theCUBE + NYSE Wired: AI Factories - Data Centers of the Future | Abhas Ricky, Cloudera

Abhas Ricky, Cloudera

In this segment from theCUBE + NYSE Wired’s “AI Factories – Data Centers of the Future” series, theCUBE’s Dave Vellante sits down with Rob Biederman, managing partner at Asymmetric Capital, to unpack a disciplined approach to early-stage investing amid AI-scale infrastructure shifts. Biederman explains Asymmetric’s founder-first model: writing $1–$10M checks (often via SAFEs), joining boards as they form and helping operators with go-to-market, operations, finance and strategy (not product/engineering). He shares why the firm avoided 2021’s lofty SaaS multiples in favor of backing proven builders earlier (single-digit pre-money), and highlights portfolio execution such as a cash-efficient LATAM e-commerce company scaling from ~$1-2M to about $50M in revenue. The discussion also explores Asymmetric’s subscale buy-and-build plays (e.g., pool cleaning in San Diego, sleep apnea clinics in Houston), where density, tech-enabled services and platform ops expand margins and enterprise value. Biederman weighs in on AI economics as enterprises race to “AI factories,” cautioning that not every AI workload creates ROI and that overbuilt compute assumptions could face a reckoning. He argues that winners will prove a clear 10× value equation and avoid scaling go-to-market before product-market fit. Additional insights include early liquidity discipline (returning $0.20 on the dollar before the fund’s third anniversary), portfolio survivability (34 of 35 companies still operating; three positive exits), and guidance to founders: make your value proposition relevant, credible and differentiated. Tune in for candid perspective on how capital efficiency, ownership discipline and anti-thematic sourcing intersect with a world where GPU-dense data centers and AI-scale software are reshaping enterprise infrastructure and economics.

Share this session

Clips
More from theCUBE + NYSE Wired: AI Factories - Data Centers of the Future

Abhas Ricky

Chief Strategy Officer

Cloudera

play_circle_outline Emergence of AI agents enables domain intelligence and production-grade workflows across industries.

play_circle_outline Cloudera launched an enterprise AI ecosystem and partnered with major firms like NVIDIA and AWS.

play_circle_outline Importance of strategic partnerships and co-design in developing effective AI solutions.

play_circle_outline Exploring Secure AI Ecosystems: Bridging Wall Street and Silicon Valley Through Governed Access to Private Data

play_circle_outline Focus on private AI to reduce operational costs and enhance data governance and security.

Info
Transcript

Abhas Ricky, Cloudera

Abhas Ricky

Chief Strategy Officer Cloudera

In this theCUBE + NYSE Wired coverage of "AI Factories - Data Centers of the Future," Cloudera Chief Strategy Officer Abhas Ricky joins John Furrier to discuss the rapid evolution of enterprise AI from simple RAG pilots to complex, multi-task reasoning agents in production. Ricky outlines Cloudera’s expanded ecosystem strategy, detailing how partnerships with industry leaders like NVIDIA, AWS and ServiceNow are providing customers with the optionality to deploy AI across hybrid environments. The discussion underscores the necessity of a modern data platform t... Read more

explore Keep Exploring

What are the telltale signs indicating the development of domain intelligence in large-scale distributed computing systems, and what is the outlook for Cloudera's business towards the end of this year and into next year? add

What advancements or initiatives have been undertaken in the AI ecosystem to enhance customer experience and strategic partnerships? add

What does the modern AI ecosystem look like? add

What is the significance of private data in the context of AI development? add

What is the significance of private AI in relation to compute costs and infrastructure usage? add

bolt Powered by CUBE AI

Abhas Ricky, Cloudera

search

John Furrier

>> Welcome back everyone to theCUBE's live coverage here. We are in the NYSE Studio of theCUBE. Of course, we have our Palo Alto Studio where we bring Wall Street and Silicon Valley together, talk about all the latest trends in tech. This is our AI Factory series, our most popular one. The AI infrastructure is continuing to accelerate into next year. We see still more massive growth, enabling the data layer, enabling agents, enabling the value creation and extraction from these AI factories. Distributed computing, all hybrid cloud, all looking good. Abhas Ricky's here. He's the Chief Strategy Officer of Cloudera, continuing to provide the data platforms. Abhas, great to have you on theCUBE for our AI Factory series. Appreciate it.

