Fintech giant to train and serve models locally, advancing data sovereignty and powering AI-driven payments, risk, and personalization tools.
Tabby, the GCC-based buy-now-pay-later and financial services platform, has announced the launch of a new AI factory in Saudi Arabia built on NVIDIA’s HGX systems, part of the company’s push to strengthen sovereign AI infrastructure and accelerate its AI product roadmap.
The factory will serve as Tabby’s local hub for training and deploying AI models across customer support, fraud detection, risk scoring, and shopping personalization. By hosting systems within Saudi borders, Tabby said it will reduce latency, improve compliance with data residency rules, and retain full data sovereignty.
“The AI factory enables us to innovate faster while ensuring that user and transaction data remain in the Kingdom, a key step in scaling responsible AI infrastructure” said a company spokesperson.
The move follows growing interest in regional AI infrastructure as the Kingdom positions itself as a hub for advanced computing and data innovation. Saudi Arabia has announced plans to deploy thousands of NVIDIA GPUs as part of its national sovereign AI strategy.
Tabby, which operates across the GCC with over 20 million registered users and 40,000 merchants, said the AI factory will drive measurable improvements in operational efficiency and customer experience. The company’s growing AI footprint also enhances its risk controls and personalization engines, key differentiators in the competitive fintech market.
The initiative comes months after Tabby’s $160 million Series E round, which valued the company at $3.3 billion, and amid reports of preparations for an eventual IPO. NVIDIA’s HGX systems—engineered for high-throughput AI and HPC workloads, will allow Tabby to handle complex model training at scale.
The launch aligns with Saudi Arabia’s broader goals to localize digital infrastructure, advance sovereign AI capability, and attract private sector investment into the Kingdom’s expanding tech ecosystem.