AI Chips news and analysis on Web Pulse: 6900 curated articles tracking the latest AI Chips developments, tools, and research, updated continuously from vetted sources.
A developer compared five GPU cloud providers for LLM inference workloads, finding Vast.ai cheapest at roughly $1.20/hr for an A100 80GB, ahead of Lambda Labs (~$1.50), RunPod (~$1.89), GCP (~$2.90) and AWS (~$3.20). For…
Reflection AI Inc. introduced Beam, an open-source large language model with 501 billion parameters, which the company says matches or beats GLM-5.2 on some tasks while using between one third and one fourth the hardware…
AMD introduced Typical Attained FLOPs (TAF), a third GPU performance metric that measures FLOP-weighted sustained throughput across a fixed representative GEMM set selected before testing, in the third blog post of its s…
Cisco released ORv3-compatible trays for its N9000 Series switches, letting operators mount the networking hardware in Open Rack Version 3 environments built around a 21-inch equipment bay, Open Units of 48 mm, and 48V D…
A home server can be built on decade-old hardware, with reused Dell OptiPlex 990 and Precision Tower 3620 PCs and older GPUs still handling media transcoding and roughly 30-billion-parameter AI models, according to a Lev…
A Stanford study of local AI found that intelligence per watt across more than twenty local models and eight accelerators improved 5.3-fold from 2023 to 2025, while the share of single-turn chat and reasoning queries ans…
Synopsys struck a revenue-sharing deal with OpenAI to co-develop GPT-Synopsys, an AI model trained and tuned for chip design tasks, Synopsys said at its Investor Day in New York City, as reported by Reuters on September …
IBM is expanding mainframe networking options for IBM Z and LinuxONE, adding a 100GB Ethernet adapter with two 100 GbE ports and QSFP SR optics that plug directly into the Central Processor Complex drawer, eliminating ex…
StorageReview connected an ASUS ExpertCenter Pro ET900N G3 and an HP ZGX Fury AI Station — two NVIDIA GB300 DGX Stations — directly with two 400G direct attach copper (DAC) cables between matching QSFP112 ports, giving t…
CoreWeave Inc. plans to offer standalone Nvidia Corp. Vera CPU capacity alongside its existing infrastructure, with CoreWeave senior director of product Harsh Banwait reporting about a 3x improvement in Sandbox startup t…
HIVE Digital Technologies ran large language model pre-training for Columbia University's Department of Industrial Engineering and Operations Research on Nvidia A40 GPUs located in Asunción, Paraguay, operated remotely f…
Lola Vision Systems, a fabless chip startup founded in 2023 and based in the Washington, DC and Los Angeles areas, is developing the LVS-250, a five-chiplet edge AI processor it says delivers 230 TOPS at INT8 precision w…
DeepSeek and Huawei released open-source programming tools for Huawei's Ascend accelerators, announced on DeepSeek's WeChat account on Wednesday, including compute libraries, chip-to-chip communication libraries, and Asc…
Synopsys announced GPT-Synopsys at its 2026 Investor Day in New York, a specialized AI model built with OpenAI's frontier models that operates directly inside Synopsys EDA tools for chip design and verification workflows…
A developer known as Niko1221 has published Strata, a repository showing that the 125B-parameter Qwen 3.8 Flash Next model can run on a single consumer RTX 4090 at roughly 100 tokens per second. The setup combines 4-bit …
An open reference architecture proposes a physical-AI counting system for Pakistan's steel mills, using IP66 HDR cameras and sealed GPU-accelerated industrial PCs running fine-tuned RF-DETR detection and ByteTrack-class …
Apex Compute, the California-based startup founded in 2024 that is developing high-efficiency AI accelerators for real-time edge AI inferencing, is building an open-source Vulkan driver based on Mesa for its hardware, wi…
Typesafe.ai built MCJev, a Jev-style agent that makes each Minecraft PvP decision in a single forward pass using the DJev model, and its DJev-serve engine runs at about 24 ms per decision on NVIDIA GB200 NVL72 and B200 G…
A technical case study reports that a single NVIDIA RTX 4090 with 24 GB of VRAM can run a 125-billion-parameter Qwen 3.8 Flash Next model at roughly 100 trillion tokens per second, a throughput the writeup says was previ…
DeepSeek and Huawei Technologies unveiled a suite of open-source programming tools for Huawei's Ascend AI chips on September 30, 2026, centered on TileLang, a high-level language pitched as a simpler alternative to Nvidi…