that there is definitely a vibe shift towards like Jensen selling confidence and that there's been kind of a shift from the opportunity that's being presented and being able to kind of future sell different solutions to, as in like, you know, application layer down to models, down to like chips, to data centers. And now it's much more about the fact of like I have continued to over deliver both on the engineering front and on the finance, you know, kind of returns front.

4:09And I think that the message is that, you know, that you can trust NVIDIA, that when you build on our platform, no longer you can make money, but in that this is like a real kind of permanent part of the future digital infrastructure. But it's something that you can bet on, that you can bet on his word and his technology, and that he's going to do the best possible, both to deliver that, you know, as you mentioned earlier, the$1 trillion, which would be obviously incredible. but he has delivered every other single time. And the Grok integration, which I agree with Niko, it was surprising to see how fast they were able to kind of like deliver something.

4:47But it goes to the continued same theme of like, NVIDIA is going to build the best chip possible. And it's not abnormal for Jensen to talk about how he's constantly going to out-innovate his own company. So yeah, you could take it one direction, which is like the acquisition of Grok was like an admission that the GPUs aren't the best in every single use case, but it's also from another trajectory. Like Jensen also bought a company, built a, you know, integrated a new type of chip architecture and then launched it in like, what, six months. So it's like, it's pretty incredible. Let me ask you this though, Aaron.

5:23So, you know, I hear you on the narrative piece of this and these types of conferences, you know, you often get the biggest cheerleaders for the companies showing up. I mean, the stock yesterday didn't really move on the keynote, and we already knew some of the developments that were coming from previous reporting and stuff like that. Is anybody on the ground there talking about competition to NVIDIA and the fact that inference-focused chips, I mean, there seems to be a lot more competition from the likes of Google and from Amazon. Again, we're not saying that it's eating giant market share yet, but there is competition, right, Aaron?

6:04Well, yeah, of course. But until those companies start reporting those chip rental revenues separately publicly, that's not that interesting. Because who actually knows the breakout within their own claw fabric? And the answer is it's pretty low. If you talk to anyone actually in Google or any Amazon, the actual demand for those is not extremely high. And they're definitely servicing workloads, just like I've been on the show talking about Huawei. Huawei is, I think, probably the most legitimate competitor to NVIDIA. But Huawei is also taking a very similar now approach to the way NVIDIA is approaching.

6:40It's getting market share, which is like NVIDIA is becoming one of the largest investors in open source models. They're looking at different kind of lower cost chip architectures. and that type of approach is basically very similar to the way Huawei is going to market. So there is legitimate competition. I think Jensen would fully acknowledge that he's not in a position of – he's the only person hanging around. I think you would argue that he also, though, going back to the market argument about there was a spike and then fall. So it looks like some people – some day traders having some fun. Right.

7:17And when we say, yeah, when I say the stock didn't move and it did move a little bit, we should acknowledge it's back down. Yeah, a trillion dollar. Right. Yeah. So what I would say to that is like, I think it goes to the kind of like the market likes exciting and Jensen's delivered like incredible returns every single quarter. And it's almost like, oh, he's doing it again. So I think there's some sense of like, yeah, we get it. Like you're the best and you're going to continue to blow through numbers and things like that. So I think it's some of that. And I think there's also the kind of capitalization problem with the space of like, how are people going to continue to pay for this in a way that's structurally aligned to like returns, like basic CapEx returns and things like that.

8:00Like how to structure this debt that goes into buying this stuff. I think that's kind of an open question still on the market. Nick, I want to come back to you here. So let's go back to the keynote and what they talked about. So NemoClaw was another thing that Jensen mentioned that intrigued me. Also, the discussion around Feynman, their next generation of chip. Talk a little bit about what you made of those two announcements, or if there was anything else that stood out to you about the autonomous driving developments, stuff like that. The NemoClaw thing was really interesting. I mean, the software stack that NVIDIA has been building over the last few years has really caught my eye.

8:40I guess the best way I would characterize what he announced yesterday in terms of software is an argument that the kind of the agentic AI era has an infrastructure problem and NVIDIA is here to solve it for you. because they had that agent toolkit, which has these three elements, one of which is the Nemo Claw secure runtime, and then the AIQ open research agent blueprint, and then the NemoTron family of open models, each addressing a sort of different kind of failure mode, you might say, of what makes enterprises reluctant to kind of deploy autonomous agents at any production scale. Obviously, it's very early in the genetic stuff, so we're not expecting people to be doing that, But there's a lot of issues around agentic security that they're kind of taking a secure first approach to that.

