'In two years of 100% growth, we added a net total of just four people,' says Umesh Sachdev, CEO, Uniphore

Umesh Sachdev, CEO, Uniphore

In a rare moment of alignment in the fiercely competitive AI ecosystem, Uniphore has pulled off what few enterprise startups can, bringing NVIDIA, AMD, Snowflake and Databricks into the same funding round. At a time when each of these giants is battling for dominance across GPUs, data layers and LLM infrastructure, their simultaneous backing of a single company signals an unusual degree of conviction in Uniphore’s sovereign, open and enterprise-focused AI platform. The firm raised $260 million in a Series F round from these tech giants at a valuation of $2.5 billion.

This milestone caps a journey that began much before generative AI became the industry’s obsession.

Incubated at IIT Madras, Uniphore spent over 16 years investing deeply across knowledge AI, emotion AI and conversational AI, giving it a first-mover advantage when enterprises worldwide began rethinking their automation and data strategies. Today, the company is valued at more than $3 billion, works with over 2,000 large enterprise customers across 20 countries, and continues to expand its footprint as global demand for AI-driven transformation accelerates.

In an interview with Moneycontrol, CEO Umesh Sachdev explains how the latest round came together, why Uniphore is doubling revenue year-on-year, how it has achieved this scale while adding a net headcount of just four people in two years, and what makes its Business AI Cloud central to the next phase of enterprise adoption.

Edited excerpts:

The backstory is this: we went out to raise capital because the new products we rolled out, specifically our Business AI Cloud, triggered an inflection in the business. We grew roughly 100% last year, and this year we are on track to register nearly the same growth. The platform, a complete end-to-end data and agentic system, is now used by some of the largest enterprises in the world. To keep fueling that momentum, we needed capital.

We approached the usual financial VCs, and the round was led by NEA. But as we began raising several months ago, we found the VC community was confused by the rapid shifts in AI. LLMs, new startups, hyperscalers, everyone was making announcements. The feedback was: your revenue and customers are great, but who will win this space? Will LLMs win everything? Will hyperscalers win everything? Will startups still have a role? They didn’t know.

Even though we were clear about our position, with over 2,000 large enterprises, including the number one bank, telco, and insurer, it wasn’t enough for me alone to tell the story. I realized someone with more credibility had to validate it. That’s when we reached out to NVIDIA, a deep partner of five years, and they participated to validate our position.

Around the same time, Snowflake, another long-time partner, also wanted to join. We are an open platform, neutral to any LLM, GPU, or data layer, so having one compute provider and one data provider made sense.

News travels fast in Silicon Valley, and AMD soon began conversations as well. John Chambers on my board has long-standing relationships with Lisa Su and leaders at AMD. Finally, Databricks came in through a common VC.

It wasn’t that we went after competitors. We wanted one thing, validation of our position. And I think that objective has been achieved.

Let’s take an example. With over 2,000 large enterprises, agentic AI adoption isn’t widespread yet, but it’s now very deep. Each company has picked a few areas, and by 2026 we’ll be ready for broad agentification.

If you’re a large insurer, you don’t want to think about GPUs or LLMs, you care about a claims process that takes 27 minutes and four people. The main issue is that claims data sits in four systems. Without connecting that data, agentic AI won’t work. Once it’s connected, a general LLM may hallucinate, so you use a small language model fine-tuned on claims.

Then you build AI agents to mimic human tasks, connect them to the fine-tuned model, and the model to your data. That gives you accurate, predictable, agentified claims processing.

The Uniphore platform does exactly this: a data layer, a fine-tuning layer for small language models, and an agentic layer for end-to-end workflows. Our inferencing layer connects to NVIDIA or AMD, and we integrate with data sources like Snowflake, Databricks, BigQuery, Informatica, and Palantir. We fine-tune the foundational model you choose—Gemini, OpenAI, Llama—while keeping everything sovereign and open.

We sit at the tip of the spear of enterprise AI adoption, which is why NVIDIA, AMD, Snowflake, and Databricks have invested in us, because if we do this job well, GPUs, data, and LLMs get consumed.

Two pricing models are emerging. The per-user SaaS model doesn’t apply here.

Uniphore uses a consumption model. We sell Uniphore credits. Each credit can be used for a certain number of queries, data, or agentic actions.

