Catherine Perloff

Amazon has cut staff from its division developing proprietary large language models, according to a company spokesperson. The spokesperson indicated that while AI models remain a priority, Amazon is refocusing on initiatives deemed most critical. The move signals a potential shift in Amazon's internal AI strategy, though the company has not disclosed the scale of the reduction or specific details about affected teams.

Arcee, an open-source AI developer that committed to spending $8 million over three years with AWS, has been unable to access sufficient Nvidia-powered servers on the cloud provider. As a result, the startup is running most of its models on cloud alternatives including Hugging Face and Together, signaling that AWS capacity constraints are pushing some AI workloads elsewhere.

Amazon Web Services announced a 20% price increase for its EC2 Capacity Blocks service, which provides AI compute rental capacity. The increase affects customers using Nvidia compute resources through AWS. The company cited rising costs across the tech supply chain driven by global compute demand as the reason for the adjustment.

Walmart acquired New York-based advertising technology startup Vibe for $1.4 billion on Tuesday, including an $180 million executive retention payment. The deal values Vibe at more than three times its last private valuation of $410 million from September. Vibe helps small businesses purchase ads on streaming platforms, and the acquisition will expand Walmart's ability to reach advertisers seeking to buy ad inventory on its own site.

Amazon is positioning its custom AI chips, Inferentia2 and Trainium, as lower-cost alternatives to Nvidia's H100 for data center inference workloads. According to a consultant at Co Driver Labs, Amazon's chips can deliver 80% cost savings for comparable tasks. A growing number of enterprises running their own data centers are evaluating Amazon's offerings as Nvidia chip availability remains constrained.

Snowflake has committed to spending $6 billion on Amazon Web Services over the coming years, with a focus on Amazon's Graviton chips and AI infrastructure. The deal represents a significant deepening of the cloud database vendor's reliance on AWS for compute resources. Graviton CPUs are becoming strategically important as businesses seek alternatives to traditional processors for general computing workloads.

Amazon's custom AI chips, Trainium, are gaining adoption among developers after years of positioning as an Nvidia alternative. Major AI labs Anthropic and OpenAI have committed to using significant Trainium capacity through their infrastructure deals with Amazon, and recent software improvements are now attracting smaller developers to consider shifting workloads to the platform. The shift signals that Amazon's hardware efforts may finally be reaching competitive viability in a market long dominated by Nvidia.

Amazon is rebranding its Rufus shopping chatbot to Alexa for Shopping, consolidating its AI assistant strategy around the Alexa brand. The move comes roughly a year after Amazon launched Alexa+, an enhanced version of its decade-old assistant powered by large language models. The rebrand signals Amazon's intention to unify its conversational AI offerings under a single brand rather than maintain separate product lines for shopping and general assistance.

Amazon has announced a deal to bring OpenAI's models to AWS through a new offering for AI agents, but the move comes as AWS customers have increasingly adopted competing models from Anthropic and Amazon's own Nova through the Bedrock service. Six firms working with AWS said they and their clients are satisfied with existing model options available on the platform, and while some may evaluate OpenAI products upon launch, others are already accessing OpenAI through alternative cloud providers. The announcement highlights a shift in customer behavior over the three years since OpenAI sparked the generative AI boom, with enterprises now comfortable diversifying their model dependencies rather than defaulting to OpenAI.