Abhas Ricky

>> John, thank you for having me. It's a pleasure.

John Furrier

>> We had a chat at AWS re:Invent. We saw obviously new level of agents, frontier agents, Nova Forge. All these are tell signs that the agents are starting to come in with domain intelligence. You're starting to see the value of these large scale systems. It's just distributed computing in a hybrid cloud environment. On-prem, edge, coming fast, we expect that to be a big theme in 2026. You guys are continuing to be at the center of the data platform. Give us the update on Cloudera. How do you feel about the end of the year and what's your thoughts on going into next year?

Abhas Ricky

>> Absolutely. I think it's been a great year. On two fronts. One on normal business, which is the lakehouse business, so that large organizations can run analytics wherever they like at the price point of their choice and deployment form factor of their choice. That's continued to grow significantly. We've had significant use cases which are not only mission critical, but also forward-looking that came through on the platform in a production environment this year. And all kinds of AI use cases, including when people who were doing RAG, they've now started to do fine-tuning. A lot of customers who were doing function calling. AI agents have now moved on to the point whereby they're doing multi-task reasoning agents in production across industries, financial services, telco, insurance, oil and gas, healthcare and life sciences. So it's been a momentous era in terms of the usage of AI and production for our customers and helping become their strategic partner of choice. So it's been great. And on the second front, we launched the enterprise AI ecosystem. I think you guys were there two years back when I announced that with NVIDIA. We have an inferencing service with them. So we take the NVIDIA microservices architecture. We add things like model registry, model catalog, high availability, so that customers can then get the best TCO advantage for accelerated compute. But we also had a partnership with AWS Bedrock so that you can access any model for Cohere or Anthropic using a single click on collateral machine learning within the confines whereby it's a user experience that is way, way better than what used to be before. That's how we started off two years back, and then we added Pinecone for semantic . Last year, we bought along Google with the Vertex Model Garden suite. We actually partnered with Snowflake for S catalog interoperability. And then there was a series of startups that we brought along, whether it was Kore.ai for the purposes of orchestration of agents or even Dell. So we have pre-validated prefabricated reference architectures that we actually work with Dell and NVIDIA so that customers can now go from zero to production in a matter of days and weeks, which erstwhile used to take six months or more. But the one thing that I've been super excited about this year is the new age partnerships that we have done. So we announced our partnership with Fundamental, which is a foundation model company for Tableau data. We're super excited about that. We announced a partnership with ServiceNow. We have built connectors because then the ServiceNow agents will now have access to the operational context sitting on the lakehouses so that they'll have a higher fidelity output, which their customs want and our customers want as well. We have a partnership with Galileo in the AI agent observability space. Tracing, debugging, and a lot of other use cases are becoming very important, but also LLM evals. LLM, as a judge, as a concept is becoming more mainstream, and we want to provide our largest customers the ability to be able to drive that through. So there's a series of things that we've been working through and we cannot be more excited and well positioned going into the new year.

John Furrier

>> I love the ecosystem strategy. Congratulations. And I want to get your thoughts on that because one of the things that's come out this year from theCUBE interviews and all the data we've been gathering is there's a new modern ecosystem out there. I'd love to get your reaction to that because it's not just do a deal, integrate in. There's a lot of co-design now with these data platforms. You mentioned NVIDIA and AWS, two great partners. Okay, great infrastructure enablers. Talk about how the ecosystem has changed in this modern era. And what are the new standards, what's the bar for being a good ecosystem partner? NVIDIA has a term they call radical co-design. Obviously they have a supply chain, a little bit more deeper integration, but we're seeing a similar co-design thinking with ecosystem. What's your reaction to that? What would you say to someone who said, "What's the modern ecosystem look like?"