9:29So I think that was really interesting. The Feynman stuff I thought was pretty interesting as well as Vera Rubin, obviously. Vera Rubin is a kind of full stack architecture as a single computer with seven chips and five racks. So I thought that was really interesting. But the agentic stuff and the, as Aaron mentioned, the open source models, I thought was interesting. There was one throwaway remark that he made about Nemotron, which I thought was pretty telling. He said something about, we're not going to stop working on them. Nemotron 3 will be followed by Nemotron 4. I think that was a bit of a dig at sort of Meta and DeepSeek, whose version 4 hasn't come out yet.

10:10And potentially what happens to Alibaba's QN? You know, there's rumors it might be going closed source. It might end up that NVIDIA is the dominant provider of open source models. Who knows? Aaron, does that echo what you've been hearing at dinners on the ground? Yeah, it makes sense, right? That the more open source continues to proliferate through the market, the more people buy different types of NVIDIA chips to run on it. Versus closed source, there's someone between Jensen and the end customer. And you can see with every single technical direction that he makes, and he's outlined it very clearly with Vera Rubin, is that he wants to be in the business of delivering a turnkey compute product, like a turnkey AI factory turnkey of turning energy into tokens.

10:51And he wants to be the definitional layer for that. And I agree with Nick on what was very interesting around the way he talked about agents was one of the most clearest signals to revenue versus key equipment. And so leaning really heavily into that and providing, I guess, continually iterating, saying, hey, we're going to deliver the best compute architecture to change watts into tokens. And now we know that agents is clearly like the agent era is like coming upon us. He has to introduce something like new to support that. Outside of just having like training architecture. Keep pushing the envelope.

11:29Yeah, exactly. So last question for you, Nick. Was there anything that you wanted more clarity on? I mean, this keynote was more than two hours long. There was a lot there. He touched on it. Any questions that went unanswered? I don't think so. One of the beautiful things about GTC is, and I've been many times, is that there is only one keynote, really. So that does make a bit of a refreshing change. He's so dominant. I guess the Omniverse stuff was quite interesting. I think that was, he sort of says something along the lines of, us vendors have never really met until we get to the data center and now we have Omniverse.

12:08And I think, yeah, that kind of technology has been seeking a problem to solve for many years. And this may be part of that. So I guess something about, you know, I mean, there was a lot there, to be honest. I mean, there wasn't much missing, but a bit more about how it all manifests itself in the data center where it's not just the NVIDIA kit. It's all the partners kit as well. Aaron, what are the big questions that you are trying to get answers to there for your own company and for the sector at large for the next few days as you're there on the ground? Yeah, it's a good question. I think the big question I have is like, when will Stargate actually be in the stars, maybe, since that's what he opened up is.

12:46I'm frankly bearish in all of those ideas. Jensen is somehow going to figure out a way of actually running GPUs in space from his own technology perspective, maybe it would be more interesting. But having we're one of the largest neoclods in the world and running this stuff in a data center is not easy. Saying we're going to put it in space and just let it float, float around, float around the Earth. since the opening picture of his keynote was GPUs in space. So maybe that's going to be one of the next iterations is actually a GPU designed to be in space to where it doesn't break. It doesn't break, it isn't super light, it can't handle all the G-forces to go up there.

13:31So that's my biggest question. I was like, okay, if we're going to actually really do this, then maybe it actually is working on it. Like that we're going to have a new type of technical architecture for space. Well, and, you know, I can't think of a better ad in some ways for the SpaceX IPO. The more that Jensen talks about putting chips in space, even if SpaceX's name is not on it, I mean, this is the whole narrative behind that IPO. Anyway, I want to thank you both for coming on. That is Nick Patience from the Futuring Group and Aaron Ginn from Hydra Host here on TI TV. open ai has signed a deal with aws that will see it sell its ai tools to the government in the u.s this deal comes amidst a rift between anthropic and the department of defense open ai's aws contract will include work with both classified and unclassified information i want to bring on our open ai reporter shri mupiti to help us break it all down shri welcome to the show it's great to have you back excited to be here okay help me understand this so open ai is signing a deal with AWS or with the government?

14:37So OpenAI signed a deal with the Pentagon after Antropoc failed to move on its red lines. And so essentially this deal is to provide ChatGPT and custom AI to the Pentagon. This is a single digit million dollar revenue deal, which is a new piece of information, which I don't think people have really understood. It's actually very, very small and is a small bucket in the$30 billion of revenue the company expects to bring by the end of this year. But simultaneously to support that, OpenAI signed a deal with AWS, which supports lots of national security customers to actually power the, or provide the computing power to provide chat GPT and the custom AI products to the Pentagon.

