Enterprise data infrastructure was messy long before generative AI entered the picture. Most large organizations have accumulated warehouses, catalogs, governance tools, transformation engines, orchestration platforms, and observability products, yet many still struggle to turn all that data into something the business can use quickly. AI has only widened the gap. Giving a model access to more data does not mean it understands what that data means, whether it can be trusted, or how it is allowed to be used.
That is the problem The Modern Data Company set out to solve with DataOS, an AI-native Data Operating System that sits across a company’s existing infrastructure and turns distributed data into governed, context-rich data products. Those products can then be reused by AI, applications, and analytics, without a rip-and-replace migration or rebuilding the same context for every new use case. In this VMblog Q&A, co-founder and CTO Animesh Kumar explains how the platform grew out of early design-partner work with Gap, a Fortune 500 company with a capable data science team but only a couple of models in production.
Kumar also shares how the company measures success, including initial production use cases delivered in four to six weeks and a customer that offset the full cost of implementing DataOS with a single month of Snowflake savings. He discusses where he sees growth heading, from financial services and healthcare to manufacturing, retail, and the public sector, and names market noise and organizational inertia as the company’s biggest threats. And, in a distinctly personal answer, he reveals that if he weren’t building a startup, he would be writing about the history of Indian mathematics.
++
VMblog: What is your company in 2–5 words?
Animesh Kumar: Creator of an AI-native Data Operating System
VMblog: Why is now the time for your company to exist?
Kumar: Enterprise data infrastructure was already fragmented before generative AI arrived. Companies accumulated warehouses, catalogs, governance tools, transformation tools, orchestration tools, and observability platforms, but still struggled to turn their data into something the business could use quickly.
AI has made that gap impossible to ignore. Giving a model access to more enterprise data does not mean it understands what that data means, whether it can be trusted, or how it is permitted to use it. That context is often scattered across systems, teams, policies, and undocumented institutional knowledge.
We built DataOS to provide the shared data layer that was missing. It works across a company’s existing infrastructure and turns distributed data into governed, context-rich data products that can be reused by AI, applications, and analytics. Companies do not need to replace the technology they already own or rebuild the same context for every new use case.
AI has raised the stakes, but the underlying need is larger: enterprises need a more productive way to use their data.
VMblog: What do you love about your team, and why are you the ones to solve this problem?
Kumar: Our team combines deep technical knowledge with firsthand experience of how difficult it is to turn enterprise data into something the business can use. We have built data platforms, led engineering and product teams, and worked directly with large organizations managing complex data and tech environments.
What I value most is that we do not treat this as a collection of isolated technical problems. We understand that architecture, governance, organizational structure, and business priorities all determine whether data ever becomes useful.
That perspective shaped DataOS. Instead of adding another point solution, we created an operating system that works across the existing stack, manages data like software, and makes governed, reusable data products available to AI, applications, and analytics.
We also built DataOS in response to real enterprise challenges. Our early work with Gap showed us what happens when valuable data is spread across disconnected systems and teams cannot use it quickly, consistently, or safely. That experience helped shape the platform from the beginning.
VMblog: If you weren’t building your startup, what would you be doing?
Kumar: I would write about the history of mathematics. Probably the Indian part. Most people skip past it. I like working out how someone reached an idea. Back then, nobody knew yet if it was right.
There is a book from around the year 628. Brahmagupta wrote it. It is the first one we know of that treats zero as a real number. A number you can add, subtract, and multiply. He explains the rules using money. A debt is money you owe. A fortune is money you have. That is a nice way to make it clear. Then he asks what zero divided by zero is. He says the answer is zero. That is wrong. Five hundred years later, Bhaskara argued back. He said dividing by zero gives something endless, not zero. He did not settle it either.
That is the part I like. A man works with no map. He gets most things right. He gets one thing wrong, and nobody can see it yet. Later, we look back and it all seems simple. A neat list of discoveries. It was never neat.
Honestly, I have the same job now. I make choices without knowing which one is wrong. The mess people deal with is mostly old wrong choices. Nobody went back to fix them. My work is finding the simple answer. The one that should have been clear from the start.
VMblog: At the moment, how do you measure success? What are your metrics?
Kumar: We measure success by how quickly customers put trusted data to work and what changes for the business once they do.
Time to production is one important measure. Enterprise data programs have historically taken quarters or even years to show value. With DataOS, we typically put an initial use case into production in four to six weeks.