Abhas Ricky

>> I'd say the first thing is people have to understand that AI is a team sport, and majority of the organizations are trying to enable different parts of the stack, but also different parts of the workflow itself. And that's largely because of the fact that I sincerely believe workflows will be reimagined. It won't just be automated or digitalized. You don't know what you don't know. So we have to look at it that our largest customers, or anybody's largest customers, will reimagine workflows and how do you actually help them on their journey. So the way we've gone about that is we have partnerships at every layer of the stack. So we obviously partner with NVIDIA. We have been for the period of time. We started off with library integration and RAPIDS and Spark, but now we have an inferencing service. We have the hyperscalers for the infrastructure provider because a cloud solution is a PaaS solution. We don't manage the infrastructure per se. So we have Amazon, a better partnership, but also we are doing work with them on a series of capabilities, including the new work that they're doing for legacy code migration. Because I do think that a lot of these workflows will need to be migrated. A lot of people will need to go to applications and move from Cobalt to Python, Python to CUDA, and a series of other use cases. So can we accelerate that journey? How can we do that in a secure fashion, at a price point of their choice? So there's a series of things that will come through. But on the application layer, we've chosen to partner with the model providers. So you can access any model on our platform. So whether it's Cohere or Anthropic, whether it's Mistrial, whether it's Fundamental, which I just talked about. So we're trying to provide flexibility and optionality to the customers because that's what customers want. Some people will just use SLMs and do federated learning or knowledge distillation at the edge. Some people will go all the way in and use an API for their favorite LLM, whatever they might like to use. So we wanted to provide that optionality.

John Furrier

>> Yeah.

Abhas Ricky

>> Similarly, on the framework side, we support a series of frameworks, including DaaS and everything else. And on the application side, that's where I think majority of the net new organic revenue streams will come through. So our goal is to be able to help large customers build enterprise AI applications better, faster, cheaper. But there's a caveat. And the big caveat is, what's the moat? So I do think that the moat is private data. So public data, everybody under the sun has five years to Sunday. And by everybody, I mean the large model providers. So the race is about getting access to that private data because you want to be able to train your models with your domain specific context rather than a public API from a model provider. And that's where we've introduced this concept of private AI. Private AI is not about doing AI on private cloud. Private AI is about making sure you can enable AI wherever you like, public cloud, private cloud, the edge, the desktop, whatever you like. But the fundamental reason why we've done that is because, as you know, compute is super expensive. And then until quantum arrives and until the cost of electricity and other energy resources starts to go down or somebody is able to make a breakthrough innovation around that, that compute cost is a big factor and determinant for a large part of the decision making. And if you look at where that compute cost is coming from, it's largely coming from inferencing. That's where the battle frontier is. So private AI collapses the cost curve by orders of magnitude because the cost is just tied to the infrastructure. It's not tied to the usage. So the simplest way to think of private AI is you're not getting any token fees. You're not being charged any tool calling premiums. You're not being charged any long context surcharges. There is no egress cost. And therefore, when you're running models in production in a private AI environment, after a certain point in time, the incremental cost of bringing an additional model in production tends towards zero. Unlike in public environments where it's linear and starts to get exponential after a certain point in time. So that is the core thing whereby I think the future ecosystem will have to drive towards, which is how do we provide feature function parity, but also making sure that the economics of AI is not prohibitive to building enterprise AI applications.

John Furrier

>> Yeah. Abhas, that's a great bet, by the way. And also I think the correct one. In the past month, I've interviewed Matt Garman, CEO of AWS, Jensen Wong, the CEO of NVIDIA, just recently. Both have a similar take with you. One is Amazon clearly says, "Hey, you know what? With the Nova Forge, you can get a frontier model without paying the price." They're basically saying, "Here, here's all the open weights, here's all the checkpoints. Make your data available in a frontier capability," which means is the tokens learn your business. NVIDIA, Jensen talks about specifically domain intelligence, AI factories pumping out intelligence. And you're hitting on is exactly that, which is the private data is the domain raw materials for that intelligence, right? And I like how you called out private cloud isn't private AI because it's hybrid cloud. Now, there's multi-tenancy involved, you can have private environments. This is where I think the agents are really going to shine. So great call out there. I love the private cloud isn't private AI. I think that's super, super important. I have to ask you on that thread, as you see 2026, you talk about an enterprise AI divide. I like that direction because if you believe Matt Garman and Jensen Wong, and what you're saying, there will be new ways to learn or tokenize your business with intelligence. Then that then becomes a value enabler. So this means that the enterprises are going to wake up this year with agents and start to normalize and refactor some of their data platforms to think data centrically around how to use that data. So what is this AI divide? Is that on the same ... Am I getting it right? Explain what you mean by there's an enterprise AI divide.

