Paying it forward: PayPal’s journey to sustainability

Online payment pioneer PayPal annually performs tens of billions of transactions totaling more than $1 trillion — giving it more than 50% market share. 

Given the nature of its business, the company is moving untold amounts of data back and forth at any moment, while concurrently seeking to derive real-time insights from it. PayPal’s main priority is doing both as quickly and efficiently as possible — latency, gravity, capacity and performance are all top concerns. 

The online payments giant is juggling all this alongside aggressive sustainability goals: It has committed to 100% carbon neutrality by 2040 and aims to reduce its operational greenhouse gas emissions by 25% by 2025 (from 2019). 

“As we grow, the reality is that we do need more computational power, more storage — how do we do that in a responsible way?” PayPal EVP and CIO Archie Deskus told VentureBeat, adding that the journey is never complete. “Our aspiration is to keep being as efficient and optimized as possible.”

Scaling and ‘bursting’ to meet demand

PayPal has a hybrid data center infrastructure: It is both on-premises via data center partners while also leveraging the public cloud. As Deskus described it, the brand was “born cloud native,” and part of its strategy over the last few years has been implementing more into the public cloud. 

This was accelerated with a significant uptick in digital commerce — and subsequent traffic — during the pandemic. PayPal worked with Deloitte to exit from its non-strategic data centers and shifted the horizontally scalable applications of its payment platform and transactions to Google Cloud.

PayPal’s greatest processing need is runtime for applications in deployment, Deskus explained. The critical piece there is that the company has peak periods — such as Black Friday and Cyber Monday — during which it sees significant multipliers in its transactional volumes. During these peak periods, it processes an average of 1,000 payments per second. 

Now with Google Cloud, the company can “scale and burst,” allowing it to handle such increases without experiencing idle capacity (when servers are essentially incapacitated and not delivering information or computing services). 

As Deskus explained, the company relies on its data center resources for cost reasons for baseline needs, and then leverages this bursting ability whenever exceeding those. 

Analytics at (or as close to) the source

Another critical processing need is analytics. 

Data gravity — the tendency of data to attract more data and applications and make it more difficult to move — is important here, Deskus noted. Because the company has committed to renewables and efficiency, it wants to ensure that they are close to analytics workloads. 

For this reason, PayPal has paired with Google’s cloud region in Salt Lake City, Utah, and is migrating key elements of its infrastructure into the region. The core stack is “tightly coupled” to keep latency “at an absolute minimum,” Deskus said.

“At PayPal, we process a tremendous amount of real-time data for our customers,” said Deskus. “Having data separated all over the place is going to cause problems in terms of performance and latency.”

The big part of the company’s journey with Google Cloud is bringing those analytics together, she said The company anticipates that migration to be completed in the first half of 2024, she added. 

Additionally, either due to acquisition or decentralized operating models of the past, PayPal has a multitude of analytics tools across its portfolio that it is working to converge and consolidate to get a better view of customers across its brands. 

The company is continuously looking at asset utilization and determining which can be deprecated. However, decoupling investments without impacting latency is critical, Deskus said. 

While PayPal has previously committed to a zero data center ownership, Deskus acknowledged that a uniform structure could prove problematic, notably due to acquisitions with varying data center constellations. 

Evaluating generative AI’s use, massive data requirements

Following close on the heels of the explosion in cloud computing are technologies and processes such as artificial intelligence (AI), machine learning (ML) and big data analytics. 

Generative AI in particular creates massive amounts of data and requires high-performing and efficient data storage and retrieval.

“This has been building,” said Deskus. “Gen AI certainly is upping the game in terms of how much more capacity is going to be needed.”

However, she noted that the technology is still in its formative stages, with many companies only now experimenting and determining its value and required resources. PayPal, for its part, is looking at ways to scale for performance — such as via high-performance computing (HPC) tools, GPUs or distributed computing clusters. 

Moving forward, as the company learns through experimentation it will be able to identify the most efficient algorithms and data requirements, she said. It is also important to analyze complexity: When is GPT-4.5 (or forthcoming iterations) more appropriate versus GPT-4?

“The belief is always that more data is better, and I think as we get more mature we find that that’s not always true,” she said. She added that the tendency is to have the “latest and greatest and best,” when technologies should ultimately be evaluated on a case-by-case basis.

“It’s not just creating new capability, but how are we looking at these other aspects and true lifecycle management?”

Another key component in adopting such tools is having the right skill sets in place. The complexities of gen AI and ML require investments in training and hiring to ensure people can tackle new, evolving opportunities and challenges, Deskus said.

“We saw this with cloud,” she said. “We started on that journey, we didn’t have all the skills we needed. We learned. We learned through some of our mistakes, we learned that we didn’t always do things in an optimal way.”

Generative AI’s power grab

Together, cloud, AI, big data analytics and other technologies are driving ever-increasing demands for computational resources. A growing challenge for large enterprises is keeping up with those power-hungry technologies. 

Computing — particularly lots of it going on all at once — gets hot and requires lots of water to cool down. It has been reported that in Iowa, for instance, OpenAI nearly ran waterways dry in developing its groundbreaking ChatGPT. 

“We’re seeing stress in the system around power,” Deskus acknowledged. “Everybody’s trying to grab the available power out there.”

Ideally, companies plan out far enough to not be in a position where they aren’t able to scale as a result of power shortages, or failing to factor in delayed delivery times, she said.

“It’s prudent to understand where those constraints are,” Deskus said, “and [ensure] that we’re planning appropriately so that we don’t end up in a situation where it prevents us from growing our business.”