Mysuru Deep-Tech Firm Vigyanlabs Launches Waterless FEMTO Sovereign AI Platform

Mysuru-based deep-tech company Vigyanlabs Innovations has launched FEMTO, a compact sovereign artificial-intelligence infrastructure platform designed to allow enterprises, government institutions and other organisations to run powerful AI workloads within their own premises instead of depending entirely on external cloud data centres.

FEMTO was formally unveiled in Mysuru on September 5, 2026, with former ISRO Chairman Dr S. Somanath attending the launch as chief guest. The system combines high-performance AI computing, storage, networking, virtualisation, orchestration and power management inside an integrated platform that Vigyanlabs describes as an AI-in-a-Box.

One of FEMTO’s most unusual features is its approach to cooling. Large AI computing installations increasingly rely on liquid-cooling systems because modern high-performance processors generate substantial amounts of heat. FEMTO has instead been engineered to operate using conventional air cooling, removing the need for dedicated water-based cooling infrastructure at the appliance level.

According to Vigyanlabs and its project partners, even the highest-performance FEMTO configuration consumes less than 6 kW of electrical power, making it possible to cool the unit using air despite the demanding computing workloads it is designed to support. The company argues that this could make AI infrastructure easier to deploy in locations where large liquid-cooled data centres are impractical or undesirable.

Water consumption has become an increasingly important issue for the global data-centre industry as computing requirements rise. Large facilities can require considerable quantities of water for cooling, particularly in hot climates. By designing FEMTO around an air-cooled architecture, Vigyanlabs is attempting to offer an alternative for organisations that want significant AI computing capacity without constructing specialised water-intensive cooling systems.

The company also says FEMTO can reduce overall energy consumption by around 50% compared with conventional infrastructure configurations, although this remains a company-reported figure and actual savings will depend on workload, hardware utilisation and deployment environment. Vigyanlabs has spent several years developing power-management technology aimed at reducing unused computing capacity and improving data-centre efficiency.

FEMTO integrates this capability through Vigyanlabs’ proprietary technology stack. The platform incorporates Virtune, the company’s AI-oriented virtualisation and orchestration system, alongside its IPM+ intelligent power-management technology. Together, these systems are intended to dynamically manage computing resources and energy consumption while providing organisations with a private AI environment.

Virtune provides the software layer responsible for managing virtual machines, containers, AI workloads and computing resources. Vigyanlabs describes the platform as hardware-agnostic across standard x86 infrastructure and designed to provide an integrated environment for AI operations without requiring multiple independent management systems.

The result is effectively a compact micro data centre capable of hosting artificial-intelligence workloads locally. FEMTO is delivered with the computing, storage, networking, virtualisation and AI infrastructure software already installed and configured. Organisations can therefore rack the appliance, connect it to power and their network, and begin configuring AI workloads without constructing a conventional AI data centre from the ground up.

The platform can also function in air-gapped environments, where the computing infrastructure is deliberately separated from the public internet. This is particularly significant for organisations handling sensitive or classified information because AI models and organisational data can remain inside the user’s premises rather than being transferred to external cloud servers.

Vigyanlabs is consequently positioning FEMTO around the growing concept of sovereign AI. The term generally refers to an organisation or country’s ability to control the infrastructure, models and data underpinning its artificial-intelligence systems rather than becoming completely dependent on foreign cloud services or externally controlled computing platforms.

The company says the complete FEMTO software stack has been developed in India, while more than 50% of the overall platform consists of indigenous content. Technologies associated with the system are also protected through patents filed or granted in India and the United States, according to the project partners.

FEMTO has been designed to support a broad range of established open AI models, including Llama, Mistral, Gemma, Qwen, Phi and GPT-OSS, together with multimodal and embedding models. This allows organisations to select models appropriate for their applications while keeping their data and inference infrastructure locally controlled.

Potential deployments range from conventional generative-AI assistants and document-analysis systems to computer vision, analytics, enterprise automation and specialised industry models. Government agencies could potentially use locally hosted systems for sensitive administrative workloads, while banks, hospitals and industrial companies could use them for applications where regulations or security policies restrict movement of confidential information outside organisational networks.

The system is also intended to scale beyond individual appliances. Vigyanlabs describes its technology architecture as capable of expanding from a single FEMTO installation at the edge to much larger private and sovereign computing environments, allowing organisations to add infrastructure as their AI requirements increase.

