How Booking.com Serves 400 Billion AI Predictions Daily

TL;DR: Booking.com revealed how it built an AI platform that handles 400 billion predictions daily. The company shared its 20-year journey from older systems like Hadoop to a modern, high-speed cloud architecture for advanced personalization.
Key facts
- Category
- Infrastructure
- Impact
- High
- Published
- Source
- InfoQ
Full summary
Booking.com detailed its 20-year journey to an AI platform serving 400 billion daily predictions, moving from Hadoop to the cloud.
Booking.com has pulled back the curtain on the two-decade evolution of its machine learning infrastructure, revealing how it built a system capable of handling 400 billion predictions every day. In a presentation covered by InfoQ, engineer Jabez Eliezer Manuel detailed the company's long and complex journey from simple databases to a sophisticated, high-performance cloud platform. This massive scale is what powers the site's personalization features, from hotel rankings to travel recommendations. The story provides a rare, unpolished look at the engineering challenges involved in operating AI at the highest level, highlighting a strategic migration away from legacy technologies to support modern, real-time demands and future AI ambitions.
Technically, Booking.com's journey mirrors the evolution of the big data industry itself. The company initially scaled by optimizing traditional MySQL database topologies but eventually hit a wall. Like many tech giants of its era, it then adopted the Hadoop ecosystem to manage its vast and growing datasets. However, the presentation's core lesson is about the company's decision to eventually sunset its entire Hadoop infrastructure. This was not a simple “lift and shift” to the cloud but a fundamental re-architecture. The team built new, specialized systems from the ground up, including high-throughput model serving platforms and centralized feature stores. These components are critical for delivering fast, consistent, and reliable AI predictions, enabling the platform to respond to user interactions in real time across its global user base.
This transition from on-premise, batch-oriented systems like Hadoop to real-time, cloud-native platforms is a defining trend for established technology companies. Booking.com's experience is a case study in managing significant technical debt while maintaining a business-critical service. Many organizations are grappling with similar challenges as they find their older data stacks are too slow and rigid for the demands of modern AI and machine learning. The move reflects a broader industry understanding that successful AI products depend as much on robust, scalable infrastructure as they do on cutting-edge algorithms. Companies that fail to modernize their underlying data platforms risk being outmaneuvered by more agile competitors who can iterate and deploy new AI-powered features more quickly.
For developers, CTOs, and architects, the key takeaway is that scaling AI is fundamentally an infrastructure and data engineering problem. Booking.com's story demonstrates that long-term success requires a willingness to make difficult, large-scale changes to foundational systems. The investment in a modern platform not only solves today's performance challenges but also unlocks future possibilities. Manuel noted that the company is now building a new “agentic AI foundation” on top of this modernized infrastructure. This signals a strategic shift towards developing more advanced, autonomous AI agents, moving beyond simple personalization. Having done the hard work of rebuilding its core platform, Booking.com is now positioned to compete in the next wave of AI innovation.
Why it matters
This case study provides a rare, detailed look at modernizing a massive, business-critical ML platform. For engineers and architects, Booking.com's journey from Hadoop to a high-throughput cloud architecture offers a practical blueprint for tackling technical debt and scaling real-time AI systems.
Business impact
Booking.com's investment in its AI platform directly powers its core business of personalization and recommendations at immense scale. This infrastructure modernization reduces operational overhead, increases development velocity, and enables the company to build next-generation AI features, strengthening its competitive advantage.
Tags
Related on Notifire
Related stories
Primary source: InfoQ