Zoho built a server. In-house, over five years, in Nagpur. That is a different kind of infrastructure investment than anything most enterprise software companies have attempted.
Zoho has deployed 1,000 Nathu La servers in its Indian data centers today. The vertical integration strategy announced at ZohoDay 2026 is now running production workloads. For enterprise buyers evaluating long-term AI pricing stability, this is the architecture that makes Zoho's cost commitments possible.
Five years of hardware research in Nagpur became a public product today. Zoho named the platform Nathu La: a server designed in-house, built around Intel Xeon 6 processors, currently running in Zoho's own Indian data centers. I am covering this from Info-Tech LIVE in Las Vegas, where both Zoho and ManageEngine are on the expo floor — a reminder that this is a company increasingly present where enterprise IT decisions get made. At ZohoDay 2026 in February, Ramprakash Ramamoorthy, Director of AI Research at Zoho and ManageEngine, had described the hardware initiative as a "kitchen tour," a preview of something not yet shipping. I covered that session and noted the target of 10,000 custom units by 2026. That target is now a deployment.
The name is not incidental. Nathu La is a high-altitude mountain pass in Sikkim where Indian Army soldiers held their ground against Chinese forces in 1967, in one of the most significant border engagements in post-independence Indian military history. Naming a sovereign infrastructure platform after that pass signals how Zoho frames this investment: not as a cost-reduction exercise, but as a statement about technological independence. For a company that has spent thirty years building its own data centers, its own models, and now its own servers, the name fits the ambition.
The announcement matters because it arrives in an environment where inference cost is no longer a startup budget line. For Zoho, which serves more than 150 million users across a suite of over 55 applications and has committed to bundled AI pricing that does not pass token variability to customers, the cost of running inference at scale is a strategic exposure. Zoho has promised flat AI pricing to customers. Whether it can keep that promise depends on controlling the cost of running AI internally.
The processor at the center of the platform is an Intel Xeon 6, imported. Ramamoorthy was precise about this when speaking to Business Today this morning: "The base chip is imported, but there is a lot of deltas to it." The deltas are the real story. Zoho designed the motherboard, the chassis, the firmware, the power delivery subsystems, the DC-SCM, or Data Centre Secure Control Module, and the NIC, or Network Interface Card. The intellectual property sits in the architecture built around the silicon.
That matters because Zoho controls the cost variables that sit above the processor: how the server draws power, how workloads are scheduled, how the firmware is updated. Those are the layers where inference cost is won or lost at scale, and Zoho owns all of them.
Five-plus patents have been filed covering thermal management and cost-optimized server architecture, according to the company. Over 300 engineering students have contributed through SETU, which stands for Student's Engagement for Transformative Upskilling, a skilling program aligned with India's Ministry of Electronics and Information Technology's ESDM, or Electronics System Design and Manufacturing, policy. Both signal a structural commitment to maintaining this capability in-house rather than purchasing it from a contract manufacturer. Zoho founder Sridhar Vembu has spoken publicly about interest in deeper chip investment, and Nathu La is the kind of platform that makes that next step sensible rather than speculative.
Zoho has been explicit about the commercial logic since ZohoDay 2026. The company uses a "right-sized model" approach, deploying smaller contextual models for specific tasks rather than routing every workload through a large generalized model. That approach, combined with owning the serving infrastructure, is how Zoho sustains flat AI pricing for customers even as inference compute costs rise across the industry.
Ramamoorthy put the pressure point plainly in a prior interview: "With infrastructure and compute already among the biggest expenses after human resources, if compute costs suddenly double, you cannot simply raise customer pricing by 30 to 40 percent overnight." Nathu La is the architectural response to that constraint. By optimizing the full stack, hardware through model serving, Zoho intends to absorb cost pressure in the layers it owns rather than passing it forward.
The 85-plus percent figure for customers preferring Zoho's bundled AI capabilities is one Zoho cites. Zoho has been integrating its AI platform, Zia, across its product suite since before the generative AI period. The launch of the company's own large language model, ZLLM, last year, now deployed across the product suite with 32-billion and 100-billion parameter models in development, means the serving volume on Zoho-controlled infrastructure is growing regardless of what the exact attachment rate is today.
The scale number is specific and important. Zoho is not announcing a prototype or a pilot. One thousand servers are running in Indian data centers today, across a global estate of 20 data centers. The company has not set a timeline for global deployment, and Ramamoorthy was careful to note that OEM servers will coexist with Nathu La rather than be replaced by it. "We have built this very purpose-built for our own use cases," he said. "We know the use case, we narrow it down, we fine-tune it."
The coexistence of Nathu La and OEM servers is a sign of operational precision, not hedging. Zoho is deploying its custom platform where it has the highest impact, on AI inference workloads where the same model runs millions of times against predictable input patterns. Purpose-built hardware optimized for that specific load delivers a genuine advantage. General enterprise workloads will migrate as the platform matures.
The platform is built on the Open Compute Project design philosophy, referred to as OCP, which emphasizes modularity and ease of maintenance. That lineage matters for enterprise customers thinking about long-term serviceability. OCP-aligned hardware can be maintained, expanded, and partially sourced from a broader ecosystem of compatible vendors, which reduces the operational risk of committing to a proprietary platform built by a software company.
Zoho owns the design and integration layer of its server infrastructure. For enterprise buyers, that means the 20-30% total cost of ownership reduction Zoho claims is tied to architecture the company controls and can improve. As the company adds 32-billion and 100-billion parameter models to its serving infrastructure, having that hardware purpose-built for the workload is a genuine advantage.
Buried in today's coverage is the detail that carries the most weight for CIOs thinking beyond Zoho's current customer base. Ramamoorthy identified a potential next step: preloading Nathu La servers with Zoho software and offering them to enterprise and government customers. No timeline exists. But the architecture this would require, a software-defined appliance running Zoho's full application and AI stack on Zoho-designed hardware, is the logical extension of everything the company has built over the past five years.
That is not a hardware sales play. It is an on-premises deployment model for regulated industries and data-sovereign governments that cannot send workloads to a public cloud. It is also, if it materializes, a different kind of competitive surface than anything Zoho currently offers.
Zoho has been building toward this for thirty years, owning its data centers, writing its own models, and now designing its own servers. Nathu La is not a detour from the software business. It is what allows the software business to make pricing commitments that vertically dependent competitors cannot. For enterprise buyers, the relevant question is how fast Zoho scales this across its 20 global data centers, and what that expansion means for the inference cost gap between Zoho and vendors still paying hyperscaler margins.
Zoho has been investing in full-stack infrastructure while most SaaS vendors were outsourcing it. Nathu La is five years of that investment becoming a production deployment. The question for enterprise buyers is not whether this works today, it clearly does, but how quickly Zoho can expand Nathu La across its 20 global data centers and whether the inference cost advantage compounds as model sizes grow.
- Zoho Corporation. "Zoho Corporation Unveils Nathu La, a Designed-in-House Server." Business Wire, 10 June 2026. businesswire.com
- Agarwal, Palak. "Zoho Unveils India-Designed Nathu La Server, Deepens Push to Own AI and Cloud Infrastructure Stack." Business Today, 10 June 2026. businesstoday.in
- Ramamoorthy, Ramprakash. "Foundations that Scale." ZohoDay 2026 Session, Feb. 2026. zoho.com
- Bellamkonda, Shashi. "Building From the Ground Up: Zoho's Bet Against AI Infrastructure Dependency." shashi.co, 21 Feb. 2026. shashi.co
- Open Compute Project Foundation. opencompute.org
