Alex Triplett on theCUBE + NYSE Wired: Why Enterprises Are Unbundling the AI Stack

Summary

TheCUBE's Gemma Allen talks with You.com COO Alex Triplett at the New York Stock Exchange studio about why he left an investing-and-operating career to join You.com, and how the company went from a consumer chatbot to a web search API for AI. Triplett calls You.com "Google for AI" and describes four kinds of customers: agent-native businesses, frontier labs, consumer businesses, and developers.

He explains how You.com sells to enterprises, why zero data retention is a draw, and why more companies are separating web search from the model they use. They also discuss open-weight models and cost, You.com's thinking on M&A and consolidation, and why he says the market is currently rewarding messaging and speed over product depth.

Key highlights

• Why he joined. Triplett cites the founder, Richard Socher, and You.com's position below the agent layer, which he calls underappreciated and less competitive.
• From chatbot to API. You.com started as a consumer chatbot, then enterprises began asking for the API behind it to ground their agents and models in the web. He dates that pull to late 2024 and early 2025.
• Four customer groups. Agent-native businesses such as Harvey and Factory, frontier labs, consumer businesses such as Alibaba.com and Amazon, and developers building on tools like LangChain and Replit.
• Enterprise approach. He describes a sales team, a technical team, and executive sponsorship, with 10 to 15 API variants tuned along a speed-versus-accuracy trade-off.
• Zero data retention. He says You.com does not store prompts or answers, which matters most in regulated industries like healthcare and banking.
• Unbundling the stack. Separating web search from the model gives companies zero data retention on search, a choice of models, and lower total cost of ownership, which he ties to a coming "token optimization era."
• Open-weight models. He says it may not matter where a model is built, and suggests companies keep core data on closed-weight models and run long-tail jobs on open weights.
• Consolidation. Triplett expects the market to consolidate, says You.com could be a home for that, and says the company has a target list and active discussions.

Watch the interview

https://www.youtube.com/watch?v=2BqhG7QVJlY

Transcript

Lightly edited for readability: timestamps follow the source transcript, speaker labels are not included, and obvious transcription errors and name spellings are corrected.

[0:00] Palo Alto studio connection: Silicon Valley and Wall Street. I'm John Furrier here with Dave Vellante, my co-host.

[0:14] Welcome to theCUBE studio here at the New York Stock Exchange. I'm Gemma Allen, co-host of NYC Wired, and today we are talking Mixture of Experts. Joining me now is Alex Triplett, COO of You.com. Alex, great to have you.

[0:26] Gemma, thank you so much for having me.

[0:28] So, I want to get into You.com and the business and everything that's happening in the industry. But first, I want to talk a little bit about you. You've had an interesting career. You have been an investment banker, a long-term M&A scout, I guess I could say, right, at a couple of firms. Now you're in this very interesting point in the world of AI infrastructure. Tell me about, first of all, the decision to join You.com. What part of your M&A brain thought, yeah, this is a great move for me?