15:19Okay, and so I'm sort of, I'm trying to sort of wrap my head around this idea that they have signed the deal with the Pentagon and the deal with AWS simultaneously. I mean, is it that both of these parties are paying OpenAI for the software here? Do you need one to have the other? How does it work? You definitely need one to have the other because national security agencies such as the Pentagon mostly run on AWS. AWS was one of the cloud providers that services a lot of these types of players in this space. And so before, OpenAI couldn't actually provide its services to national security partners because of its exclusivity clause to Microsoft.

16:04And so last fall, when it restructured itself, it actually created a carve-out so that it could work with other cloud providers such as AWS to provide these types of services. And so this exclusivity carve-out really led the way to OpenAI, now landing this deal with the Pentagon and then also having this deal with AWS to provide the Pentagon with its services. Got it. And last question here, I'm going to get into big picture questions here in a second. But so is AWS paying OpenAI then in this scenario? Or is AWS just saying, hey, we will agree to host your software on our cloud? What is the exact arrangement there?

16:45I don't know the specifics of the deal itself. But as you know, Amazon also invested$15 billion or committed to invest$15 billion in OpenAI. and it could be up to 50 billion in OpenAI if the company, for example, achieves AGI or goes public. But as part of this, OpenAI also has an agreement of expanding its computing power that it buys from Amazon. So this was originally 38 billion over the course of seven years that they signed last fall. And then this most recent deal is part of the 50 billion investment with Amazon. OpenAI expanded it to add on 100 billion more. And so my understanding is maybe it could be a mix of sort of being able to work with this, but it could also be a separate deal that they signed with the government, which is my understanding.

17:30But we don't know the specifics of the deal. What we do know is that Amazon is actually going to also sell OpenAI's products to its broad customer of national security agencies. And then as part of that, any revenue that is generated will actually be shared. And so this will be some sort of revenue share model as the companies co-sell together. Now, one of the things I found interesting about your story is you made the point that the government contracts actually are a pretty small portion of revenue. And if you look at the aggregate dollar amount, I mean, these are not the biggest contracts in the world.

18:04But there is sort of this notion here that if you can get government contracts, that is seen very highly from the perspective of corporate customers. And this is, I mean, you mentioned Palantir, the story as one example of this. I mean, just talk a little bit about how much corporate customers look to government contracts as sort of a sign off in some ways of credibility. Totally. OpenAI is definitely moving into the enterprise game. They launched lots of different products to service them. They have Frontier, which is a forward deploy engineering motion so that they can partner with consulting partners and get in front of enterprises.

18:42But I think one of the strongest indicators or signals that this is a trustworthy, secure, compliant software is by actually working with the government. It's a side off, as you said, where enterprise buyers might feel comfortable working with OpenAI because if the DOD or other important agencies within the government are working with OpenAI, then they feel more comfortable. And so I think it's a strong signal that even though that it's not generating that much money, for example, single digit dollar revenue. I think another classified deal that OpenAI had was 20 million for servicing national labs.

19:17So as you see, these are very, very small deals compared to the larger amount of money that the company is generating. And so again, I think it's a sign that the government believes in OpenAI and then maybe enterprise buyers will feel more secure in buying it. I think another point to that is that actually what my reporting or like when I've been talking to folks across, it's that OpenAI also believes that it's like to their mission of benefiting humanity. And so like they believe that by providing government officials access to its products, these officials will also be able to benefit from OpenAI as well.

19:58And this is, we should make a distinction here. I mean, this is sort of the first time that we've started talking about classified information with respect to OpenAI's work with the government, right? I mean, previous to this, it was, there may have been some work, but it was largely dealing with unclassified information. Exactly. It was largely dealing with unclassified information. So last August, as did other AI labs, OpenAI provided Chack GPT for a dollar per agency across the federal government. And then they've sort of pushed ChatGPT to other branches of the government, as well as to the state and local levels.

20:31The other classified deal beyond this Pentagon deal that the company just landed a few weeks ago was this Labs deal, which I was mentioning. And so that was roughly$20 million. And that was unique in that OpenAI was able to deploy its models on, or sorry, deploy it on-prem, which means that the labs had servers directly that they were able to run it. And so that was due to, again, security constraints. But this is primarily as a result of the exclusivity clause. And so with Microsoft, and so now that OpenAI was able to get out of that or have a carve out of sorts that they're now able to really move into this space.

