Jump to Content
AI & Machine Learning

Cash App uses Google Cloud to power mobile payments innovation and research

February 26, 2021
Christin Brown

Global Financial Services Strategy & Solutions Lead, Google Cloud

Mobile payments are creating opportunities to reach and benefit more people worldwide by providing services to underbanked communities, and empowering streamlined services in e-commerce and brick-and-mortar stores.

Square, a U.S.-based financial services company that specializes in payment software and hardware products, currently stands at the cutting-edge of this industry.

The company regularly pursues more powerful, advanced financial services solutions, like its burgeoning consumer finance service Cash App. Cash App has been a particularly active source of innovation. Last year, Square acquired an artificial intelligence (AI) research firm Dessa to bolster Cash App’s existing solutions and drive innovative new mechanisms to improve customer experience and increase accessibility to banking services.

Cash App opted to use Google Cloud AI and machine learning (ML) solutions and NVIDIA’s graphics processing units (GPUs) to handle the immense compute demands of its applied AI efforts.

Establishing a foundation for breakthroughs in Artificial Intelligence

Dessa has a long history of applying AI to what it calls “Bananas” – novel and ambitious projects that use emerging machine learning technologies to solve problems in new ways, ultimately driving real-world impact. 

Dessa has used Google Cloud’s AI Platform services which were configured and made available by Square’s Platform Infrastructure Engineering group to Square’s internal needs. The services enable data scientists at Square to carry out these data-heavy, processing-intensive initiatives. Dessa works with Deep Neural Networks (DNNs), which come with long training times and data volume requirements that can make new experimentation and ideation challenging. DNNs are more resource intensive, but help to solve many of the computer training problems that AI/ML practitioners sometimes face. 

While Cloud Storage helped to alleviate some of the challenges associated with storage of raw and analytical data, the speed with which information could run through and between GPUs was also a sticking point. 

“Google Cloud gave us critical control over our processes,” said Kyle De Freitas, a senior software engineer at Dessa. “We recognized that Compute Engine A2 VMs, powered by the NVIDIA A100 Tensor Core GPUs, could dramatically reduce processing times and allow us to experiment much faster. Running NVIDIA A100 GPUs on Google Cloud's AI Platform gives us the foundation we need to continue innovating and turning ideas into impactful realities for our customers.”

NVIDIA stepped in to identify bottlenecks in these processes and implement the A100 to experiment with large datasets and push out new models more quickly. The NVIDIA A100 GPU delivers 20X more compute capacity than the previous generation, along with a new TF32 precision, Multi-Instance GPU (MIG) feature and support for accelerating structural sparsity.

Google Cloud AI and NVIDIA were able to deliver a roughly 66% improvement to the processing time it takes to complete a core ML processing workflow. 

NVIDIA has also provided Dessa with developer support to improve ML engineer skills, remove bottlenecks, and overcome challenges in real time. NVIDIA developer support, GPUs, and AI Platform on Google Cloud have also improved the speed and quality of Cash App services to customers.

For example, Dessa would generally need about six hours to process two terabytes of data and complete training for a single epoch, or the total passes of a dataset that an ML algorithm has completed. Now, it can complete processing seven terabytes of data in under two hours. Considering the fact that Dessa runs between 10 and 20 epochs at a time, some of which involve training with 350 million parameters, this 10X acceleration delivered by the NVIDIA A100 has proven invaluable.

“NVIDIA GPUs and AI Platform have given us value by scaling up to deal with data and the volume associated with it, while Dataflow gives us the speed to capitalize on event data in real-time,” said De Freitas.

Further embedding AI into Cash App

Because Cash App has put so much effort into maturing its AI/ML capabilities, it is now better positioned to effect real change in the communities it serves.

“We’re focused on providing financial support across communities, like the ability to share resources in a secure, inclusive, and traceable manner through advanced ML technologies,” said De Freitas. 

Through Dessa’s experimentations and innovations, Cash App and Square are furthering efforts to create more personalized services and smart tools that allow the general population to make better financial decisions through AI.

Posted in