English
English
English
India unveils its first sovereign AI box- CommandCORE that runs enterprise AI locally, keeps sensitive data private, and can scale from small tasks to massive workflows. Could this be the future of secure, on-premise intelligence?
India’s first sovereign AI product built around the concept of ‘AI in a box’
New Delhi: At the India AI Impact Summit 2026, Indian AI transformation foundry Arinox AI and agentic AI specialist KOGO unveiled what they are calling India’s first sovereign AI product a compact, enterprise‑grade system designed to run powerful artificial intelligence locally without relying on the internet.
Called CommandCORE, the system is being positioned as a secure, private alternative to cloud‑dependent AI. Built on partnerships with Nvidia and Qualcomm, the latest iteration of CommandCORE runs on Nvidia hardware and packs significant computational power in a physically compact unit.
What to expect from Emmanuel Macron’s India visit for India AI Impact Summit 2026?
As organisations increasingly adopt AI agents, concerns around data security, privacy, and operational costs have grown. Vendors often rely on large, cloud‑hosted foundational models but this exposes sensitive business information and operational insight to third parties.
“The future of AI is private, on an enterprise level too. You simply cannot farm out your intelligence,’’ said Raj K Gopalakrishnan, CEO and co‑founder of KOGO AI. “Organisations must own the AI if they want to truly increase their internal intelligence and learning.”
CommandCORE is designed to compute locally, keeping sensitive workflows on‑premise. Rather than sending raw data to remote servers, enterprises running AI workloads on CommandCORE can filter and process information in situ, sending only essential summaries externally. This approach not only reduces privacy risk but can also cut cloud bandwidth and compute costs.
The product comes in multiple configurations, each suited for different enterprise needs. The smallest option can run models between 1 billion and 7 billion parameters, suitable for tasks like batch processing or human resources automation. Mid‑tier configurations support models between 20 billion and 30 billion parameters, enabling more complex inference workloads.
‘Will Reshape, Not Replace Jobs’: Tech leaders at Delhi AI Summit; Full story here
For larger deployments, interconnected units including Nvidia’s DGX Spark and Blackwell‑series server systems can scale to support enterprise‑wide AI workflows. Multiple interconnected units can handle models up to 405 billion parameters, according to Nvidia documentation.
Industry data has shown that enterprises are increasingly wary of vulnerabilities introduced through external AI tools. A security analysis found that a significant majority of organisations cite concerns over third‑party integrations, including widely used platforms such as ChatGPT, Copilot, and Gemini.
Arinox and KOGO are targeting sectors where data sensitivity is paramount such as finance, government services, defence, and regulated industries positioning CommandCORE as a secure, on‑premise solution that brings intelligence closer to the edge of business operations.