Abhas Ricky

>> Yeah. Look, I think, John, in my role, I have the luxury of speaking to almost a dozen large enterprises a week, and that's across industries, and that's global customers we're talking about. And that is where I learned the biggest problems our most strategic customers are facing. And you are spot on when you say that this is the year when we think the agentic workflows will turn into production grade workflows. And a lot of the conversation will be, how do I make sure that the AI inside my enterprise can scale? Because the inference costs are continuing to go through as well. So the way I look at it is a typical agentic loop is you have to do some planning, you retrieve, you reason with it, and then that agent acts and it revises, and that's a complete circle and a loop that carries on. But that burns a lot of tokens. Just to give you an example, we were doing a study where we realized that one of the largest financial services customers in the world, they're processing 1.8 billion to 2 billion transactions a day with us. And their goal is to get to a trillion transactions today with us.

John Furrier

>> Wow.

Abhas Ricky

>> And they're burning almost 4,000 to 25,000 tokens on small tasks. These tasks can take between 8 to 20 to 25 LLM calls per task. So if you were to do the math, if you take a public API, pick your favorite hyperscaler, so Azure, OpenAI, AWS, Anthropic, et cetera, this can easily become 50 cents to $2 per task. So if you apply all of the tasks that you're doing, that's hens of millions of dollars a year for one use case. Now, if you compare that on a private AI deployment, and if you choose an alternative model, so for example, if you're choosing a 7 billion or 13 billion parameter model on enterprise GPU clusters, in certain instances, we saw the cost per task drop to less than five cents. In certain cases, dropped to less than 0.5 cents. So the result is you're getting a 20X to 100X cost reduction. You're getting a throughput jump. You're getting predictable spend, so your CIOs and the business owners can plan better for the P&L requirements that you might have. But more importantly, you're getting the full data governance because the data assets are not leaving your data center. The data assets are fully secure. And over a period of time, that security investments that you've been making through finally are helping you in an enterprise gate environment. I'll give you a real example. The first use case that anybody does, any industry, telco, financial services, retail, et cetera, is what? It's a simple customer 360 agent because ever since big data started, I remember when I joined 10 to 12 years back in this company, at Hortonworks and Cloudera, the first thing everybody wanted to do was Customer 360. And that is one of the core use cases that will get, quote unquote, "agentified," if that's a word. And one of the largest insurance companies, they were burning between 20 cents and $2 per query on a public API. And when they started doing that with us, it turned out that it was 20 to 50 times cheaper. So I do think, in this new world, John, a lot of the questions that will be asked is, "Do you have the right metrics? Can you actually quantify your cost per query? Can you quantify how many GPRs per task? Can you quantify what is the margin erosion risk if I stay on my continued path of being on public APIs?" These were questions that weren't necessarily being asked last year or even the year before, because we weren't getting into production environments and a lot of the RAG and the fine-tuning use cases were still experiments that were running in businesses. So I think that's the paradigm shift that will happen as we start to see gazillions of agents across different sets of use cases and functions come through in an industry.

John Furrier

>> I haven't seen this in my career that spans now 30 years in tech. There's always the shiny new toy. There's always the new model here at AI and other cool things. Really a lot of cool stuff happening in AI for sure. But what's interesting this year, and you pointed it out, there's a dual conversation. There's the innovation side and there's the efficiency cost side and things like migration you mentioned. That's not like a top line zeitgeist conversations. Like, one, they're slow as hell, they cost money, they disrupt operations, but not with AI and agents. When you talk about cost, I was talking, again, with AWS and also with NVIDIA, the cost per token per watt is reducing, but also the resource allocation orchestration, meaning I don't need to use the GPUs for certain tasks. I can use another resource. Combination of solid state memory, storage, networking computing, database, all play into those infrastructure things. So you're seeing the parallel theaters of conversations, innovation, which is moving at rapid pace, and then this ROI cost centric efficiency narrative. I've never seen it. Usually when you talk about efficiencies, it's like usually a mature market, you're reigning in the chaos, you're operationalizing things. You got an operational day two operations mindset with full innovation explosion. This is the reality. So what's your take on how that maps to the customer conversations? Are they similar and do they weigh one over the other? Because when you talk governance and sovereignty, it's not just blocking and tackling cost structure issues enabling. What do you see on this? What's your reaction?

