DUM360India's citizen-powered sovereign supercomputer
DUM360 pools the nation's already-paid-for idle compute — phones, gaming PCs, university & PSU labs, enterprise & MSME machines, and solar GPU rigs — into India's largest sovereign inference engine. Built for AI inference and parallel workloads, not frontier training — and reclaimable by its owner in minutes.
Indian-hosted platform · Earn payouts via UPI · Confidential, sandboxed compute
Most of India's computing power is sitting idle right now
At any moment, hardware the country has already paid for is doing nothing. DUM360 puts it to work — safely — and gives it back instantly.
Phones at night
Hundreds of millions of devices, charging and idle.
Gaming PCs after hours
Powerful GPUs unused while owners sleep.
Labs on weekends
University & PSU labs over evenings & vacations.
MSME workstations
Business machines idle outside working hours.
Surplus daytime solar
Clean GPU cycles from GJ/RJ solar rigs.
Affordable compute for India's builders
Localized, affordable AI inference and rendering for Indian startups, researchers and academics — using cycles the country already owns.
- ✓Localized AI inference at a fraction of public-cloud cost (Phase-2 target: ~½).
- ✓Direct micro-incentives paid to everyday contributors via UPI.
- ✓Solar arbitrage routing for cleaner, cheaper daytime cycles.
A designed sovereign emergency protocol
A designed protocol — not operational today — in which a cryptographically-authenticated government credential could mobilize the national pool for public-good crises. Governance is being developed with relevant authorities.
- ✓Disaster & climate modeling for NDMA and response teams.
- ✓Bio-defense & genetic sequencing for medical agencies.
- ✓Cyber-defense at national scale, on demand.
From idle device to national compute
Enroll once, set your hours, and let DUM360 safely put your spare cycles to work — on your terms.
Install & verify
Citizens, campuses, enterprises and PSUs install a lightweight sandboxed node (K3s/KubeEdge) or mobile WASM client. Hardware integrity is verified via Secure Enclave.
Pre-inform availability
Book recurring windows — e.g. 11 PM–7 AM. The orchestrator maps workloads hours in advance.
Compute safely
Jobs run inside encrypted, confidential sandboxes you can't read — and we never read your files. Tasks mirror across ≥3 machines for resilience.
Get paid
Usage is metered to the rupee — paid straight to your UPI VPA, or as credits/grants for institutions. Graceful drain returns your device 30 min before your exit.
You stay in control — pause, edit your hours, or uninstall anytime. Your device is always yours.
How work flows across the network
Every tier of India's idle compute — campus labs, government & PSU centres, enterprises, solar rigs, gaming PCs and phones — feeds one sovereign core. Workloads, results and UPI payouts flow continuously; switch to Rashtra Seva to see how the network would mobilize for a national emergency.
Illustrative — a diagram of how work, results and payouts route across the mesh, not a live network feed. Toggle to see emergency mobilization.
Hover any node to trace its flows · switch modes above
Who powers the mesh — and what each tier is good at
Country-wide mesh for throughput; tightly-connected clusters for big models. Each tier is scheduled for the work it's actually good at.
Institutional
IIT/NIT/IIIT & university labs
- Interconnect
- LAN + NKN
- Best-fit work
- In-cluster sharding, cross-campus pipelines
Government · PSU
Sovereign labs & data centres
- Interconnect
- DC-grade LAN
- Best-fit work
- High-trust inference, sovereign capacity
Enterprise · MSME
Off-hours servers & workstations
- Interconnect
- LAN / broadband
- Best-fit work
- Batch inference, ETL, rendering
Renewable
Solar GPU rigs (GJ/RJ)
- Interconnect
- Site LAN
- Best-fit work
- Daytime batch & inference (clean cycles)
Consumer
Phones & gaming PCs
- Interconnect
- Consumer broadband
- Best-fit work
- Request-parallel inference (the burst edge)
How big models actually run on the mesh
DUM360 matches model size to interconnect quality — so inference scales without pretending home broadband is a supercomputer fabric.
Small models → whole requests to whole nodes
Models that fit one node (7B–34B quantized) are replicated; whole requests route to whole nodes — scaling to millions of concurrent inferences nationwide.
Big models → split inside one cluster
Genuinely large models (70B–400B+) are split only inside well-connected clusters — a campus lab, a GPU rig, a PSU data centre — where bandwidth and latency can sustain it.
Biggest models → pipeline across campuses
For the biggest models, pipeline parallelism spans institutions over the National Knowledge Network — a loosely-coupled academic grid, not consumer broadband.
Built for massively-parallel, latency-tolerant work — inference, rendering, simulation. Not for training frontier models across home internet — the physics doesn't allow it, and we won't pretend it does. Plus speculative decoding: small models draft tokens, big cluster-hosted models verify, to stretch scarce big-model capacity. Read the honest comparison →
The cleanest, cheapest cycles — honestly scoped
Solar arbitrage runs today. Coastal green compute is an early R&D track for round-the-clock clean baseload — promising, not yet at scale.
