Amazon SageMaker Feature Store introduces two new APIs to improve efficiency and record discovery. The BatchWriteRecord API can handle up to 25 records per call, reducing API calls needed for high-throughput pipelines.
AMD's LMCache extends 4-bit KV quantization beyond HBM to CPU DRAM, boosting goodput by 2x and reducing p95 TTFT by 2.6x in tests.
AMD outlines how LDS optimizations improve performance on MI450 GPUs by addressing layout mismatches and partition contention.
Salesforce reduced infrastructure costs by 8x using SageMaker Inference Components to host multiple models on shared GPUs, but needed a new approach to ensure high availability across Availability Zones.
Decathlon, with 400 million users, adopted Chronos-2 to improve demand forecasting accuracy across 25,000 products globally.
Hugging Face introduces a new storage solution for ML workflows, enabling efficient data handling with 500GB datasets.
Intel's new GPU AI skills tool helps users run, benchmark, and migrate Hugging Face models on Arc and Arc Pro GPUs with one install command.
Creative teams face growing demand for more assets, formats, and revisions, with 78% of leaders reporting capacity issues. Amazon Quick and fal now offer a unified agentic workflow solution.
Anthropic announced a new initiative offering 10,000 free and discounted Claude subscriptions to scientists worldwide starting August 27, 2026.
Hugging Face introduced the Microduck, a $399 duck-shaped robot set to ship before Christmas, as part of its open-source AI hardware efforts.
AMD releases ROCm 10.0, a major update to its open-source GPU compute stack, built for agentic AI applications. The platform now supports a unified developer experience across multiple architectures.
AWS and NVIDIA reduced ASR inference costs by 75% using NVIDIA MPS on Amazon EC2, cutting GPU instances from 16 to 4.