Startup Deep Dive : NeevCloud — sub-Rs 10 crore revenue and a $1.2 billion GPU bet

NeevCloud reported revenue of less than ₹10 crore in the 12 months to March 2025, according to startup data platform Tracxn. Its founder, Narendra Sen, is trying to raise up to $1.2 billion. That gap between what the business earns today and what it wants to spend is the whole story, and it is the reason NeevCloud is one of the most closely watched, and most argued-about, names in India’s AI infrastructure push.

The pitch is simple to say and hard to do: build a “sovereign” AI cloud, owned in India and run on Indian soil, so that Indian companies training and running AI models are not renting every GPU-hour from an American hyperscaler. The reality check is just as blunt. In November 2025, the research firm SemiAnalysis put NeevCloud in the lowest bucket of its ClusterMAX cloud rankings, the “Unavailable Tier”, because it could not independently verify that the service was live and testable. This is a deep dive into a company that is part real infrastructure, part order book, and part promise, and into how you tell those three apart.

Quick facts

What NeevCloud actually sells

NeevCloud sells rented AI compute. The core product is on-demand and reserved access to clusters of graphics processing units (GPUs), the chips that train and run large AI models, packaged with storage and networking and aimed at Indian startups, enterprises and government-linked buyers who want their data to stay inside India.

• What it offers, per the company and third-party listings: bare-metal and virtualised GPU clusters, “AI Supercomputing-as-a-Service”, plus IaaS, PaaS and an inference platform.
• Chips it advertises: NVIDIA H100, A100, L40, H200, B200 and B300, and AMD’s Instinct MI300X (Analytics India Magazine, May 2024; SemiAnalysis ClusterMAX review, November 2025).
• Cluster sizes advertised for pre-reservation: 1,000 to 16,000 GPUs connected over InfiniBand, per NeevCloud’s own site as recorded by SemiAnalysis in November 2025.
• Price claim: the company advertises GPU access from $1.69 per hour and calls it among the world’s lowest, saying it cuts cost by up to 50% (company-stated). This is a company claim, not an independently audited price.
• Positioning: a “sovereign” cloud that keeps data and processing within India, aligned with the Digital Personal Data Protection Act and the IndiaAI mission (company-stated).

The origin: from RackBank to a supercloud

NeevCloud did not appear from nowhere. It is the AI-cloud layer on top of a data-centre business that Narendra Sen has been building since 2013, when he founded RackBank Datacenters in Indore, in Madhya Pradesh, far from the Mumbai-Bengaluru axis where most Indian infrastructure money pools. RackBank’s early idea was unfashionable and stubborn: that a serious data centre could be built and run out of central India, and that the country should own more of the hardware its digital economy runs on.

The founding insight behind NeevCloud is a bet that the same argument now applies to AI. Sen’s contention, in his own framing, is that Indian buyers do not care about the brand on the cloud so much as whether it works. “Indians don’t care about the brand. What they care about is that the API should work, latency should be fast, and they should get immediate tokens,” he told Analytics India Magazine in May 2024. NeevCloud is the attempt to turn a colocation and hosting company into a full-stack AI compute provider, using RackBank’s buildings, power and cooling as the foundation, hence the name Neev, Hindi for foundation.

The struggle years

The hard part of this story is not the vision, which is easy to narrate, but the distance between the plan and what has actually been switched on. The struggles are structural, and the company has been candid about the scale of what still has to be financed.

The first difficulty is capital intensity. A GPU cloud is not a software startup; it is a balance-sheet business where the product is bought before a rupee of revenue arrives. NeevCloud set out in May 2024 with a plan to deploy 40,000 GPUs and storage worth about $1.5 billion by 2026, beginning with an order for 8,000 GPUs from Hewlett Packard Enterprise for delivery in the second half of 2024 (Analytics India Magazine, May 2024). Orders of that size have to be paid for long before the customers using them show up, which is why the company has leaned on debt and a revenue-sharing model with data-centre operators rather than pure equity.

The second difficulty is credibility. In November 2025, SemiAnalysis reviewed NeevCloud for its ClusterMAX ranking and placed it in the “Unavailable Tier”, the label it uses for services it cannot verify because they are not launched, are sold out, or are government-only. The reviewers wrote that they could not test the Kubernetes and Slurm clusters, and noted that the company did not want its offerings tested. For a business whose entire pitch is reliable, sovereign compute, an independent “we could not verify this is live” is a real setback, and it lands in the same window as the fundraising drive.

The turning point

The single event that turned NeevCloud from a slide deck into a physical asset was the Raipur data-centre park. In May 2025, RackBank inaugurated an AI data-centre park at Nava Raipur, in Chhattisgarh, with an initial investment of ₹1,000 crore, roughly $73 million at $1 ≈ ₹96.0, spread across 13.5 acres that include a 2.7-hectare special economic zone (Evertiq, May 2025). The company described it as India’s first dedicated AI data-centre park, a claim that traces to RackBank itself rather than to an independent register.

