NVIDIA's RTX 5090 offers 32GB of VRAM and 78% more memory bandwidth than the RTX 4090, but the extra performance is most noticeable for specific local LLM workloads such as running 32B models at Q4 wi…
Samsung, SK Hynix, and Micron control 90-95% of the global DRAM market, with a class-action lawsuit alleging price manipulation as conventional DRAM prices surged 700% over four years. The HBM shortag…
A developer building an open-source AI agent runtime called AgentProof argues that trust in autonomous systems requires full observability and evaluation, demonstrating how a flight recorder revealed …
Chinese AI startup DeepSeek is developing its own inference chip to reduce reliance on NVIDIA and Huawei, according to Reuters. The move aims to cut API costs and stabilize supply amid export restrict…
SK Hynix is listing on the Nasdaq in a record $26.5 billion share sale to fund high-bandwidth memory production, betting that AI demand will break the semiconductor industry's historic boom-and-bust c…
GPU architect Raja Koduri's startup Oxmiq raised $35 million on July 1, 2026, to scale its OxCore architecture, a licensable GPU IP block that fuses CPU, GPU, and tensor engines for AI workloads. The …
Researchers introduced two CUTLASS-based kernels that accelerate the SwiGLU activation function in large language models by up to 2.47x on NVIDIA H100 GPUs, shifting workloads from memory-bound to com…
Google Quantum AI published a reinforcement learning system in Nature on July 8 that calibrates its Willow quantum processor in real time, achieving a 3.5x improvement in error stability and a record …
SpaceXAI launched Grok 4.5 on July 8, pricing it over 60% cheaper than Anthropic's Claude Opus 4.8 and OpenAI's GPT-5.5 at $2 per million input tokens and $6 per million output tokens. The model featu…
NVIDIA and Google researchers have developed a host offloading technique in JAX that reduces high-bandwidth memory bottlenecks during LLM training, achieving up to 57% throughput improvements on NVIDI…
Enterprise AI agents are runtime products requiring a control plane, API, release path, rollback story, and ownership for regulated workflows. Focused Labs highlights the NemoClaw Deep Agents Blueprin…
CoreWeave CEO Michael Intrator sold nearly 370,000 shares for $31 million, bringing total insider sales since the March 2025 IPO to over $2.3 billion. The GPU cloud provider reported $2.08 billion in …
The AI industry is facing a shift from computation to memory as the fundamental primitive for intelligent systems, according to Andrej Karpathy. Just as GPUs replaced CPUs as the dominant compute arch…
Amazon is offering the MSI Vector 16 HX AI gaming laptop with an RTX 5070 Ti, Core Ultra 7 255HX, 32GB DDR5 RAM, and a 240Hz QHD+ display at $1,899.99, its best price in 30 days. The deal targets game…
Researchers introduced VTC, a tensor compilation framework that eliminates unnecessary data movement in deep neural networks by using virtual tensors and index mappings. VTC outperforms existing ML co…
The Linux Foundation launched Akrites on June 25, 2026, a shared Security Incident Response Team and coordinated vulnerability disclosure process backed by 19 organizations including AWS, Google, Micr…
Mini PCs with unified memory architectures, such as those using AMD's Strix Halo chip, can run 70-billion-parameter language models that exceed the VRAM capacity of discrete GPUs like the RTX 5090, bu…
NVIDIA is introducing a revenue-sharing and credit-support model for specialized AI cloud providers, financing GPU deployments to bypass hyperscaler custom silicon and secure distribution for inferenc…
ASUS gave a tour of its thermal testing lab in Taiwan, where it tests AI servers in environmental chambers that simulate years of thermal stress in compressed timeframes. The lab includes a walk-in ch…
NVIDIA published a technical blog post explaining how kernel fusion in CUDA can optimize GPU code by combining multiple operations into a single kernel, reducing memory traffic and launch overhead. Th…