21:10What do you think that Dario Amadei is thinking of this news as he sees it come out right now? In terms of... I've seen OpenAI moving in on the turf that Anthropik has sort of had. Yeah, this is totally my own speculation, but I'd imagine he's pretty angry or maybe he sees it as opportunistic in terms of OpenAI coming in just as Anthropik was unable to reach a deal. But I will say that OpenAI had, as I said, been working on creating this carve out back in the fall and summer of last year when they were negotiating with Microsoft. So I think we actually, like I personally felt like I saw this coming because they had a very specific carve out to be able to service national security customers.

21:58And so when this happened, I was like, oh, that makes sense because Microsoft is allowing it. And so that was really interesting to me. Great. Well, before you go to the cloud, Shree, I do want to ask you very quickly. I mean, we've seen headlines this week about a memo that Fiji Simo sent out last week, sort of echoing some of the comments that we've reported have been surfacing inside OpenAI, the idea that we need to really focus on our core product in ChatGPT, and maybe the everything, everywhere, all at once strategy is not the right approach. I mean, look, we've reported on the CodeRed previously.

22:34Obviously, you've reported on some of these developments. But I mean, was there anything different about the comments last week? Is there anything that strikes you as really moving the needle forward from Fiji's SEMO this time? I think that the comments Fiji was really focused on was to really, as you said, focus the company. And we've reported on some of this. So for example, we reported that Sora's video generation products are getting rolled into ChatGPT. We reported on instant shopping being pivoted into directing shoppers to actually do direct links to actual merchants rather than partnering with like Shopify.

23:18And so these initiatives that really sort of like sprung out of OpenAI last year are now sort of getting more focused and getting rolled into ChatGPT or like being very, very proactive into focusing on ChatGPT, which I think is interesting. And I think what we'll expect to see is probably like lots of frustration because maybe folks that thought they could work on innovative projects are sort of getting really honed in on ChatGPT and then likely ads because ads is like going to be one of the largest revenue drivers. And so I think that'll be interesting to see in the next couple of weeks as well for folks to actually focus at OpenAI.

23:56Yeah, no, I mean, I certainly I didn't think about the human level of being recruited for a company and saying, hey, you will get to work on maybe the fifth, sixth, seventh largest product at the company. And then a little bit of this pivot saying, hey, we're still focused on one, two, three. So, you know, maybe you joined the company with a certain perception and now you're being put to work elsewhere. All interesting developments all around. I want to thank you for coming on. And that is Shreem Upidi, our OpenAI and Anthropic reporter here at The Information. MasterCard announced it will acquire stablecoin infrastructure startup BBNK for up to$1.8 billion.

24:34This comes after deal talks between BBNK and Coinbase fell apart back in November. Traditional payment platforms like MasterCard are increasingly trying to keep up with changing tides in the payment space. I want to bring on our finance editor, Ken Brown, to help us make sense of all of this. Ken, welcome back to the show. It's great to have you here. Hi, Akash. What did you think of the, and maybe let's just start, what is BVNK? I mean, it's another crypto startup we have to wrap our heads around. What do they actually do for stablecoins? They are an on-ramp and an off-ramp. You can go from regular currency from dollars into crypto, and you can go from crypto into dollars.

25:12I mean, that's essentially their main purpose. They do a lot of the technical stuff with blockchain and all that, but that is what they do. Okay. And so MasterCard is out here saying, hey, we use credit cards, but we want people to be able to transact essentially in stable coins. Is that the idea here? It's all stable coins. And yes, it is, you know, MasterCard has this incredible business like Visa where they, you know, do all these card transactions and get a cut of everything. But it's all dollars to dollars, dollars to pesos, dollars to euros, whatever. And so, yeah, there's this big growing area of dollars to stable coins or stable coins to stable coins.

25:54And, you know, MasterCard doesn't want to be left behind. so you said left behind which is a good segue here i always try to think about these deals as offense or defense here and look i i guess you could make the argument for both in some ways because it's not like stable coins are that big right so maybe it's a little bit of offense and getting ahead of it on the same time i mean i don't know other other payment companies have made stable coin acquisitions before so maybe it's defense how do you think about it well you You've got to come down on one side or the other, right? Okay, yeah. You got me.

26:28You got me. I think it's defense. And the reason is the world is moving really fast. I mean, MasterCard said today, you know, we could have built it ourselves, but, you know, there's a speed to market element of this. And so, you know, they don't want to be left behind. That's really their thing. I mean, they're paying, they have plenty of money, but they're paying$1.8 billion to not be left behind. I mean, BBNK gives them, you know, a good entree point there. They're well-respected. They've done a lot of work in this space. I mean, it's only a five-year-old company. It's not like it's, you know, there's a lot of legacy to any of this stuff.