The numbers on either side of that opening show the leap being attempted. Before it, NeevCloud’s public footprint was a plan and a handful of test GPUs. After it, the group had a live site engineered for high-density AI racks: Phase 1 designed for 5 MW of power, with an expansion path the company puts at 150 MW, direct-to-chip and immersion “Varuna” liquid cooling it says cuts cooling costs by up to 70%, and a projected 500-plus jobs (Evertiq, May 2025). It is the difference between promising sovereign compute and pouring the concrete for it, even if most of the capacity is still to be built.

The money behind it

The funding picture is split across the two entities and mixes equity, debt and a very large ambition that is not yet closed.

• RackBank seed round: $16.5 million, about ₹138 crore, in a single round dated March 2025, from D-Street investors including Lucky Investment Managers and KRIIS, alongside angel investors (Tracxn; corroborated by market reporting on the ₹138 crore equity round).
• Capital already mobilised for Raipur: about ₹700 crore, made up of roughly ₹150–160 crore of promoter equity and ₹540 crore of debt from lenders, to build and operate the flagship park (Narendra Sen, as reported by BW Businessworld, April 2026).
• Committed GPU order: 1,000 NVIDIA B300 GPUs worth $50 million, which the company says has already been funded (BW Businessworld, April 2026).
• Round now underway: ₹6,000 crore, which BW Businessworld reported as up to about $700 million, to deploy a further 11,000 GPUs for three international enterprise customers that have signed letters of intent (BW Businessworld, April 2026).
• Headline target: up to $1.2 billion of capital for GPU deployment and AI infrastructure across the financial year, per Sen (BW Businessworld, April 2026). This is a stated plan, not money in the bank.

The named backer worth watching is Lucky Investment Managers, the firm associated with veteran public-markets investor Ashish Kacholia; its participation is what turned RackBank’s 2025 round from a founder-and-friends raise into one carrying a recognised name.

How it makes money

Strip away the vocabulary and NeevCloud is an asset-rental business. It buys expensive chips and buildings, then sells time on them. The economics live and die on utilisation, the cost of capital, and power.

• Money in: GPU-hours, sold either on demand or on multi-month reservations, plus storage and networking; the advertised entry price is $1.69 per GPU-hour (company-stated).
• Money out: the GPUs themselves (an $50 million cheque for just 1,000 B300s, per the company), plus power, cooling, real estate and the interest on the debt used to buy the hardware.
• Where the margin sits: in keeping the chips busy. Idle GPUs still cost interest and power, so signed reservations, the letters of intent from three international customers, matter more than any single price.
• Capex-light layer: to move faster than it can build, NeevCloud has also used revenue-sharing partnerships with third-party data-centre operators, initially in Chennai and Mumbai, rather than owning every rack (Analytics India Magazine, May 2024).
• The part people get wrong: a low headline price per GPU-hour is not a moat. Because everyone buys the same NVIDIA and AMD chips, the real edge is cheaper capital, cheaper power and higher utilisation, not the sticker rate.

The numbers

This is a young business, so the honest version of “the numbers” is a short history and a long forecast. NeevCloud was incorporated in 2024; a three-year audited revenue series simply does not exist on the public record yet, and inventing one would defeat the point of the exercise. What can be sourced is set out below, with actuals and projections kept strictly apart. Figures are in ₹ crore unless a dollar figure was reported directly by the source.

The shape is stark: reported revenue in the low tens of crore in FY25, and a company-stated aim to reach $200–220 million of annualised revenue by FY27. Every rupee of that jump depends on capital being raised and chips being switched on, so treat the FY26 and FY27 lines as targets, not results.

Where the money comes from

Because the revenue is small and forward-looking, the more useful split is where the capacity and demand sit, not a historic segment breakdown that does not yet exist.

• By site (planned power): the group is building toward roughly 300 MW of AI-focused capacity across Navi Mumbai, Raipur and Indore (BW Businessworld, April 2026). Raipur’s Phase 1 is 5 MW with a stated path to 150 MW (Evertiq, May 2025).
• By customer type: the near-term revenue driver is three international enterprise customers whose letters of intent underpin the 11,000-GPU tranche (BW Businessworld, April 2026), alongside Indian startups and sovereign or government-linked buyers.
• By chip vendor: demand is split across NVIDIA (H100, A100, L40, H200, B200, B300) and AMD (MI300X), a deliberate hedge against single-vendor pricing and supply (Analytics India Magazine, May 2024; SemiAnalysis, November 2025).
• The surprise: for a company that sells itself on Indian sovereignty, the money that gets it to scale is disproportionately foreign, the three international customers behind the ₹6,000 crore tranche, and imported, since the GPUs come from American chipmakers via HPE and NVIDIA.

