Amazon is quietly advancing a new Alexa AI project codenamed "Moonraker" aimed at allowing the assistant to perform more complex, multi-step tasks and process multiple related requests at once, according to internal documents revealed in July 2026 [1, 2]. The project is designed to push Alexa deeper into the AI agent race by enabling it to execute several interlinked actions in a single user interaction, such as simultaneously booking a ride and sending a text message [2, 3].
Moonraker represents one of the most expensive elements of Amazon's Alexa+ platform overhaul, with internal estimates forecasting over $100 million in GPU costs for 2026 alone [2, 3]. The project relies on deploying hundreds of NVIDIA GPUs and testing advanced AI models, including the Anthropic Sonnet model, which provides enhanced reasoning abilities and visual response capabilities [3].
Although Amazon officially launched the Alexa+ assistant nationwide in the US earlier this year, the rollout faced delays and some performance issues, including hallucinations, inconsistent replies, and difficulties managing certain types of requests [2, 3]. Some internal leaders at Amazon have expressed concerns about the high operational costs of running the AI models behind Alexa, feeling that the spending may be excessive given current challenges [2, 3].
Despite these hurdles, Amazon remains committed to developing the Alexa+ platform, continuously adding new features like diverse personality styles and improved natural language order processing [2, 3]. CEO Andy Jassy acknowledged Alexa's ongoing development by stating, "Alexa is still early in its journey to be the world's best personal assistant" [2].
Details about the Moonraker codename and its scope were previously unreported until internal documents surfaced this year [1, 2, 3]. The project timeline shows plans to deploy GPU infrastructure and test AI models as early as December 2025, followed by the national Alexa+ launch in early 2026 and the disclosure of Moonraker’s costs and aims in July 2026 [1, 2, 3].
