Resources · Blog
Notes on sovereign inference.
What we are learning building an NPU-native gateway in Pakistan, and what it means for the teams that call it. Short, practical, and free of benchmarks we have not run.
Page 2 of 2
8 min read
Serving LLMs on Ascend NPUs: what actually changes from CUDA
Import constraints mean much of the accelerator capacity arriving in Pakistan is Huawei Ascend, not Nvidia. The differences are real, they sit in the toolchain rather than the maths, and almost none of them should reach an application team.
Read the post6 min read
Pakistan's AI will run on NPUs. Here is what that changes for your stack.
Nvidia import constraints mean the accelerators actually available in Pakistan are Huawei Ascend NPUs. That is a different toolchain, and a gateway is how a team stops caring which one it is.
Read the post5 min read
What changes when your AI bill is in rupees
A dollar-denominated API turns every model call into a small foreign-exchange position. Prepaid PKR credits, per-key caps and fail-closed budgets change how a team plans and controls AI spend.
Read the post5 min read
Moving an OpenAI-compatible codebase to P/9 in one change
The gateway speaks the OpenAI chat completions API, streaming included. Change the base URL and the model string, scope a key, and read the cost of each request back from the response.
Read the post