To boost the optimization of ride-sharing and on-demand delivery services, two segments where speed and matching accuracy matter dearly, Uber expands its continuous collaboration with Amazon’s AWS, testing new chip models for smarter AI predictions and handling demand spikes in real-time.
AWS today announced a scale-up of its cooperation with global ride and delivery giant Uber, offering Graviton4 and Trainium3 chip capabilities to back Trip Serving Zones, customer personalisation and faster rider and delivery matching.
The freshest update is Uber’s experimentation with AWS Trainium3 to train some of the AI models that help power its apps. These models are supposed to learn better prediction and matching skills the more trips they analyze. This way, Uber hopes to deliver even faster matches, more accurate arrival time estimates, and more personalized recommendations to customers worldwide. As this requires enormous computing power, Trainium was chosen as an efficient, cost-effective option to power the improved app experience. At the same time, Uber continues using Graviton tools as well.
In essence, AWS Graviton is a family of processors designed specifically for cloud workloads running in Amazon ecosystem. Uber uses them for real-time infrastructure behind every ride and delivery: the one which has milliseconds to decide which driver is closest to the starting point or delivery place, which route should be used for a planned ride depending on the changing road traffic conditions, what are estimated time and price of the trip, or even which option would a particular customer prefer based on their history.
Known as Trip Serving Zones, this unseen decision-making engine is part of Uber’s larger system that makes sure every ride and delivery runs smoothly, meanwhile requiring to make millions of predictions and processing location data in milliseconds. In the expanded collaboration scenario, Uber will rely on AWS Graviton4 latest processor series with more of these real-time workloads that require AWS compute, storage, and networking.
“Uber operates at a scale where milliseconds matter,” said Kamran Zargahi, vice president of engineering at Uber. “Moving more Trip Serving workloads to AWS gives us the flexibility to match riders and drivers faster and handle delivery demand spikes without disruption.”
Besides faster decision speed, the ride-hailing company expects reduced energy consumption while the engine is scaling rapidly during demand spikes from a new chip functionality. As one knows, AI infrastructure needs lots of energy to maintain its rela-time efficiency and self-improve existing models.
Per rough estimations, increased AI usage at different stages of corporate operations is predicted to triple U.S. data-centre usage by the end of the decade, increasing energy consumption from 150-175 terawatt hours (TWh) in 2023 to 560 TWh, or 13% of current U.S. electricity demand.
To handle increasing demand for cloud and AI services, AWS is regularly pouring billions of dollars into its data centre network across the globe. Despite the absence of official data, some reports estimate the number of data centres supporting AWS ambitions in hundreds (possibly ~900) locations worldwide.
Uber, in turn, uses AI for all areas of its operations, be it ride-matching or payments. In the latter case, the company partnered with Checkout.com to access its AI-powered payment optimization technology, improving authorization rates, reducing payment failures, and ensuring a faster, more seamless experience for customers worldwide.