[0:57] Oh, it's a great question. Yeah, I did have an interesting journey. So, starting in banking, went to private equity, had this realization in 2010: I can be a better investor if I get a little bit of operating experience. 16 years later, that's all I've been doing. And so when I first went to Ion Group, a fintech company, part of my job was M&A, raising capital, integrating acquisitions. Part of my job was operating business lines that once we acquired needed somebody to go run them. And so I did both of those jobs and did that for 12 years. We grew the business quite a bit, from 150 million in revenue to 3 billion, 600 people to 12,000 in 12 years. So it was quite a run. Saw a lot of phases of growth of a company. I spent three years as CFO and COO at Appfire, a small application provider around the software development life cycle, and that was fantastic. Grew that from 100 million of revenue to 300 million. When I was thinking last summer about what to do next, and so to your question, why You.com, what part of my brain was stimulated? The first thing I was thinking about was, given my experience, where can I make an outsized impact on a business that is ripe for change, growth needs a catalyst. So that was first. Second, I wanted to participate in AI. How do you do that? Can you do that as an investor or can you do that as an operator? And I felt like being an operator within AI you just get closer to everything. You get closer to your suppliers, your customers. And for me, the decision on You.com was really centered around two things: our founder and our position. So founder first. Richard Socher has been doing this for 15, 16, 17 years. When I say this, it's AI, he was writing papers in 2009, 2010, 2011 at Stanford contemplating hooking up neural networks with natural language processing, thinking about some of the things that ultimately became large language models as we know them today. Heavily cited researcher, started a business, chief scientist at Salesforce, did all these things. When he spun out of Salesforce in 2021 to start You.com, he was the first to take an LLM and hook it to the web, because an LLM's training data ends. It needs accurate, fresh information from the web to make answers for anybody that's prompting it. And when I looked at his track record and his tenure and the fact that he's been doing this, he's got a North Star, he's a visionary, I thought aligning myself with somebody like him, probably a good move. And then the second thing, to the point of what do we do? We're in this interesting little niche, which I can talk about later, where we supplement and complement the LLM and we sit below the agent layer. And I felt like that was a really good space to gravitate towards because it's underappreciated. It's very, very important and less competitive.

[3:36] Let's talk about that niche for a second. The decision to really hone in on a niche, which has been the You.com journey, right? You mentioned that this started as, I guess you could compare it to maybe a Perplexity type of model, but it was too early for the market at that time. Now we have a lot of companies coming on the show and they're saying, we were told we're a little bit early, we don't want to be a fast follower now, we want to be a very niche player. How would you describe You.com's journey and You.com's market message at this moment? Because it has shifted somewhat, right?

[4:07] Yep. So, we are Google for AI. We are agentic web search. We provide web search APIs for anybody to search the web that's an LLM, that's an agent, that's a large consumer business, that's a portal, whatever. So we do that, right? That's the market message. That's all we do today. But you're right. We had an interesting journey to get here. When Richard started the business in '21, you're absolutely right. We started life as basically a chatbot like Perplexity, like OpenAI was when it started. We had an interface. We had LLMs in the background. We had web search hooked up with that LLM and we could provide great answers for people, typically consumers. Two things happened in the consumer space. One, you need incredible scale through marketing dollars or incredible scale through compute dollars. And others got there first. And obviously OpenAI has been phenomenal in that space, and so has Perplexity. And so what we had was this amazing search index, the search engine that was underpinning the chat. And all of a sudden enterprises started to pull us and say, wait a minute, do you have an API? I don't want the interface. I don't want any of that stuff. I just want the API, because I want to ground my agents or my model to the web. And so we got pulled into the enterprise space, pulled into the web search API space. And this is kind of late 2024, early 2025. And, Eureka, we discover this is an amazing market. There's this huge need. Enterprise is a great space to be in. It's underserved. Most people right now cannot be served by Google and Bing, the traditional search engines. We are next. We are it. And so that's what we've been doing ever since.

[5:43] Let's talk about that need for a second, because this is essentially tailored endpoints, right? It is about making sure that you get the best possible return that's available to you on the web. You have an interesting profile of customer, though, because on one end, and I'm guessing here, it's an assumption, you are I'm sure serving many enterprise customers, you're also serving some of the frontier models. Break that down. Are you a competitor in one space and a collaborator in another? Profile the customer base for me here.