The risks

The risks here are concrete and mostly financial or physical, not reputational hand-waving.

• Financing risk. The entire growth case rests on raising up to $1.2 billion, of which only about ₹700 crore was reported as mobilised for Raipur as of April 2026 (BW Businessworld). If the ₹6,000 crore round does not close, the FY27 ARR target of $200–220 million does not happen. The mechanism is simple: no capital, no GPUs, no revenue.
• Utilisation and demand risk. GPUs depreciate fast and cost interest whether or not they are used. Letters of intent are not binding contracts; if the three enterprise customers slip or shrink, the chips sit idle while the debt still has to be serviced.
• Credibility and verification risk. SemiAnalysis placed NeevCloud in its “Unavailable Tier” in November 2025 because it could not test the service. Until independent users can benchmark live clusters, the gap between advertised capacity and running capacity is itself a risk to winning demanding customers.
• Competition and price risk. NeevCloud competes with global neoclouds and better-funded Indian rivals for the same chips and customers. SemiAnalysis openly questioned “how many neoclouds will make it going forward”; a low advertised price of $1.69 per GPU-hour is easy for a deeper-pocketed rival to match.

The takeaway

The transferable lesson from NeevCloud is about the order in which infrastructure businesses have to prove themselves. A software startup can raise on a story and build the product afterwards. A GPU cloud has to buy most of the product first, on debt, and only then find out whether the customers and the utilisation show up, which means the founder’s real job is not selling a vision but sequencing capital, chips and contracts so that none of the three gets too far ahead of the others. NeevCloud has done the hard, unglamorous first step, pouring concrete in Raipur and signing letters of intent, but it is being judged, fairly, on a much larger promise it has not yet financed. Whether it becomes India’s sovereign-cloud backbone or a cautionary tale about buying ahead of demand will be settled by one number: how much of that $1.2 billion actually closes, and how quickly the GPUs it buys start earning.

Frequently asked questions

What is NeevCloud?

NeevCloud is an Indian AI cloud provider that rents access to GPU clusters, storage and networking for training and running AI models, marketed as a “sovereign” service that keeps data inside India. It sits on top of the data-centre business of RackBank, and both are led by founder Narendra Sen.

Who founded NeevCloud and what did they do before?

Narendra Sen is the founder and CEO. He founded RackBank Datacenters in Indore in 2013 (per Tracxn), building data-centre and colocation infrastructure in central India before launching NeevCloud’s AI SuperCloud, publicly announced in November 2023.

How much money has NeevCloud raised?

As a separate entity NeevCloud is listed as unfunded by Tracxn. The related RackBank raised a $16.5 million seed round (about ₹138 crore) in March 2025, and the group had mobilised about ₹700 crore for its Raipur park as of April 2026. Founder Narendra Sen has said he plans to raise up to $1.2 billion in total (BW Businessworld, April 2026).

What revenue does NeevCloud make?

NeevCloud reported revenue of under ₹10 crore for the year to 31 March 2025, and RackBank ₹10–50 crore over the same period, both per Tracxn. The company projects group annualised revenue of $200–220 million by FY27, but that depends on its fundraising and GPU deployment going to plan.

Is NeevCloud’s AI cloud actually live?

Partly. RackBank inaugurated its Raipur AI data-centre park in May 2025, but in November 2025 the research firm SemiAnalysis placed NeevCloud in its “Unavailable Tier”, saying it could not independently verify or test the clusters. So there is real physical infrastructure, but independent verification of a fully live, testable service was still lacking as of late 2025.

Sources

Figures are as of September 2026. Currency converted at $1 ≈ ₹96.0 as of 18 September 2026 (Trading Economics).

• Tracxn — NeevCloud and RackBank company profiles (revenue bands, incorporation years, funding, board), accessed September 2026.
• BW Businessworld — “NeevCloud, RackBank CEO Plans $1.2 Bn Raise For AI Infrastructure”, April 2026.
• Evertiq — “RackBank inaugurates $118 million AI data centre in Raipur”, May 2025.
• Analytics India Magazine — report on NeevCloud’s 40,000-GPU plan and HPE order, May 2024.
• SemiAnalysis (ClusterMAX) — NeevCloud cloud review, “Unavailable Tier”, November 2025.
• The Tech Capital / PR Newswire — NeevCloud AI SuperCloud launch announcement, November 2023.
• Crunchbase — Ashish Kacholia / Lucky Investment Managers and RackBank profiles, accessed September 2026.

Found an error? Write to us and we’ll correct it in the open, dated, on the piece.