7 min readparanine
What a sovereign AI platform in Pakistan has to prove
A sovereign AI platform in Pakistan runs models on hardware inside the country, bills in rupees and keeps no prompts. Five things it should prove, call by call.
A sovereign AI platform in Pakistan is one that runs AI models on hardware inside the country, bills in rupees, and can prove both for every request it serves. The word has been stretched to cover a local sales office or a logo on a slide, so the useful definition is the one a buyer can check.
This post sets out five things such a platform should be able to prove, in the order a compliance team or an engineering lead will ask for them, and what paranine shows for each.
Proof one: where each request was executed
Sovereignty is a property of a single call, not of a company. A platform that serves some models in the country and relays others abroad is common and perfectly reasonable, but only if it tells you which is which before you send anything.
On paranine the model catalogue marks the pool behind every route. PK NPU-native routes are served on Paranine's own hardware inside Pakistan; global pass-through routes are served on partner infrastructure outside it and are labelled that way. You choose the residency of a workload by choosing the route, and you can scope a key so it cannot call anything else.
The PK Endpoint: what is served in Pakistan, and on what hardware
Proof two: whose hardware, and where it sits
Renting capacity in a foreign region and calling it local is the most common way the word is misused. The question that settles it is physical: which accelerators, in which country, operated by whom.
paranine's PK pool runs on Huawei Ascend 910B NPUs, 64GB accelerators racked in Pakistan rather than rented by the hour abroad. Paranine built the serving stack for these chips and operates the fleet directly, and the same route ID can be served from Ascend NPUs or Nvidia GPUs without the calling code changing.
Proof three: what is kept after the answer
Residency of the computation means little if the prompt is then copied into a log somewhere else. Ask what is retained, for how long, and where.
paranine's gateway stores no prompts and no completions on any route. The request log keeps the model, the key, the status, the timings, the token counts and the cost, which is everything billing, debugging and an audit need, and nothing of what was said.
Proof four: what currency the invoice is in
A platform billed in dollars keeps the business exposed to the exchange rate and to the card and approval process that a foreign payment needs, whatever its servers do. Sovereignty that stops at the data centre leaves the finance function where it started.
paranine quotes, bills and settles in Pakistani rupees: per-million-token rates published in PKR, a prepaid balance, and the exact rupee cost of each request in the request log. New organisations start with Rs 1,000 of free credits.
Proof five: who controls the keys
The last proof is governance. A platform should let the customer decide which models a key may call, how much it may spend and when it stops working, and enforce those decisions on the request path rather than in a policy document.
- Each API key can be scoped to approved models; a call outside the scope is refused with a 403
- Each key can carry a rupee credit limit; at the cap that key stops with a 402 and everything else keeps running
- Each key can carry an expiry date
- Every administrative change is recorded in an append-only audit log, exportable as CSV
What a sovereign AI gateway is, and how to tell a real one from a badge
How to use this list
Put the five proofs in the procurement questionnaire and ask for them per route, not per company. A platform that answers all five for a specific model ID is sovereign for that workload; one that answers with a mission statement is not.
If none of the five matters to your workload, a global provider with a larger catalogue is a reasonable choice and the honest recommendation. If even one does, it is usually the requirement that decides the project.
Common questions
- What is a sovereign AI platform?
- A sovereign AI platform serves AI models on hardware inside one country, under that country's law, and bills in its currency. For Pakistan that means inference on machines physically located in Pakistan, invoices in rupees, and records a Pakistani regulator can be shown. The test is whether the platform can say, per request, where that request ran.
- Is there a sovereign AI platform in Pakistan?
- Yes. paranine serves open-source models on Huawei Ascend 910B NPUs it operates inside Pakistan, through one OpenAI-compatible API billed in rupees. Its catalogue marks which routes are served in Pakistan (PK NPU-native) and which pass through to partner infrastructure abroad, so residency is chosen per route.
- Does a sovereign AI platform store my prompts?
- It should not have to, and on paranine it does not: the gateway stores no prompts and no completions on any route. The request log keeps metadata only (the model, key, status, timings, tokens and cost), which is what billing and audit need and nothing of what was said.
- Is sovereign AI only for banks and government?
- No, though they feel the requirement first. Any team whose customers ask where their data is processed, or whose finance function cannot hold a dollar balance, has the same two problems: jurisdiction and currency. A software house selling abroad often needs the residency answer for vendor qualification.