Amazon Details Technology Powering Amazon Luna’s Cloud Gaming Infrastructure

Amazon has shared a rare behind-the-scenes look at the technical architecture powering its cloud gaming service, Amazon Luna. In a new AWS for Games blog post co-authored by Tyler Ng of the Amazon on AWS Marketing team and Richard Lawrence, Director of Engineering for Amazon Luna, the company detailed how the service manages latency, optimizes hardware utilization, and scales its infrastructure to support millions of game sessions every night.

GPU Sharing and Fleet Utilization

One of the most significant revelations is Amazon Luna's use of a multi-tenancy system to maximize hardware efficiency. Rather than dedicating an entire GPU to a single player, Luna can run four or more simultaneous game sessions on a single GPU. Each session is allocated a specific slice of the GPU's resources.

This resource-sharing model has allowed Amazon to achieve a 76% active-time utilization rate across its GPU fleet. To ensure that sharing resources does not compromise the player experience, Amazon subjects every game in the Luna catalog to a strict performance certification process. This certification tests whether a game can consistently maintain its target frame rate (typically 60 frames per second) within its allocated GPU slice under peak load. If a game fails to maintain 60fps during testing, it is held back from release until the performance bottleneck is resolved or the game is assigned to a higher-performing hardware configuration.

The Streaming Path and Input Latency

When a player starts a session on Amazon Luna, the service selects an available GPU-equipped Amazon EC2 instance from a global fleet spanning dozens of AWS Regions. This selection is based on network latency to ensure the data travels the shortest distance possible. The selected instance handles game logic, frame rendering, and memory management.

Games are typically executed inside lightweight Linux containers, though Windows instances are utilized in rare cases. The rendered frames are encoded and streamed to the player's device at 1080p and 60fps. Once a gaming session ends, the environment resets completely, making the instance available for the next user. Save data is synchronized across devices via the user's Amazon account, and most games allow players to export their saves to a local PC.

To address input latency, Amazon uses asynchronous input delivery and adaptive bitrate encoding. While standard controller inputs travel back separately from the video stream, the official Amazon Luna Controller utilizes a direct Wi-Fi connection. By connecting directly to the cloud rather than passing through the display device, the controller bypasses Bluetooth overhead and avoids processing delays from the television.

Infrastructure and Future Growth

The blog post officially confirms that Amazon GameLift Streams is the core technology powering Amazon Luna’s game-streaming infrastructure. This underlying technology supports a massive scale, with Amazon reporting that Luna handles millions of game sessions nightly.

Looking ahead, Amazon aims to expand Luna's reach to tens of millions of people within five years, a goal first reported in July. The company plans to integrate the streaming service deeper into its existing ecosystem, including Prime Video, Fire TV, and Twitch. Currently, Amazon has the potential to surface Luna games to more than 200 million Amazon Prime members across various platforms, including the Prime Video app.