[6:12] Yeah, absolutely. Astute observation, enterprise customers and frontier models. So I think about our customer set as basically being four categories. Category number one, agent-native businesses. This could be Harvey, which is a great customer for us, right, the ubiquitous legal agent. Could be Factory, coding droids, amazing company. I would put Salesforce in that category with Agentforce, Slackbot, we serve them as well. So, agent-native businesses. Second would be the frontier labs. So there we could serve them at inference, because every frontier model needs web search. All of them have it. Sometimes they use us, sometimes they use other people, but all of them need web search at inference when they have a prompt or make a call. So we can serve frontier models. We also can help them with pre- or post-training as well. The third would be any consumer type of business. Typically it's an agent application. So think Alibaba.com. They have an AI mode. Those are summarizing specs around a product. We serve them. We serve Amazon to discern trends out in the marketplace that they then will reinforce how they position products in their marketplace. So that can be consumer-based businesses. And then the last is developers. Anybody that's a developer that wants to build, that goes on LangChain or Replit or OpenClaw or Hermes or whatever it might be, [unclear] Studio, whatever the application is, and they want to just build and they want to take an API and plug it into the agent that they're building, that could be us.

[7:38] I mean, GitHub is certainly a [goldmine?], so I can certainly see the value in that. I want to talk about enterprise, though, because we talk to a lot of folks who come on the show and talk about nailing enterprise, right? The enterprise win is the big bet of this next era.

[7:50] Yeah.

[7:52] You're not a [unclear], though, right? You guys are very much about ensuring that what is broadly available is returned in the most succinct and, I guess, accurate fashion possible.

[8:03] Yeah.

[8:04] What sorts of TAM do you have within enterprise? Are you working with HR departments, finance departments? And I want to talk a little bit about the competitive side of that, but first I want to understand, tell me how the product is being used. Like, what niche are you building on?

[8:18] Yeah, it's a great point on how to crack the enterprise. So typically the conversation, if you're in one of those AI-native businesses, there's an engineering team that probably understands search or has somebody focused on search, and so we will cater to them directly. And there the buying motion is highly technical. It's very specific. They know exactly what they want. And so our positioning is we call ourselves a search lab. You mentioned lots of endpoints. We have a variety of endpoints that operate across the Pareto frontier. You can be fast with a little less accuracy when you sacrifice when you get fast. You can be a little slower with higher degrees of accuracy. There are lots of parameters, enrichments in the API. So we have basically 10 or 15 different varieties of our APIs. And so when we work with these agent natives, we're trying to tune the parameters of the API to exactly what they want. When we go into the frontier labs, it's typically somebody that's solely focused on data quality and search. And so that would be the buyer there. In the consumer space, it's typically a CTO, CIO, that type of person within the organization that's thinking about broader strategy. And then individual developers is just exactly that. But I think with the enterprise, the first thing that people often forget is human interaction matters. So we've tuned our team to have a sales organization that interfaces directly with the customer, that can work through their process, that understands how they want to buy, what their biggest pain points are, what type of commercials they're interested in, how their process works in general, just kind of quarterbacks everything. We've got a technical team of engineers that can interface directly with their engineers, hold their hand, help them with evaluations, help them tune and parameterize the endpoint. They're all technically capable and in theory we can just hand them the endpoint and go away. But that's not an enterprise relationship that you're trying to build. You're trying to solve business outcomes for somebody. So it's really those two teams. And then the last is we have executive sponsorship. I want to make sure that any enterprise has our commitment on reliability, uptime, zero data retention, these things that really matter. It's part of the reason why they come and choose us. They know we process a billion queries a month. They know we have zero data retention. They know we have a host of enterprise customers already. We understand how to solve their needs, service them, but you also want that executive commitment as well from the entire organization. So it's really those three facets of engagement.

[10:44] Let's stay on zero data retention for a second, because that's interesting, especially in the world of enterprise, where we hear a lot about context, about really getting inside the mind of your user, your agent. Is that a competitive advantage, do you think, over, like, a Claude for enterprise? What are you seeing and hearing in the market, especially in the world of enterprise, where there is a lot of concern about what could be living out there, right, about you and your company and how that information is protected? People don't want models training on their data, for example. What are you seeing and hearing? Give me the competitive stance on that.

