Amazon Bedrock Expands Global Claude Inference
Amazon Web Services has expanded global cross-region inference on its Bedrock platform for Anthropic's Claude models. The service is now available in the Middle East, including the UAE and Bahrain, as well as several countries in Southeast Asia such as Thailand, Singapore, and Taiwan. The expansion aims to provide lower-latency and sovereign AI capabilities for regulated industries.
- The expansion is part of a larger trend of hyperscalers deploying infrastructure to comply with growing data sovereignty regulations in the Middle East and Southeast Asia, which mandate that sensitive data be stored and processed within national borders. - Amazon Bedrock is the first and currently only managed service to offer all three of Anthropic's Claude 3 models—Opus, Sonnet, and Haiku—providing customers with a range of options to balance intelligence, speed, and cost. The newer Claude 3.5 Sonnet is also available, outperforming the more expensive Claude 3 Opus on several benchmarks at twice the speed. - AWS is competing in a fierce generative AI market, where it holds roughly 30% of the overall cloud market share, but trails Microsoft (23% share) in new generative AI projects. Google Cloud Platform follows with about 13% of the market. - To power services like Bedrock and reduce reliance on NVIDIA, AWS is heavily investing in its own custom silicon, including Trainium chips for training and Inferentia chips for inference, which can offer up to 50% better price-performance for some AI workloads. - Anthropic and AWS have a strategic collaboration that includes Anthropic using AWS Trainium and Inferentia chips to build, train, and deploy its future models, and a joint project to build a massive training cluster with over 500,000 Trainium2 chips. - The cost of using Claude models on Bedrock is based on a pay-as-you-go model, priced per input and output token, which differs from a direct Anthropic API subscription and can be optimized using provisioned throughput for high-volume, predictable workloads. - The Claude 3 models on Bedrock feature advanced vision capabilities, allowing them to process and analyze diverse visual formats like charts, graphs, and technical diagrams, not just text. - This global expansion is critical for latency-sensitive applications in regulated industries like financial services and the public sector, where near-instantaneous responses and local data processing are often mandatory.