Daytime solar & floating coastal rigs
The cheapest, greenest cycles come from daytime solar GPU rigs in Gujarat & Rajasthan, given priority routing 10am–4pm.
- ✓Solar arbitrage — priority scheduling 10am–4pm.
- ✓Seawater-cooled, floating-solar coastal rigs (deployable today).
- ✓Idle hardware reused — no new build, less embodied carbon.
Tidal & wave power
Tidal and wave power are a nascent R&D track in India toward predictable 24×7 clean baseload. We are exploring it — we are not presenting it as near-term capacity.
- ○Goal: round-the-clock clean baseload for coastal compute.
- ○Early-stage research — not available at scale at launch.
Where DUM360 actually fits
We're not claiming to out-build the world's hyperscalers — the physics of consumer networks won't allow it, and we won't pretend otherwise. We turn India's already-paid-for idle hardware into its largest sovereign inference engine, without building a single new data centre. Here's the honest picture.
|
★ DUM360 (sovereign) |
Foreign hyperscaler (AWS/Azure/GCP) |
India public HPC today (NSM/AIRAWAT) |
|
|---|---|---|---|
| Data stays in India | ✓ | ✗ | ✓ |
| Uses existing hardware (no new build) | ✓ | ✗ | ✗ |
| AI inference at scale | ✓ | ✓ | limited |
| Cost | ~½ of cloud* | baseline | n/a |
| Greener (idle + solar) | ✓ | varies | varies |
| Usable AI-inference scale | a few EF/s usable † | effectively unlimited (paid) | ~NSM baseline (~0.04 EF/s FP64) |
| Emergency mobilization (designed) | ✓ | ✗ | partial |
| Frontier model training | ✗ not us | ✓ | limited |
We're honest about what we're not. * Phase-2 cost target, not measured. † Figures are our own bottom-up, clearly-labelled order-of-magnitude estimates, discounted for the verified 2–10× distributed penalty — not measured results or a forecast. On AI-precision peak paper FLOPS the mesh could approach a single flagship AI cluster; usable throughput is a few exaFLOP/s — 1–2 orders of magnitude beyond India's entire current public HPC. We do not claim to rival hyperscaler total fleets or to train frontier models. See the honest benchmark.
What India does with a sovereign inference engine
Two things: power affordable AI for the people building here, and stand ready when the nation needs compute fast.
Affordable LLM inference
Run a million inference requests overnight — request-parallel serving at national scale for startups, researchers and academics.
Batch & video rendering
Render a film sequence across campus labs — for studios, ed-tech and media.
Scientific & ETL batch
Parameter sweeps, Monte-Carlo, and data-extraction pipelines — embarrassingly parallel, checkpointable.
Genomic pipelines
Sequencing & bioinformatics — parallel, latency-tolerant, recoverable.
Student & research compute
Labs accessing capacity they couldn't otherwise afford.
Disaster modeling (Rashtra Seva)
A standby strategic reserve for climate/disaster modeling, bio-defense and cyber-defense — a designed emergency protocol.
Largest sovereign inference in India
1–2 orders of magnitude beyond current public HPC — with data kept in India.
Economic inclusion
UPI micro-income to citizens, compute-grants to institutions, IT-spend offset for MSMEs.
Not built for: tightly-coupled frontier model training over consumer links — the network physics forbids it, and we say so plainly.
Sovereign infrastructure for the platform & your data
The DUM360 compute platform's control plane, metadata and relay traffic run exclusively on Indian clouds — so sensitive data and national orchestration metadata never leave Indian soil. Aligned with the Digital Personal Data Protection (DPDP) Act.
Indian cloud/data-centre providers (such as E2E Networks, CtrlS, Yotta, NIC) are named as illustrative examples, not claimed partnerships. Note: this marketing website is served from a global CDN; signup data is stored in Firestore Mumbai (asia-south1). See our Privacy Policy for exactly where your data lives.
This is for you — here's the ask, and what you get
Pick who you are. We'll keep the next step short — only your email is required.
Build it in the open on GitHub
DUM360 is public and MIT-licensed — no waitlist needed to contribute. Star the repo, raise issues, open pull requests, run and write tests, join the discussions, or help shape the protocol specs. Everything happens in the open.
Raise an issue
Report a bug, request a feature, or flag a gap in the docs — pick a template and tell us what you found.
Open an issue →Join the discussion
Post comments, ask questions, share ideas and weigh in on direction in GitHub Discussions.
Start talking →Open a pull request
Fork, branch, and send a PR against main. We use
conventional commits and review every change.
Run & write tests
Clone the repo, run the suite locally, and add tests that keep the mesh honest and reliable.
Set up locally →Write protocol specs
Help design the node agent, orchestrator, confidential sandbox and Rashtra Seva protocol via architecture decision records.
Shape the specs →Find a good first issue
New here? Start with issues tagged
good-first-issue and
help-wanted.
Honest answers to the obvious questions
Still have a question? Email us.
Join the mesh
Share a device and earn via UPI, register green or off-hours capacity, request affordable compute, or help build the platform — join now and we'll reach out as we open access for your segment.
Under 2 minutes · No spam · Your data stays in India.