[11:19] Yes, it's a big, big issue, and zero data retention has gotten a lot of traction for us. And so the first thing is on zero data retention, when you send us a prompt, we won't store it. When we send you an answer back, we won't store it. And that's important because, as you say, people do not want their data being used to train a model. They don't want somebody like us storing the data, training an index or something like that, right? So it's very, very important, particularly the more regulated you are, in healthcare and banks, it's vital that you have this. The second, when you mention somebody like Claude or other model companies, a lot of enterprises we see are starting to just unbundle the stack. So two years ago, or even last year, you'd say, doesn't the LLM have web search? I'll just use that. And then you realize, wait a minute, they're going to retain my data and they're going to make their model smarter on my data and I don't know what is being sent, what my employees or my agents are prompting. And so I don't really want that. If I unbundle the stack, at least I can have ZDR on web search. So that's point number one. Point number two, when you unbundle the stack, is this really fun thing called choice. You can be multi-model. All of a sudden you don't have to use one single provider or one single model and you have all sorts of flexibility. And I think the OpenRouter announcement with Stripe is just a manifestation of that in the last six months, where people want that choice. It's highly important. And so when you unbundle the stack, you get ZDR across your web search layer, you get multi-model, and then all of a sudden you cut your TCO dramatically. And so I think the next era that we're going into is this token optimization era. And token optimization could be I'm multi-model and I use a bunch of open-weight models because guess what, they're just more efficient, or use the most efficient version of Claude or OpenAI because it saves me tokens. But also, if I isolate web search and I isolate the LLM, I can then start to squeeze costs on both. So we run all sorts of TCO calculations for our customers saying, look, here's the benefit with unbundling: A, ZDR; B, full control over the web search, parameterize it any way that you want; C, multi-model; and D, lots of TCO compression.

[13:32] I mean, TCO is certainly very compelling. On the enterprise question, though, especially as relates to open-weight models, it seems as though the mood on that is changing again, like the Irish weather, right? We hear all sorts of skepticism and enthusiasm day by day on what's happening in terms of whether or not folks want Kimi in the enterprise, right? But from your perspective, from the position that You.com is coming at this from, are you in some respects the kind of throat to choke in that scenario, you know? Are you offering a level of protection and security, whether or not it's even true that there is this additional security needed around open models or not is TBD, right? We have no real proof of that, but it's certainly a narrative that lives strong. Talk me through that a little bit.

[14:16] Yeah, great questions. So firstly, definitely when you unbundle the stack and use this as web search layer, if you want ZDR and protection on your web search queries, absolutely we're the throat to choke, and happily so. I think we also can start to promote to the enterprise, and we partner with a lot of the open-weight firms, whether the American open-weights, Reflection, an amazing company, they're going to come on strong, Thinking Machines, also a great company, they're going to come on strong, Mira, she's doing it, she's making it happen. So I think those two come to mind immediately. I know Nvidia is working on the Nemotron Coalition that's going to have some interesting side effects. We're close to the Poolside folks, right, there's that relationship with Nvidia that just happened earlier in the last couple of days. So there's a lot of the US firms that are thinking about this open-weight initiative, but also with the Chinese models, you mentioned Kimi, Alibaba Qwen, MiniMax is another one, ByteDance with Seedance and others, we are completely happy to facilitate the ability for people to be multi-model with those open weights. Our belief is over time it's not really going to matter where the model is created. May the best model win. And you're sending your data somewhere if you send a prompt to a model, that's for sure. And I'm not entirely sure that it's a negative thing to be sending it to Kimi K3 versus sending it to, you know, Claude Sonnet. I'm just not entirely sure. Internally, we're users of Kimi. We're users of Qwen. We're happy. We're users of Claude. We're users of OpenAI. We're super happy with all those. So we like the choice. We like the differentiation between different tasks. We certainly like the price of the open weights. But we do encourage our enterprises: A, how can we facilitate, your web search layer is secure, so that's nice, and B, how can we facilitate TCO compression, and open weights is the way to do that. But there are also different ways you can do this. Like let's say your core jobs, some core data, is on a closed-weight model where it's more segregated, and then your adjacent jobs, your tertiary jobs, your long-tail jobs, could be on the open weights where you're a little bit more relaxed on where that data goes. There are all sorts of ways within an enterprise to figure this out.