Abhas Ricky

>> I think you've touched on two important topics. So let me answer ... And the first one is the new production AI economics, and then the second one is this sovereign pitch, which is exceptionally interesting. So I'll come to the sovereign piece in a minute. But the interesting, John, is if you look at the AI economics, majority of the people, when they think about the economics of enterprise AI, the normal instinct is to focus on training, like a one-time GPU build to build a model. But in reality, once you put the AI agents into production, the real cost drivers continue to shift dramatically. So for example, inferencing, as I mentioned, can be almost 75% of the total cost. The generating the tokens, the powering the multi-step agent workflows that I talked about, if that is taking care of three quarters of the spend, that is what you're trying to optimize for. And that's because agents don't just answer to a single question. They return multiple LLM calls, they do multiple retrieval steps, they do multiple reasoning hops for every task. So as adoption scales, you have these agents in the system, but then there's a continual ongoing token burn. In fact, the second biggest bucket in the economics piece is also not training. It's operational overhead because you have to run vector databases, as you said, we have to have orchestration layers. Somebody needs to take about the caching, the routing, the governance systems. That adds up to almost 10 to 15% of the budget. Now, this is the hidden gravity of the AI economics piece that people are starting to realize and understand. And the infrastructure needed to keep the agents reliable, governed, and compliant at enterprise scale, that will continue to grow. But if you look at the training and the fine-tuning costs, there are largely episodic costs for a majority of the use cases, oftentimes less than 10% combined. So for many of the enterprises, they feel it's a rounding error compared to the continuous spend of inferencing and operations. So I do think that enterprise tokenomics is an inference first operations heavy problem. And if we want to bend the cost curve, the levers to pull around inferencing efficiency or caching or routing to smaller models, as I talked about, quantization and streaming operations, that is something that we need to be able to focus on. And I think that reframes the conversation for majority to large CIOs who now have billions of dollars of budgets, including CFOs actually, whereby you need to understand AI isn't about a CapEx burden only. It is an ongoing OpEx continuum that will continue to grow. And that is where we need to manage the burn because that's where the real value for the engineering lies. So that's the economics piece. But the sovereign thing is also exceptionally interesting because I personally believe AI will continue to get Balkanized if it hasn't already. I mean, if you look at the trend for the model providers, whether the two large ones, OpenAI and Anthropic here in the US or Cohere in Canada, or even Mistral in Europe, I do think that there's a lot of tailwinds that Mistral is benefiting from with the EU narrative and with a lot of the political leaders becoming sensitive to this issue. The same thing's happening in Asia with majority of their own countries, whether it's China, whether it's Singapore, et cetera, starting to drive a lot of the sovereign requirements. And therefore, you will start to see sovereign cloud capabilities come to the fore. And I do think for us, the opportunity is how do you enable customers to be able to get to build an infrastructure whereby the data is not only secured and governed, but also you can provide them and expose that to any agent, any model, or any task with the highest fidelity there is. I think that's the opportunity. Any organization that can expose high fidelity private data/context to these agents, which there will be billions of in any enterprise going forward, in a secure and governed fashion, whilst they're able to align with these sovereign compliance rules, I think that organization wins.

John Furrier

>> Abhas, it's been great to have you on. I love the commentary, love the strategies, love the vision. I think you're right. We are entering in a world where the economics are going to be at front and center now. And I think the common thread that I see is people are saying, target the operation heavy areas. That's where the money is. Certainly knock down some low hanging fruit, but also the revenue enablement is going to be key. Congratulations on a great year. We've been following Cloudera this year. Love the strategy, love the ecosystem. Again, right in line with the two biggest partners, AWS and NVIDIA. We're seeing them lead the way. And then model centric architecture with optionality gives the development of AI native capabilities options to use the best model for the best job, best part of the use case. So congratulations.

Abhas Ricky

>> Well, thank you so much, John. And as I said, we cannot be more excited going into next year, given the fact that we manage 27 exabytes of data/enterprise context, and the opportunity is huge for us going forward.

John Furrier

>> Enterprise data platforms have to be smart. They have to be pumping out tokens. AI factories will do that. Thank you for joining me here in theCUBE and our new NYSE studio for theCUBE. Thanks for coming on.

Abhas Ricky

>> Thank you so much for having me. Happy holidays.

John Furrier

>> Same. I'm John Furrier with theCUBE. This is our AI Factory series. This is where we talk to the leaders who are out there making it happen and solving the problems that are going to bring in this next era of innovation, and also operational efficiency and productivity for people, work, play, society all happening at the same time. We're doing our part here in theCUBE to bring the data to you. Thanks for watching.