We also look at the progression from proof of value to a commercial relationship. That tells us customers are seeing enough measurable value to continue rather than treating the work as another experiment.
After deployment, the measures depend on the customer and use case. We track reductions in delivery time and infrastructure costs, improvements in data quality, adoption of reusable data products, and the number of analytics and AI use cases that reach production.
Ultimately, success is not how much data a company stores or how many tools it owns. It is how productively that data can be used to create a business outcome.
VMblog: In a few sentences, what do you offer and to whom?
Kumar: DataOS gives enterprises a shared operating layer for their data. It connects to the infrastructure they already have and turns fragmented data into governed, reusable, context-rich data products for AI, applications, and analytics.
We work with data and technology leaders who have invested heavily in infrastructure but still cannot deliver trusted data to the business quickly enough. DataOS allows them to get more value from those investments without another large migration or rip-and-replace program.
The result is a shorter path from a business question to a production-ready answer.
VMblog: What’s most exciting about your traction to date?
Kumar: The most meaningful traction is what happens after customers see DataOS working with their own data. Customers who start with a single use case consistently come back to expand it, adding new data domains, more business teams, and additional analytics and AI use cases onto the same operating layer, rather than treating DataOS as a one-off project.
That expansion matters because enterprise data buyers have seen plenty of impressive demonstrations. They are looking for evidence that a platform can work within the complexity of their actual organization. We demonstrate that by starting with a real business outcome, using their existing data and infrastructure, and delivering an initial production use case in four to six weeks.
The economic impact can also be substantial. One customer saved enough on Snowflake in a single month to offset the full cost of implementing DataOS.
We are also seeing the market catch up with several ideas that shaped DataOS from the beginning: treating data as a product, managing it like software, and adding a common operating layer across fragmented infrastructure. What once required a long architectural explanation is increasingly becoming an urgent business conversation.
VMblog: Where do you think your growth will be next year?
Kumar: Our growth will come from enterprises moving beyond AI experimentation and confronting the harder question: How do we put AI into production with data that is trusted, contextualized, governed, and economical to use?
That need is especially pronounced in industries where decisions must be explainable, and the cost of inconsistent data is high. Financial services, healthcare and life sciences, manufacturing, retail and distribution, and the public sector are all important areas for us.
We also expect growth from organizations that do not want to replace their existing infrastructure. Most enterprises have already invested heavily in warehouses, lakehouses, catalogs, and cloud platforms. They need a way to make those systems work together and support new AI and application requirements.
DataOS gives them that operating layer. The opportunity is not limited to one AI project. Once the layer is in place, it can support a growing number of data, analytics, application, and agentic use cases across the enterprise.
VMblog: Tell us about your first paying customer and revenue expectations over the next year.
Kumar: The idea behind DataOS became concrete through our early work with Gap.
Gap was a Fortune 500 company generating more than $16 billion in revenue at the time. It had a highly capable team of data scientists, yet only a couple of models had reached production.
The conventional answer was another large transformation program. We believed the answer was a different architecture: an operating layer that could work across existing infrastructure, apply governance consistently, and package data with its logic, policies, quality standards, and context.
Gap became our first major design partner. We then worked with other large enterprises facing different versions of the same underlying issue: valuable data existed, but the organizations could not put it to work with enough speed, consistency, or control. Those early partnerships helped us design DataOS for the realities of complex enterprises.
Over the next year, we expect growth to come from expanding within existing customers, adding enterprise accounts in our priority industries, and supporting the movement of more AI initiatives from pilot to production.
VMblog: What’s your biggest threat?
Kumar: Our biggest threat is not a single competitor. It is the combination of market noise and organizational inertia.
Almost every data platform now describes itself as AI-ready. Catalogs are becoming governance platforms, governance tools are becoming data management platforms, and infrastructure companies are adding overlapping capabilities. That makes it difficult for buyers to distinguish between adding another feature and changing how enterprise data is managed.
At the same time, the people responsible for data and AI are often working through reorganizations, competing mandates, and unclear ownership. A company may understand the technical need but still struggle to align the teams and budgets required to act.
The risk is that enterprises respond to the urgency around AI by buying more disconnected technology or delaying the architectural decisions that production AI requires. Our job is to make the difference concrete: start with an important business outcome, prove it using the customer’s own environment, and show value in weeks rather than asking the organization to believe in another multiyear transformation.