[16:31] I want to talk a little bit about You.com, the commercials, right, the kind of growth journey you guys are on. Break it down for me. I know it's valued at over a billion dollars, a billion plus maybe, give me the exact numbers, and talk a little bit about how you guys are thinking about things, especially from the perspective of M&A. You're an M&A guy by trade, right? Like that is your wheelhouse. Are you thinking about this point in the market? It's an interesting time. Do you think we're going to see a lot of convergence? Where's your head at with this, Alex, considering your background?

[17:01] You're leading the witness here, and it's appropriate. So, we were valued at a billion and a half in October of last year, and it felt a little early because we had some peers raise at higher valuations recently. So, kudos, but it just shows that this is a really, really interesting space in the market to be in. For sure we need consolidation. I think in any space, and we've seen this in software as a service, in every, whether you're horizontal, whether you're vertical, I saw it in fintech in spades, particularly in capital markets, where, you know, equities and FX and fixed income and all these places we land-grabbed and we consolidated. At Ion we were consolidators. At my last company, Appfire, we were consolidators. And certainly part of the reason I came to You.com was thinking about the inevitability of where all of these markets end up going, which is, after you get the understanding and the adoption of a market, and after players are established, ultimately you will go into a consolidation phase. And I think we're going to be a really good home, eventually, to consolidate many, many of the businesses that are out there in our space. And I think about either core providers that are doing something very similar to us, and so there it's just maybe reinforcing best-of-breed, maybe it's reinforcing a bigger customer base and scale, because scale matters and scale wins. But then also, adjacent to core web search, what are other things that can be consolidated? Interesting structured data assets. People that are doing something that is a little bit on the edge of maybe core web search but is ultimately delivering information or an answer from generally publicly available sources could be a consolidation target. So we have our list, we have active discussions, and we'll see.

[18:40] Last question. In this market, it is a speed market, right? It is act now, think later, in some respects. It seems, how it certainly feels watching it every day unfold. What do you think people are truly scouting for here? Is it product? Is it enterprise time? Where do you think the real acquisition-heavy conversations are truly honing in on, where people could be most successful at this juncture of the market, in this phase?

[19:07] And I see this in our space. I see this in the agent layer. I see this in the inference layer. So just above us and below us. It is about message and speed to market. Right now it is not necessarily about the absolute nth degree of product quality. Most of the providers in those spaces I name are generally pretty good, and all of us benchmark against each other. All the inference providers benchmark against each other. All the web search providers benchmark against each other. All the agent layer providers are trying to come up with benchmarks to benchmark against each other, which is refreshing, unlike in software, where you kind of sort of think somebody's better. We can all benchmark and we can see quite literally who is better, and so that's useful. But people aren't necessarily optimizing for in-house IP and product quality. They're optimizing for marketing message and speed to market, which I think is appropriate, right? We are in a race. It is invigorating every single day to wake up and know that there's heavy competition. You have to move fast. It's really exciting. But I think that's the short term, and so you've got to win that short-term battle. Long term, though, great IP always wins. And so I do see some providers out there that, if they aren't systematically building real IP under the hood, they may short-term win a marketing race, but long-term they will lose the ultimate battle, because the IP needs to follow. And so we're trying to do both at pace. But I think this phase is like the marketing and land-grab phase.

[20:36] So to summarize, I'm going to say folks need to remember that it is both a marathon and a sprint.

[20:39] Absolutely.

[20:40] Alex Triplett, thank you so much for joining us on NYC Wired. Absolutely a pleasure.

[20:44] Thank you.

[20:45] I'm Gemma Allen here at theCUBE studio at the New York Stock Exchange. This is NYC Wired: Mixture of Experts. Thanks for watching.