5 days ago

Vigyanlabs Unveils FEMTO AI-in-a-Box for Sovereign, Energy-Efficient AI Computing

Mysuru-based deep-tech company Vigyanlabs has unveiled FEMTO, an indigenous AI-in-a-Box platform designed to bring high-performance artificial intelligence computing directly to enterprises, government organisations and institutions.

The platform was formally launched in Mysuru on September 5, with former ISRO Chairman Dr S. Somanath attending as chief guest. The system is being developed and deployed through Kasvin, a joint venture between Krishnaa Advance Systems (KAS), a subsidiary of Krishna Buildestates, and Vigyanlabs.

FEMTO is designed as a fully integrated AI infrastructure appliance combining computing, storage, networking, virtualisation, orchestration and power management in a compact unit. The platform is intended for on-premises and air-gapped deployments, allowing organisations to process sensitive AI workloads within their own infrastructure rather than relying entirely on external cloud platforms.

FEMTO-4 Targets High-Density AI Computing

The highest-performance configuration, FEMTO-4, is designed to support AI models exceeding 400 billion parameters, with eight GPUs and 768 GB of GPU VRAM. Company materials state that the system operates at less than 5.5 kW and uses air cooling rather than liquid cooling.

Vigyanlabs has also reported performance tests involving a two-GPU FEMTO configuration running a 70-billion-parameter Llama model, achieving 3,862 output tokens per second at 512 concurrent users, with approximately 4.8 output tokens per second per watt of GPU power. These figures are company-reported benchmark results rather than independently verified performance claims.

The platform incorporates Virtune, Vigyanlabs’ orchestration and virtualisation technology, along with its IPM+ intelligent power-management technology. The system is designed to support a range of AI workloads, including large language models, multimodal models and embedding models.

Focus on Sovereign AI

A central objective of FEMTO is to enable sovereign and private AI computing. By deploying the infrastructure on an organisation’s premises, sensitive data and AI workloads can remain within its own physical and administrative environment.

This approach could be particularly relevant for sectors such as government, defence, healthcare, financial services and manufacturing, where data residency, security and operational control can be important considerations.

Vigyanlabs has positioned FEMTO as an alternative to relying exclusively on large centralised cloud data centres, particularly for organisations that require local or air-gapped AI processing.

Waterless Cooling and Energy Efficiency

Another focus of the platform is reducing the energy and water requirements associated with AI infrastructure. FEMTO uses air cooling, eliminating dependence on liquid cooling infrastructure.

The company says its architecture can achieve a Power Usage Effectiveness (PUE) below 1.3 with zero water usage. Vigyanlabs and its partners have also claimed that the platform can reduce energy consumption by approximately 50%, although these savings depend on deployment configuration and workload and should be viewed as company-reported targets.

The FEMTO initiative is backed by a planned ₹100 crore investment over the coming year, according to KAS and Vigyanlabs. The companies intend to use the investment to expand deployment and scale the platform across India.

One of the early applications announced alongside the launch is an AI data-centre deployment in Gundlupet, Chamarajanagar district, aimed at bringing high-performance computing infrastructure into a rural setting.

The launch places FEMTO within India’s broader push toward domestically controlled AI infrastructure, combining high-density computing with local data processing, energy efficiency and reduced dependence on external cloud infrastructure.