Amazon is reportedly working on an upgraded version of #Alexa that would expand on the capabilities of its existing Alexa+ assistant. Internal planning documents detailed a project codenamed Moonraker, designed to handle more complex interactions containing multiple simultaneous requests.
The primary focus of Moonraker is building out Alexa's agentic AI capabilities. Examples listed in the document include asking the assistant to book a ride and text a friend in a single interaction, enabling #Amazon to better compete with agentic models from Google, #Anthropic, and OpenAI.
However, this ambition comes with massive infrastructure costs. Amazon projected GPU costs of over $100 million in 2026, prompting internal discussions to either delay or scale back Moonraker's scope. Some senior executives reportedly believe the company has already overspent on the AI models powering the current iteration of Alexa.
A separate document from late 2025 indicates that Amazon planned to utilize hundreds of NVIDIA GPUs to power the Moonraker project, while integrating Anthropic's Claude Sonnet model to test advanced reasoning and visual responses.
While Alexa+ officially launched in the US in early 2026, the rollout has faced headwinds. Users reported issues with handling basic tasks and retaining context during multi-step processes, such as booking rides via Uber. Despite these hurdles, Amazon continues to invest in adding natural language features and customized personalities.
[AgentUpdate Depth Analysis] Amazon's "Moonraker" project represents a crucial pivot from a reactive voice command tool to a proactive, multi-step AI Agent. While competitors like OpenAI and Google focus heavily on desktop and web-based agents, Amazon's unique advantage lies in its ubiquitous smart-home ecosystem. However, transition to true agentic workflows highlights a massive bottleneck: high orchestration complexity and skyrocketing infrastructure costs. By leveraging Anthropic's Claude Sonnet for testing, Amazon implicitly acknowledges the gap in its in-house LLM capabilities for complex reasoning. The future of consumer-facing AI Agents will belong to those who can efficiently manage multi-agent orchestration while optimizing token costs, turning ambient computing into a truly frictionless utility.
