zypl.ai and Zinda Capital launch GPU credit card pilot in Tajikistan

zypl.ai and Zinda Capital have signed an MoU to develop what they describe as the world’s first GPU credit card, starting a September 2026 pilot in Tajikistan. The product would let users access NVIDIA H200 compute on demand and pay by GPU-hour, with a future credit version under development. Why it matters: - The partnership aims to make on-demand GPU compute available domestically in Tajikistan through a regulated financial rail. - The product is designed to reduce reliance on foreign cloud providers, which can add FX risk, regulatory friction and data-sovereignty concerns. - The pilot could create a new model for paying for AI compute, with potential use cases for researchers, startups and developers. What happened: - zypl.ai and Zinda Capital signed a Memorandum of Understanding to develop a dedicated product they describe as a GPU Credit Card. - The launch was announced in Dushanbe, Tajikistan, on June 18, 2026. - The project is set to begin in September 2026. - The companies say the product represents a global first: a credit limit that provides instant, on-demand access to GPU compute. The details: - A card transaction does not trigger a normal cash payment. - The transaction provisions a compute resource tied to a server powered by an NVIDIA H200 GPU. - Billing is measured in GPU-hours, so the user pays only for the compute time actually used. - Zinda Capital will act as the card issuer and handle customer identification, billing and settlement in Tajikistan somoni and U.S. dollars. - zypl.ai will provide the compute infrastructure and track consumption in real time. - The compute setup includes four NVIDIA H200 GPUs with a combined 564 GB of HBM3e memory. - 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. - At launch, the product works as a prepaid compute wallet. - Users top up fiat funds, which convert into a GPU-hour balance. - The companies are also developing a credit option, described as “train now, pay later,” underwritten by zypl.ai’s scoring engine. Between the lines: - The initial prepaid model lowers the risk of launching a new category of financial product tied to AI infrastructure. - The planned credit feature suggests the partners want to extend the product from metered access to underwriting-based financing. - The pilot configuration of four H200 GPUs creates about 2,880 GPU-hours per month, according to the announcement. - The companies say scaling to 100 GPUs could unlock $2.25 million to $4.25 million in annual billing potential. What’s next: - zypl.ai and Zinda Capital plan to develop the product together and provide more details as the work advances. - The September 2026 start will test whether GPU compute can be packaged and financed like a consumer or business credit product. - The partners are positioning the launch as a path to broader regional access to domestic AI infrastructure. The bottom line: - The GPU Card is an attempt to turn AI compute into a bankable, metered financial product, starting in Tajikistan.