Reinventing telcos and data centres
Is critical for ai resilience.
Data should not move to a central place.
Else thieves get a ready booty.
Distributed Learning Is Critical for Sovereign AI. Sovereign AI isn’t about building another chatbot. It’s about who controls intelligence in finance, health, energy, defence, and governance.
Most national data cannot and should not be centralized or exported. Unless it is suitably encrypted and adequately defended. Given the humongous amount of data involved, this is niether practical nor affordable. Cloud storage prices are currently cartelised and controlled by BigTech. They are the equivalent of Rs 30,000 per square feet per month rent for a data centre in the back of beyond. Remove the Rs 400 per sqft per month rent, it is still an uncontrolled unregulated loot. The data centre owner gets Rs 250 per sqft. Rest is all forex leakage.
If the telcos are made non hierarchical and the data centres are made distributed and hierarchical – it will become amenable to virtualised terminals and secured communications with cyber secure localised data laking. This would ease data orchestration and reduce GPU load.
Distributed ai learning solves a core problem. U can either send the thirsty (ai model) to the well. Or send the well to where the thirsty are. BigTech likes the latter. We need the former. Moving data is highly avoidable.
But how do you learn from national-scale data without moving the data?
Instead of pulling data into foreign controlled clouds, the model moves to where the data already lives.
Only learning updates are shared. Raw data never leaves.
This enables:
• Data sovereignty without data hoarding
• Compliance without paralysis
• National-scale learning without mass surveillance
In countries like India—where data is fragmented across states, regulators, and institutions—distributed ai learning isn’t optional.
It’s structurally aligned with how systems like UPI, GSTN, ABDM, and CBDC actually work.
More importantly, it reduces dependence on:
• Hyperscalers
• Cross-border data flows
• External control planes
Centralized AI creates data colonies.
Distributed learning enables context-native intelligence.
In one line:
Distributed learning lets a nation compound intelligence across its systems without exporting its data, its control, or its autonomy.
If an ai LLM is not built ground up for distributed learning, it should be BANNED in India.
Every big tech narrative is around “give me your data”. The idea is not to provide any ai technology but just to steal data. Every govt office is swamped with foreign VC funded cos seeking data under some pretext or the other. Govt hospitals delete data because they don’t have budgets to buy storage devices. Bigtech provides these entities with free storage for six months. The terms and conditions clearly state that data stored on their servers can be used and even sold.
Govts have to wake up
And smell the coffee
Data lost is country lost
