Amazon Trims Nova AI Models While Doubling Down on $200B Infrastructure
The e-commerce giant is winding down development on several Nova models while redirecting resources to frontier research, even as capital expenditure soars.
Amazon is reshaping its AI strategy: the company is winding down active development on several Nova models while maintaining its massive infrastructure buildout, according to Business Insider reporting.
Amazon is moving Nova Premier, Omni, Reel, and Canvas to maintenance-only status. The company continues to support and develop Nova 2 Lite, Nova 2 Sonic, and Nova Act, all available through Amazon Bedrock.
Infrastructure Spending Remains Massive
Despite the model consolidation, Amazon's commitment to AI infrastructure shows no signs of slowing. The company's 2026 capital expenditure guidance sits at approximately $200 billion, primarily allocated to data centers, custom AI chips like Trainium and Inferentia, and model training infrastructure. For context, 2025 capex totaled $131.8 billion.
This spending places Amazon squarely in the AI arms race alongside Microsoft, Google, and Meta. The key distinction: Amazon operates both as a model developer and as an infrastructure provider, hosting third-party models like Anthropic's Claude on Bedrock while maintaining its own proprietary Nova line.
Strategic Pivot or Pragmatic Consolidation?
Resources previously dedicated to the wound-down Nova models are being redirected to a new "Frontier Model Research" initiative, according to Business Insider's July 2026 reporting. The move suggests Amazon is concentrating its model development efforts rather than abandoning in-house AI entirely.
The consolidation raises questions about the economics of frontier model development. Building competitive large language models has proven more difficult and costly than many companies initially anticipated. By narrowing its Nova portfolio while expanding infrastructure capacity, Amazon may be positioning itself as the picks-and-shovels provider for the AI economy, even as it maintains selective model development capabilities.
What This Means for Investors
The $200 billion annual capex figure represents a significant escalation from prior years. Investors watching for return on this spending will be monitoring whether Amazon's AI services revenue accelerates accordingly. The company's dual role—as both model creator and infrastructure host—could prove advantageous if proprietary model development proves less critical than providing the compute backbone for the broader AI ecosystem.
Amazon did not immediately respond to requests for comment on the Nova restructuring. The company's next earnings report is expected to provide further clarity on AI-related revenue trajectories and whether the infrastructure buildout is translating into measurable business outcomes.
For now, the message appears clear: Amazon is not exiting the AI race, but it is recalibrating how it competes.