zypl.ai, a generative AI company specializing in synthetic data and decision intelligence, and Zinda Capital, a licensed microfinance deposit organization in Tajikistan, announce today the signing of a Memorandum of Understanding and the development launch of a dedicated product setting a new global precedent: a credit limit that gives the cardholder instant, on-demand access to GPU compute.

A card transaction does not initiate a conventional payment — it provisions a compute resource: the cardholder gains access to a server powered by an NVIDIA H200 GPU — and pays only for the time actually used, measured in GPU-hours.

How it works:

Zinda acts as the card issuer, handles customer identification (KYC), billing, and settlement in TJS and USD. zypl.ai provides the compute infrastructure — a fleet of four NVIDIA H200 GPUs with a combined 564 GB HBM3e memory — and tracks consumption in real time.

When a user requests compute, Zinda authorizes the transaction using zypl.ai's AI scoring. The system then provisions a server and delivers access credentials to the user, with billing accruing for each GPU-hour consumed.

At launch, the product operates as a prepaid compute wallet: the user tops up fiat funds, which convert into a GPU-hour balance to be spent as needed. Looking ahead, the partners are also developing a credit option — "train now, pay later" — underwritten by zypl.ai's scoring engine, giving users the flexibility to run their models first and settle afterward.

Why this matters for Tajikistan and the region?

Today, there is effectively no commercial market for on-demand GPU compute in Tajikistan. Researchers, startups, and developers are forced to turn to foreign cloud providers — absorbing FX risk, regulatory friction, and losing control over data sovereignty.

GPU Card closes that gap: for the first time, compute capacity is available domestically, on a regulated rail, without the need for foreign accounts or payment systems. The pilot configuration (4× H200) creates a pool of ~2,880 GPU-hours per month; scaling to 100 GPUs unlocks billing potential of $2.25M–$4.25M per year.

This project is initiating in September 2026 and, for now, reflects the parties’ shared intent to bring the GPU Card to market. zypl.ai and Zinda Capital are committed to developing the product together and will share further detail as the work advances.