Google AI Star Cloudsufi On Building AI Factories And ‘Poison’ Internet Data

Cloudsufi CEO Irfan Khan explains his company’s massive sales growth from building AI and data factories that are winning multimillion-dollar contracts for the Google Cloud partner.

As its head count and sales soar, Cloudsufi is transforming from being a top Google Cloud services provider to being an innovative AI factory builder with customer deal sizes reaching millions of dollars.

“We’re not positioning ourselves as providing traditional services; we’re building AI factories, data factories and control tower frameworks,” Cloudsufi’s CEO, Irfan Khan, told CRN. “These are three areas which are repeatable industrialized AI services, not random customer projects.”

“Some of these are large-ticket items—$5 million to $10 million a pop,” Khan said. “So we’re moving from services to building AI factories.”

[Related: Nvidia Global AI And DGX VP Hired By Google To Lead AI Infrastructure]

San Jose, Calif.-based Cloudsufi is a Google Cloud Premier partner that grew sales 70 percent in 2025 with its sights on continuing to grow revenue by 100 percent in 2026.

The company now has over 1,000 employees around the world with huge plans in store for building AI and data factories for enterprise clients that want to truly unlock the power of artificial intelligence.

Internet Data Is ‘Poison’

Cloudsufi leverages its own IP and Google Cloud AI technology to build unique AI factories and data environments that provide significant AI ROI for customers that doesn’t involve potentially corrupt data from the internet.

“We cannot use internet data—that’s poison,” said Khan.

“Even while you’re training an [AI] engine, if you’re using internet data, your engine will not be able to train well,” he said. “We are not touching internet data.”

Cloudsufi CEO Khan said his global AI and data company aims to transform enterprises into “human-powered by AI” organizations through scalable data platforms, responsible AI systems, and modular factory-based execution.

In an in-depth interview with CRN, Khan explains how his company is winning multimillion-dollar AI factory deals, Cloudsufi’s strategy and why internet data can be “poison” to a client’s AI solution.

Can you explain Cloudsufi’s AI factories strategy that is driving sales and ROI for customers?

A lot of players can just do services, and there’s a lot of AI POCs [proofs of concept] going on right now.

We’re not positioning ourselves as providing traditional services; we’re building AI factories, data factories and control tower frameworks. These are three areas which are repeatable industrialized AI services, not random customer projects.

The trend we are following right now [is] we are moving from AI experiments to AI industrialization. We have a lot of industrialized delivery IP. Every enterprise needs assembly lines for intelligence. That’s our hook there right now.

So we’re building the framework for standardized GenAI deployments, industrialized model life-cycle management that governs prompt pipelines, tracks ROI at a use case level, and then enforces enterprise grade-guardrails in the beginning.

So customers normally sign the project for a year or two years.

So it’s like a Lego block which comes together, but these Lego blocks have been created so far that we don’t have to really rebuild it again all the time.

How is Cloudsufi winning large enterprise deals?

If I break my revenue into Cloudsufi AI and platform engineering—70 percent of our growth came from Cloudsufi AI.

The reason is because the deal sizes are big. The reason being is a lot of pieces used to be separate boxes, which doesn’t make sense. For example, somebody doing connectors would say, ‘Oh, I need five connectors or 15 connectors.’ That conversation has now moved to, ‘I need 100 connectors.’

Because they realize that if they really have to grow on the AI side of it, then they need to also make sure they have foundation blocks.

So because of that, the size of the deal ask is no longer, ‘OK, let me just get two connectors and figure out the rest.’

They’re saying, ‘OK, take the top 50 enterprise software [platforms] that are highly used, and pick up five of each vertical and give me a connector factory for it.’ So then the deal size moves up now because we’re creating a factory.

So not only are deal sizes increasing, but enterprises’ AI appetite is an ongoing demand?

Because customers know they have to go through the AI journey, they also know that what they start with is not what they’re going to end up with.

