Engineering Sustainability at Scale: Uber’s Zero-Emission Push
Uber’s commitment to sustainability extends far beyond its well-known ride-hailing operations. By 2030, the company aims to operate as a zero-emission mobility platform in Canada, Europe, and the United States, with a global target set for 2040. This ambition encompasses not only vehicles but also the engineering infrastructure—data centers, cloud deployments, and hardware resources—that underpin Uber’s services. The company’s engineers have embraced the concept of “responsible ownership,” which includes designing efficient systems where energy efficiency and sustainability metrics are embedded into development processes.
In late 2021, Uber Engineering began a structured initiative to integrate sustainable practices into its operations. Progress is tracked across four categories, each representing a stage in maturity. The first stage, awareness and training, involves leadership-backed programs and sustainability questionnaires in pre-development documentation. Developers are prompted to quantify CO2 emissions and outline reduction strategies. The second stage, assisted development, uses continuous analysis and ticketing systems like Jira and ServiceNow to deliver actionable recommendations, complete with quantified impact metrics. The third stage focuses on automated sustainability improvements, such as intelligent tiering of cloud storage to shift infrequently accessed data to lower-energy “cold” storage. The final stage, sustainable by default, ensures that the most environmentally responsible configurations are the default choices during resource provisioning, with exceptions requiring explicit justification.
Central to Uber’s approach is the “Sustainability Stack,” which addresses three layers: software efficiency, hardware optimization, and energy sourcing. Minimizing time and space complexity in code reduces both execution costs and emissions. Hardware must be scrutinized for energy efficiency, avoiding overprovisioning and eliminating idle resources. Finally, energy consumption is examined in context—regulations, renewable sourcing, and environmental impact of land use all factor into deployment decisions.
This framework aligns with a “Shared Fate” model between cloud providers and customers. Providers are tasked with optimizing infrastructure sustainability, managing water resources, and sourcing renewable power. Customers, in turn, must optimize workloads, reduce resource demand, and select providers based on transparent sustainability metrics.
Uber’s best practices reflect industry guidance from Amazon, Google, and Microsoft, enriched by internal experience. These include investing in sustainability metrics, right-sizing workloads, scheduling compute only when needed, leveraging managed services, setting scaling limits and alerts, cleaning up unused resources, deploying in sustainable geographies, refactoring monolithic applications, improving code efficiency, deduplicating storage, selecting energy-efficient programming languages, and adopting new hardware and software offerings. The choice of provider should be informed by certifications, commitments, and tool availability.
Several in-house tools support these practices. The GCP Project Lifecycle service identifies inactive projects—often remnants of deprecated services or abandoned proofs of concept—using Google’s Active Assist. Activity and billing data are enriched, stored in Firestore, and used to generate ServiceNow and Jira tickets for project owners to assess and delete unused assets.
The Geography Recommender integrates sustainability data into deployment decisions. By default, only high-sustainability regions are available for provisioning, with exceptions requiring justification. Recommendations also target existing resources for migration to regions with higher Carbon-Free Energy scores. For example, moving storage from Las Vegas (CFE 19%) to Oregon (CFE 90%) or Iowa (CFE 93%) can yield significant sustainability gains with minimal effort.
The Optimal Utilization Recommender operates across AWS and GCP, using Trusted Advisor and Active Assist to identify underutilized compute instances. Tickets advise owners to scale down or terminate resources, detailing both cost savings and sustainability benefits.
Through awareness, automation, and default sustainable configurations, Uber Engineering has begun to measurably reduce its carbon footprint. The ongoing effort includes refining metrics, prioritizing impactful actions, and collaborating with industry partners to advance tools and practices that support a global transition to clean energy.