The second model, which partners like KPMG use, is value-based pricing. KPMG uses our platform to build fine-tuned SLMs and agents for industries. They then charge clients based on outcomes, for example, if they save a client $100 million, they keep a portion of that. They consume our platform purely on a credit model.

So it’s either consumption-based or outcome-based. We stay away from outcome-based. But our partners buy credits from us and offer outcome-based pricing to their clients.

Publicly, we’ve said we have crossed $200 million in ARR this year. Last year, we crossed half a billion in TCB, and we think that number is doubling this year. Our fiscal year ends April 30th. So yes, we will likely end the current fiscal with $1 billion of TCB. ARR is already above $200 million with two quarters to go.

We are still negative EBITDA because we’re growing at 100%. So we focus on gross margin as the proxy. Our gross margin is upwards of 70%, and we think it will go up further.

There isn’t much competition for an end-to-end sovereign and open platform, so we have pricing leverage. And we use our own platform heavily. In two years of 100% growth, our net headcount increased by only four people, even though we acquired four companies during this period.

Yes, that phenomenon is happening. In SaaS, vertical SaaS was the norm—Salesforce for sales, Workday for HR, Freshworks for service.

In AI, especially for large enterprises, there is no vertical AI. Business AI is horizontal. You connect data, fine-tune models, build agents, the recipe is the same across departments.

Large companies no longer buy point solutions. That shift is what’s happening.

We’ve done four acquisitions in the last year, two data companies and two pure AI companies. Not one was vertical SaaS.

We are an end-to-end data and AI platform. When we enter a bank, insurer, telco, or consulting firm, we go wide across the enterprise. We also see gaps in their data, model, or agentic architecture. Many startups have good tech but no go-to-market engine. We can deploy their innovation to 2,000 customers quickly. So we will continue deep tech acquisitions, not vertical SaaS.

No. Half our headcount is in India, but less than 5% of revenue is from India. So shifting HQ doesn’t make sense.

We have two offices in India, Bangalore and Chennai, almost equal in headcount. We will keep growing these two. We won’t open new locations; spreading out too much is an overhead.

Very important. Continental Europe is our second-largest region. We count UK revenue as part of Europe. About 65% is North America, 23% is Europe, and the rest is Asia, mainly Middle East.

I’m talking about enterprise adoption, not developer or consumer use. Yes, tools like OpenAI and Cursor have big user bases in India. But large corporations in India are not yet agentifying processes or fine-tuning models as much as the US or Europe.

Even Europe is a year behind the US. Asia varies by country. It will happen, but today the rate is different.

The AI market is forking into two paths: AGI (OpenAI, xAI) and fit-for-purpose AI (Anthropic for coding, Perplexity for search, Uniphore for enterprise). Both sides have real revenues, growth, and margins. Are valuations overheated? Likely. But it’s nowhere close to the 2000 dotcom bubble.  This time, companies have real revenue, OpenAI is at $20B, Anthropic at $5–7B. Enterprise adoption is real. The big proof points will come in 2026 when public companies start reporting earnings gains from AI.

Very likely next year will be similar growth. The inflection is because enterprises want an end-to-end sovereign and open platform. Half our revenue next year will come from existing customers expanding.

We’ve now raised $850 million. At some point, liquidity must be created for investors and employees, so going public is on our mind. But we haven’t started the clock yet. Usually preparing for an IPO takes 18–20 months. We’re not there yet.

This is also a unique time; GenAI is evolving fast. Sometimes it’s better to stay private while the underlying technology shifts weekly.

Yes. Even if we have one Indian customer, it affects us. And globally, every region has its own AI regulations. As a global company, we cannot cherry-pick. If an Indian uses our AI in UAE, we’re liable under DPDP. If a European uses it elsewhere, we’re liable under EU rules. Regulations will likely slow down innovation.

No. We started Series F with a goal of 200, and with the strategics joining, we crossed that. This is more than sufficient for the next two to three years. We’re focused on customers now.

No. I wouldn’t change anything. I’ve had so much fun. We’ve gone through good times, tough times, and learning times. From IIT Madras to selling to the top banks and telcos, that journey hasn’t been easy, but we’ve cracked it.

If we started Uniphore today, the pace would be much faster given the capital and technology available now. It wouldn’t take 17 years.