Narendra Sen, Founder and CEO of NeevCloud and RackBank, is planning to raise up to USD 1.2 billion in capital this financial year to fund a large-scale deployment of graphics processing units (GPUs) and expand its artificial intelligence (AI)-ready infrastructure.
In an interview with BW Businessworld, Sen said the company has already mobilised approximately Rs 700 crore, comprising around Rs 150-160 crore in promoter equity and Rs 540 crore in debt from lenders, to build and operationalise its flagship AI data centre park in Raipur, Chhattisgarh. That facility, he said, is within days of going live.
The broader Rs 6,000 crore (up to USD 700 million) raise now underway is intended to fund the deployment of a further 11,000 GPUs for three international enterprise customers who have already signed letters of intent (LoIs).
The Indore-based founder said that his ventures have already committed 1,000 Nvidia B300 GPUs worth USD 50 million, which has already been raised.
The Capital Stack
Sen laid out the financing structure in precise terms. To unlock the full Rs 6,000 crore of capital required for the 11,000-GPU commercial deployment, the company needs to first raise Rs 1,500-2,000 crore in equity, at which point banks will finance the remaining 75 per cent as debt, in line with standard infrastructure lending norms in India.
The additional equity raise has not yet closed. Sen said he expects it to conclude within six to eight months, before the end of FY27, and confirmed that private equity players and banks are now actively engaged, encouraged by the Indian government’s recently announced 20-year tax break for qualifying data centre infrastructure.
“Now everybody is interested because the government has given a 20-year tax break (announced in the Budget FY26-27) and the environment is right,” Sen said.
Big Revenue Ambition
For most of RackBank’s 13-year history, annual recurring revenue (ARR) stayed flat between USD 1 million and USD 2 million. The company bootstrapped through that entire period, never raising institutional capital, reinvesting margins from a retail web hosting and managed services business, and growing one server at a time.
“The ARR will come this year in the books. Net may come up to USD 220 million this year. I estimate an average of USD 200 million for FY27” – Narendra Sen, Founder & CEO, RackBank & NeevCloud
The trajectory is about to change sharply, Sen said. With the Raipur site expected to be operational shortly and the IndiaAI Mission tender in place, Sen projects RackBank’s ARR will reach USD 20 million in FY26 from current deployments. Once the 11,000-GPU commercial tranche is live, contingent on the equity raise, he expects ARR to reach USD 200-220 million in FY27.
The Cost Arithmetic
The competitive case RackBank makes to customers rests almost entirely on operating cost. Sen claims the company can deliver GPU compute at 15 cents per GPU per hour, against an industry norm of approximately 45 cents, a 30-cent gap that the company passes on in its pricing. The advantage flows from geography and energy strategy.
RackBank has concentrated its infrastructure in central India, specifically Chhattisgarh, which Sen says carries a power surplus of 22 gigawatts today, expected to rise to 30 gigawatts within five years. Power from renewable sources now costs approximately Rs 6 per unit in this region, against Rs 8 from the grid, a 25 per cent saving that compounds across the energy-intensive GPU workloads.
DRAM Bottlenecks, Nvidia B300 Orders And The Inference Alternative
During the conversation with BW Businessworld, Sen was candid about the headwinds on the supply side. GPU memory, specifically DRAM, has become the critical choke point in the global AI infrastructure build-out, with a concentrated supplier base of three to four companies and lead times stretching from six weeks to six months for some components.
“RAM companies are saying you are in a queue, you will not get supply,” he said. “A component that was USD 200-300 has become USD 2,000, a 2 per cent cost component has become a 20 per cent cost item in the supply chain.”
For its Raipur GPU cluster, RackBank has committed to Nvidia’s latest B300 chips. The 1,000-unit IndiaAI tranche is being procured immediately, requiring the full USD 50 million to be transferred upfront. The 11,000-unit commercial deployment follows once equity financing closes.
Sen’s admission about DRAM bottlenecks is the biggest threat the company’s planned revenue trajectory. If the memory supply stays tight, those 11,000 GPUs might not all arrive in FY27, which would push the revenue into FY28.
Five State-Plan And 20 MW Per Location As Standard
Beyond Raipur, RackBank has acquired land in Navi Mumbai and is awaiting permits to begin construction. An MoU has been signed in Assam for a new site this year. Construction is underway in Madhya Pradesh, where the company has acquired additional land. Discussions are also ongoing with the Karnataka government.
The company’s target is three fully operational locations within the next twelve months, each designed for 20 megawatts of capacity, with Raipur built to accommodate a larger footprint than the others.
“In this industry the only differentiator is speed,” Sen said. “Everybody has money and everyone can build the same thing maybe two years later. But speed is the edge.”
The company uses a common vendor and design playbook across sites to compress construction timelines: plinth work takes two months and cannot be shortened, but PEB (pre-engineered building) fabrication runs in parallel at the factory, and MEP equipment is ordered early given six-week-to-six-month import lead times.
Looking ahead, Sen said his companies have signed a proof-of-concept (PoC) agreement with “one of the largest banks in the country” to build an underground data centre in a decommissioned mine. The rationale: hardened infrastructure resilient to targeted attacks on critical national systems. Design work is underway and Sen said visible progress should be apparent by next year.
Even as Sen races to close FY27 financing, NeevCloud is looking to launch orbital compute nodes in partnership with spacetech startup Agnikul Cosmos. A test satellite carrying experimental load is targeted for launch before the end of December 2026, with the eventual ambition of 600 satellites providing low-latency GPU compute to autonomous vehicles, drones, defense systems, submarines and oil rigs globally. |