It’s a process. So we have an assessment, and there is a workshop. There is this whole Santa Claus list which needs to be created in milestones.

What’s happening right now is instead of creating their own software factory, they come into us saying, ‘OK, there’s a hybrid of innovation, R&D and creating a production-ready system at the same time happening.’ So conventionally, they would say, ‘This is a Statement of Work. And yes, these are three milestones.’ Milestones are very directional now.

The customer is committing to us and paying quarterly in advance.

The beauty of this is the tech is so new that the demand for the customers keep changing as they get access. … They want more speed. They want more.

We have a customer who keeps calling us saying, ‘Look, do you think we can do more? Should I invest more in the AI factory?’ He’s only investing $10 million a year. Because for him, the ROI was so high, he’s like, ‘Can I put more money into this box, which is giving me insights?’

So with that trend, each of the deal sizes are going up right now. … These are large-ticket items—$5 million to $10 million a pop.

What’s the biggest mistake businesses are doing right now with AI?

The biggest mistake people are doing is they go, ‘We’re going to get data.’ The most expensive piece is data. So we’re building a data factory, which means we are buying data.

We are buying data and the raw data, then that’s their IP. We cannot use the internet data—that’s poison.

For example, we have a use case with [oil giant] Aramco.

So every time a turbine goes out in Aramco, they’re losing $100 million a week. A week—$100 million—gone. So they’re not able to do predictions.

The only way they still predict the turbines is by people who’ve been there for 30 years by hearing the sound. They can hear the turbine saying, ‘There’s a problem.’

We are now creating AI machinery to predict the predictability of these turbines—which are like the size of buildings—by sound and vibration. So we have data factory collecting all that data.

Another example, [aerospace manufacturing company] Boeing has tons of data. They’re more than happy to share their raw data coming in.

So for us, the bigger piece we’re doing right now is this: moving from service to AI factory.

So you’re not using internet data for your AI factories. Why did you call internet data ‘poison?’

Because with the AI coming, you realize that every data [point] of the internet is so biased by people who also use it.

So if you’re training an engine and you’re using internet data, your engine will not be able to train well.

So a part of it is collecting raw data through credible sources. And that data needs to be captured in a very separate box.

All the tabs of the engine to suck data from the internet have been switched off. We’re not touching internet data.

We’ve made these mistakes before.

For example, when a customer said, ‘Oh, ChatGPT is doing this. Our guys are not doing it right now.’ Our team hurried up, saying, ‘Let’s pull some data out there.’ It will solve the problem for that moment, but the engine loses the credibility over a period of time. The engine is not pure to give you what you need.

A lot of people are privately owning their data now.

They are going to move away from the Grok AI and the ChatGPTs model right now.

They might use models which have free source code to create their own models through that, but they’re not going to be using internet data.

Can you explain the AI and data factories you’re building for customers?

So AI factory consists of—just like a traditional factory—there are pods sitting there consisting of six to seven AI engineers, a data architect, etc. so it’s actually a physical factory.

We’re creating these use cases then moving into production.

So it goes into a design team. The design team takes into an architecture team. The architecture team takes it into a prototyping team.

It’s a full-blown supply chain in the future.

The data teams come into picture to make sure the data factory starts right there and then. They all keep handing over their stuff to each other in terms of getting it done.

That’s where I feel the future will be, from our perspective, going and building it out. So for us, service to factory is important.

For the data factory, plus the control tower architecture—control tower used to be very synonymous with supply chain in the past.

So what we are seeing is most of the enterprise struggle, not with AI, but with many data ecosystems. And that is where we have an advantage because our company has started out looking to reduce the gap between human intuition and data-backed decisions.

Data injection, ingestion, integration, and building the accelerators for that is what we have been doing for a lot of companies—to rebuild enterprise connectors, connector factories, reusable transformation pipelines, or control tower observability layers to see how easy it is to connect. … We are very bullish about our future.