FEMTO is being commercialised through Kasvin, a joint venture involving Vigyanlabs and Krishnaa Advance Systems, which is part of the Krishna Buildestates group. The partnership combines Vigyanlabs’ software and deep-tech capabilities with infrastructure deployment, investment and market-development expertise.

The partners have announced a planned investment of approximately ₹100 crore to support deployment of FEMTO-based AI infrastructure. The programme is intended to target enterprises, government organisations and institutions that require locally controlled, high-performance artificial-intelligence computing.

The launch is particularly noteworthy because Vigyanlabs is based in Mysuru rather than one of India’s largest technology hubs. The company was founded by Srinivas Varadarajan and Srivatsa Krishnaswamy and has focused on energy-efficient computing, intelligent power management, edge infrastructure and private AI systems. Vigyanlabs says its technologies have been used to manage and optimise millions of computing devices internationally.

Its latest effort extends that work directly into AI infrastructure, where electricity consumption has become one of the technology industry’s biggest challenges. The rapid growth of large language models and generative AI has led to increasingly powerful GPU clusters, which in turn require significant electricity, cooling and data-centre investment.

FEMTO attempts to tackle that problem from the infrastructure level by combining high-density computing with software that continually manages utilisation and power consumption. Rather than focusing solely on the raw performance of individual processors, Vigyanlabs is attempting to increase the amount of useful AI computation produced from each watt of electricity.

The company has already demonstrated configurations running sizeable large language models on compact GPU installations. Vigyanlabs has reported testing a 70-billion-parameter model on a two-GPU FEMTO configuration, highlighting its effort to extract substantial AI inference capacity from relatively small physical installations. Such performance figures remain company benchmarks and will vary depending on model, hardware and operating conditions.

FEMTO is also moving beyond laboratory demonstrations. One of the first projects associated with the platform is an AI rural data centre at Gundlupet in Karnataka’s Chamarajanagar district, established by Cynefian. The project is intended to demonstrate that advanced AI computing infrastructure does not necessarily have to remain concentrated in Bengaluru, Hyderabad, Mumbai or other large metropolitan technology centres.

Distributed micro data centres could potentially allow artificial-intelligence computing to be placed closer to universities, factories, hospitals, government facilities and rural technology clusters. Local processing could reduce dependence on long-distance data transfers while improving data sovereignty and creating access to advanced computing outside India’s largest cities.

The sustainability argument is equally important. Vigyanlabs says its broader green computing technologies have been designed to reduce infrastructure costs and power consumption significantly, while its micro-data-centre architecture targets considerably lower Power Usage Effectiveness than conventional data-centre installations.

However, FEMTO’s real-world impact will depend on how its performance, cost and energy efficiency compare with conventional GPU servers and cloud-based AI systems when deployed across different workloads. Claims of large energy savings will ultimately need to be assessed under independent and comparable operating conditions.

Even so, the platform represents an interesting development within India’s emerging AI hardware and infrastructure ecosystem. Much of India’s recent artificial-intelligence activity has focused on software models, applications and access to imported GPU capacity. FEMTO instead addresses the physical and software infrastructure required to operate those models privately.

This gives the technology relevance to India’s broader ambition to develop sovereign AI capability, where sensitive workloads can be processed within domestic and institution-controlled infrastructure.

For sectors such as defence, government, banking, healthcare, manufacturing and strategic research, the ability to operate AI models inside an isolated network could be as important as the performance of the model itself. In such environments, the movement of confidential information into an external cloud may simply not be acceptable.

By combining on-premise AI computing, air-gapped operation, Indian-developed infrastructure software and air-based cooling within a compact appliance, FEMTO represents an attempt to make high-performance sovereign AI easier to deploy without building an entire specialised data centre around it.

The product also demonstrates that India’s deep-tech ecosystem is beginning to move beyond AI applications and models into the underlying infrastructure on which artificial intelligence operates.

From its base in Mysuru, Vigyanlabs is therefore targeting a fundamental question facing the next generation of artificial intelligence infrastructure: how to deliver increasingly powerful AI computing without simultaneously requiring ever-larger quantities of electricity, cooling infrastructure and externally controlled cloud capacity.

---