We built our own cloud, and we publish the parts where it loses
CloudNX is our own cloud compute platform: dedicated bare-metal NVMe instances in Mumbai and Finland, provisioned in about a minute, with browser SSH and full root access. It exists because our own client deployments kept landing in the gap between shared hosting that is too slow and hyperscaler infrastructure that is too expensive and too complex for the size of the job.
The problem
Every Indian developer or agency picking infrastructure lands in the same gap. Shared hosting at a few hundred rupees a month is oversold and slow enough to end any hope of competing on Core Web Vitals. The hyperscalers are fast but priced and shaped for teams with a platform engineer, and the bill arrives in a currency your accountant has to convert.
In the middle sits a lot of ordinary work: a Laravel app, a Node backend, a staging box, a CSMS gateway. It needs real hardware, root access and predictable pricing, and it does not need a service catalogue with two hundred entries in it.
We were having that conversation on nearly every client deployment. So we built the thing we wanted to put those deployments on, and then had to run it well enough to keep our own production systems on it.
What we built
Instances run on real dedicated NVMe hardware rather than a shared slice of a busy host, which is the difference between a benchmark number and the same number on a Tuesday afternoon. Provisioning takes about a minute, with browser SSH and full root access from the moment it is up.
India in Mumbai and EU West in Finland, both dedicated bare metal. Mumbai keeps Indian traffic in the low tens of milliseconds instead of an overseas round trip, and keeps Indian user data in India, which makes DPDP Act 2023 a simpler conversation. US regions are on the roadmap rather than live, and the site says so.
International cards, net banking, UPI and SEPA, with autopay across all of them so a machine never stops because a top up was forgotten. Every payment generates a downloadable tax invoice for customers who need to claim input credit.
You top up a wallet and pay per minute for what actually runs, with no card required before you start. Postpaid with a month-end invoice is available to established customers on a credit line. There is no free signup credit, which the product states plainly rather than burying.
IAM-style access keys, S3-compatible object storage and standard Linux VMs, so rsync, scp or a container workflow moves an existing AWS, DigitalOcean or Linode deployment across. The exit is documented too: everything sits on Proxmox, so a qcow2 image of any VM exports on demand and imports elsewhere in a couple of hours.
99.5 per cent on compute, written into the MSA rather than into marketing copy. It is single-host until the second box lands, at which point it moves to 99.9 per cent with multi-AZ, and prospects are told that before they sign rather than after.
If any of that is what you need, the scope and the price go in writing first.
Get a fixed-price proposalHow the architecture fits together
The platform runs on dedicated bare-metal hosts under Proxmox, in two regions, with per-minute metering and wallet billing built into the platform rather than bolted on beside it.
The design decision that drives the rest is refusing to oversell. Dedicated NVMe rather than shared slices costs more per customer and removes the noisy-neighbour problem that makes cheap hosting unpredictable, which is the specific failure the product was built to avoid.
We run CloudNX on our own infrastructure rather than reselling someone else's, and that is the part that makes this page checkable: two of our production systems already sit on it.
Where it stands
- Two live regions today, India in Mumbai and EU West in Finland, both on dedicated bare metal. US regions are on the roadmap and are described that way rather than as available.
- Priced about 35 per cent under AWS on the m-class workhorse SKUs, and comparable to DigitalOcean and Linode on raw price while giving bare-metal NVMe instead of a shared VPS.
- CloudNX publishes a line-by-line AWS comparison that includes the SKUs where it loses, which is a decision we would defend on any product we build: a prospect who finds the weakness themselves trusts nothing else on the page.
- PlugEV, our production OCPP platform, runs its stack on CloudNX, including the gateway that holds the live charger WebSocket connections.
- Jai Shri Balaji Store, the Delhi NCR kirana we built a three-app grocery system for, runs its Node.js backend on CloudNX.
- Running an infrastructure product changed our client defaults. Backups, SSL renewal, monitoring and log retention stopped being line items to argue about and became things we assume.
We still run this one. Ask us what it actually took to keep it running.
Ask about this buildQuestions we get about this build
Why build a cloud platform instead of reselling AWS or DigitalOcean?
Because a reseller cannot fix the two things that were actually wrong. It still passes on someone else's pricing and someone else's hardware decisions, so it cannot put you on dedicated NVMe instead of an oversold slice, and it cannot answer a support email inside Indian business hours. Running our own hardware is what makes the price and the response time ours to set.
Is CloudNX cheaper than AWS?
About 35 per cent cheaper on the m-class workhorse SKUs, and comparable to DigitalOcean and Linode on raw price while giving bare metal rather than a shared VPS. It is not cheaper on everything, and CloudNX publishes a line-by-line AWS comparison that names the SKUs where AWS wins, so the answer is checkable before you sign up rather than after.
What uptime does CloudNX commit to?
99.5 per cent on compute today, written into the MSA. That number is the contractual commitment rather than a measured result, and it is deliberately conservative: the platform is single-host until the second box lands, and it moves to 99.9 per cent with multi-AZ at that point. PlugEV has held 99.7 per cent in production since January 2025, but that is measured across the whole PlugEV system rather than the hosting layer, so it is not a CloudNX uptime figure and we will not present it as one.
Is there a free trial?
No. There is no free signup credit. You top up a wallet when you are ready, no card is required before that, and per-minute billing means you only pay for what actually runs. An earlier three-month free offer ran alongside the 2026 launch and has ended, and this page used to still advertise it, which is why it now says this instead.
Can I move an existing site or app onto CloudNX, and off it again?
Both, and the second one matters more than most providers admit. Coming in, there are IAM-style access keys, S3-compatible object storage and standard Linux VMs, so rsync, scp or any container migration workflow works. Going out, everything runs on Proxmox, so you can export a qcow2 image of any machine whenever you like and import it into another cloud in about two hours.
Do you put all of your client work on CloudNX?
No, and that is deliberate. We pick infrastructure per project rather than defaulting to our own product, and we will say when a global provider is the better answer, which is usually when most of your traffic is outside India. What we do run on it is our own production work, which is a stronger signal than a preference.
What this project demonstrates
The services below cite this build as proof. If you need one of them, this is what it looks like when we deliver it.
Want something like this built?
We run this in production ourselves, so the estimate you get is based on having done it rather than having read about it.