MIT researchers unveil Attention Matching, a groundbreaking KV cache compression technique that slashes large language model memory usage by up to 50x in seconds, delivering near-identical accuracy without the slow, compute-heavy optimization methods that have long bottlenecked AI deployment.
Apple's 18-core M5 Max stuns in Geekbench 6 benchmarks, outscoring AMD's 96-core Threadripper Pro, though experts caution the test favors fewer, faster cores and doesn't scale beyond 32 threads — still, the chip's 29,644 multi-core score and 614 GB/s memory bandwidth signal a major leap forward for Apple silicon.
Red Hat is making major strides in open AI infrastructure, emerging as the third-highest global contributor to PyTorch while driving innovations in multi-cloud inference, distributed model serving, and hardware-agnostic AI deployment through projects like vLLM and llm-d.
SRAM-centric AI chips are challenging GPU dominance as NVIDIA's $20B Groq IP licensing deal and Cerebras's 750 MW OpenAI partnership signal a major shift toward faster, lower-latency accelerators purpose-built for AI inference workloads.
Arm's technology, found in 99% of smartphones worldwide, is revolutionizing on-device AI by bringing real-time generative features like creative style transfer at 30 frames per second to mid-range and entry-level devices, with a TECNO collaboration and KleidiAI optimizations delivering 30% faster AI compute across all price ranges.
Nvidia CEO Jensen Huang signals a major shift in private AI investment strategy, confirming the company's $30 billion stake in OpenAI — part of a massive $110 billion funding round — will likely be its last before OpenAI's anticipated IPO later this year, with a separate $10 billion Anthropic investment also expected to be Nvidia's final private stake in the competing AI startup.
Alibaba debuts two AI smartglasses at MWC — the Qwen Glasses S1 and G1 — boasting a Snapdragon AR1 chip, 12MP camera, real-time translation, and a swappable dual-battery system, though the monochromatic green HUD display raises questions about whether the hardware can keep pace with full-color competitors.
Alibaba launches the Qwen3.5 Small Model Series, a lineup of efficient AI models ranging from 0.8B to 9B parameters built for edge and on-device deployment, with the 9B flagship using Scaled Reinforcement Learning to rival models up to 10x its size and a native multimodal 4B model delivering strong visual reasoning at a fraction of the compute cost.
Apple's new M5 MacBook Pro delivers a massive 4x AI performance boost with upgraded M5 Pro and M5 Max chips capable of running local AI models up to 90B parameters, but buyers will pay $200–$400 more, with prices now reaching up to $3,899.
Alibaba's newly released Qwen3.5 Small Model Series stuns the AI world as its compact 9B flagship model outperforms OpenAI's massive 120B system on key benchmarks, running on a standard laptop while offering free commercial use under Apache 2.0.