Zoho has launched an in-house server platform called ‘Nathu La’ to reduce AI inference costs and strengthen control over its hardware and software stack. The company says the server is designed to deliver 20% to 30% lower total cost of ownership and up to 18 percent lower power consumption, which it expects will improve operational efficiency across its business.
Zoho launches Nathu La to cut AI inference costs
The server was developed over five years by Zoho’s internal R&D teams and manufactured with Indian electronic manufacturing partners. Built around Intel Xeon 6 processors, it includes in-house engineered motherboards, DC SC modules, and network cards, with Zoho saying it also reduces exposure to foreign licensing dependencies and external security audits.
Zoho has also filed five patents related to Nathu La, covering thermal management and modular server architecture. The company plans to use the servers across its application and SaaS infrastructure for AI inference, virtualisation, high-performance computing and storage, while improving data governance for global customers.
What the move means for Zoho’s wider AI strategy
Zoho said the server push is part of its effort to lower inference costs for its own AI usage as enterprise adoption of AI pushes up spending. The company has already deployed a few hundred Nathu La servers, with 1,000 in production and pre-production, and a target of 2,000 by year’s end, according to reporting cited by the company.
Shailesh Davey, Zoho Corporation Chief Executive Officer (CEO), said, “With Zoho’s strategy of using contextual, right-sized models, running on our own platform, now on our own servers, accelerated by our own GPU database, we are compounding the benefits accrued from owning and operating our entire technology stack.”
