The Information's TITV
00:39:27 · Catherine Perloff
[00:39:27] S5: So there isn't one now, but I could totally see that. I kind of... One of my like... I suspect maybe part of the reason we haven't heard more about this is because Amazon hasn't fully worked out the business model. Maybe once they do work out the business model, it'll become more pressing. They could take a cut. They could also have this as an ad opportunity. Maybe they could start chart like if you want to show up as shop direct, maybe it's an ad. You know, Mark Zuckerberg has already said that they envision taking a cut from used transactions. I mean Amazon, the you know, the reason people don't want to sell on Amazon and the reason why they keep products on their Shopify site, even if they do sell on Amazon, is because Amazon takes out a ton of fees and the margins are not very good on Amazon. So I can't imagine if this becomes popular like they, you know, they wouldn't. They wouldn't want to give an opportunity for, say, merchants to take all their products off Amazon and take them on. Put them on Shopify. Okay, because I'm not going get the fees from Amazon and but they'll still show up there like they're not going to want to do that. They I think they will eventually collect, But right now, I think they're just trying to get the program to have traction and have consumers catch on.
00:40:38 · Akash Pasricha, Gopal Srinivasan
[00:40:38] S0: Right. Great. Well, I want to thank you for coming on. That is Catherine Perloff, who covers Amazon here at The Information. Our next segment is with our partner, Gemini, and we are speaking to one of their implementation partners at Deloitte. Gemini released Gemini IV Argonne a few weeks ago as the model race continues to wage on Fast and Furious. I want to bring on Gopal Srinivasan, Principal and Global AI and Data Lead at Deloitte, who is working with businesses to integrate Gemini into their systems. Gopal, welcome to the show. It's great to have you here. [00:41:13] S4: It's a real pleasure. Thank you for having me, Akash. [00:41:16] S0: So, we're kind of at this interesting point in the AI race right now, Gopal, where you have Google, which they've been releasing models. They've been under pressure show that they can compete with the other frontier labs at the very top of these benchmark leaderboards. So that is one story that's playing out. On the other hand, you have the Anthropic S 1 out this week and you have their own admission that, Hey, our models are really good and we're still trying to get people to grow into these capabilities. These feel like two contrasting realities because on one hand, we're so focused on having the best model. On the other hand, people can't even use the best models. How does that coincide with what you're seeing on the ground right now in your business?
00:42:10 · Gopal Srinivasan
[00:42:10] S4: It's a it's a terrific question. There is certainly a lot of narrative both in the media, as well as if you go into various social forums around who has the best model, who's ahead, who's behind. And these things change every few weeks. Somebody gets up to the top, somebody catches up, and that's been a roller coaster ride for more than a couple of years now. However, where I operate predominantly in in working with Google is serving your typical Fortune thousand plus enterprise. And when we go inside of those environments and look at where can AI create impact and value, the capabilities that come with the more mid tier models are serve many of the use cases. The most frontier, highest tier models are valuable in certain use cases, particularly if we are applying it towards large scale code migration. If we are, say, using it in very advanced use cases like drug research. In those kind of use cases, there is certainly a need for the most capable model. But if you look at most real world application within enterprises, the mid tier models, so in Google's case, that would be the flash tier of models. They're more than been doing the job. And candidly, whether we had a current pro model or the lack of it, it did not really impact anything that our mutual customers were doing, and we continue to see significant implementations and rollout of various AI use cases.
00:43:58 · Akash Pasricha, Gopal Srinivasan
[00:43:58] S0: Right. And, you know, I I I wanna ask you about the traction that we're seeing around TPUs, because this is a topic we've been talking about on the show today. I mean, now OpenAI is using TPUs, Oracle using TPUs. I expect it's going to be front and center tonight for Google's quarterly results as well. Good morning. When you are talking to customers about Gemini, how much are they thinking about whether or not they can use an end to end offering from Google with the cloud, with TPUs, also the model. Is that an important consideration for them? Is it sort of a second consideration? Where does that come up? [00:44:45] S4: The full stack approach that Google has is certainly a big consideration. The fact that Google brings its own first party products at every layer of the stack from the silicon at the bottom to the application at the top of the layer and everything in between from the platforms to the models is a major differentiator. And many of my clients absolutely consider that as an advantage for many reasons. One, it manifests itself in the seamlessness of being able to build and deploy agents and AI applications. Second, because it's optimized end to end, it's highly performant. And third, it also translates into better economics because of that end to end integration and the fact that they designed their silicon themselves means the overall cost to any end customer tends to be more optimal than, say, considering alternatives.
00:44:45 · Gopal Srinivasan, Akash Pasricha, Umesh Padval
[00:44:45] S4: The full stack approach that Google has is certainly a big consideration. The fact that Google brings its own first party products at every layer of the stack from the silicon at the bottom to the application at the top of the layer and everything in between from the platforms to the models is a major differentiator. And many of my clients absolutely consider that as an advantage for many reasons. One, it manifests itself in the seamlessness of being able to build and deploy agents and AI applications. Second, because it's optimized end to end, it's highly performant. And third, it also translates into better economics because of that end to end integration and the fact that they designed their silicon themselves means the overall cost to any end customer tends to be more optimal than, say, considering alternatives. [00:45:44] S0: Now, one of the interesting transitions that we've seen from Google Cloud is that they're now offering access to other companies' models as well. Right? [00:45:54] S2: Mhmm. Mhmm.
00:45:54 · Umesh Padval, Gopal Srinivasan, Akash Pasricha
[00:45:54] S2: Mhmm. Mhmm. [00:45:55] S4: That is true. So Google has this layer in the platform called the model garden, which offers a variety of models from not just the first party Gemini models or their own open weight Gemma models, but models from other labs as well. And, they've offered, for instance, Anthropic for a long time. And as, many of you and your viewers may have seen recently announced that OpenAI models will also become available. And, I think if you've been following what they've been saying forever is that Google Cloud as a hyperscale platform for building agentic and AI applications will take the approach of offering optionality to customers. Their their differentiation is centered on we wanna offer the widest choice to our customers and create the most seamless place where you have the choice to use different models, different frameworks. They support third party frameworks, first party frameworks. They came up with the a two a protocol, enables agent to agent interoperability. So that has always been the philosophy. And so the news itself perhaps doesn't come as much of a surprise. And at this point, you can get models from all the leading labs out there Right. Available on the Google Cloud Agent Platform. [00:47:25] S0: Although, I mean, the the question that does come to mind for me is, does that not then cannibalize use of Google's own models that it is developing and trying to convince customers to use?