Artificial Intelligence
-
Demis Hassabis says recent AI-driven security incidents are early warning shots. The bigger issue is that frontier agents increasingly need to be treated like privileged insiders, not ordinary software.
-
A full comparison of CPUs, GPUs, NPUs and TPUs: architecture, programming models, latency, throughput, energy efficiency, memory and the workloads each processor handles best.
-
The supercomputer leaderboard changed in 2026. China is back at No. 1, Europe has crossed the exascale line, and five machines now sustain at least one quintillion floating-point operations per second. This countdown uses the June 2026 TOP500 and ranks systems by HPL Rmax — measured sustained performance, not theoretical peak. For more on the…
-
Tensormesh has launched an enterprise platform that reuses LLM KV cache across GPU memory, CPU memory and external storage to reduce repeated inference work.
-
Salience Labs has raised $125 million to scale silicon-photonic optical circuit switches designed to cut latency and power use inside AI data centers.
-
Airoha is bringing a 3-TOPS AI accelerator into its 25G PON broadband platform so gateways can diagnose and repair some network faults locally.
-
NVIDIA is committing $250 million in computing access over five years to support UK research into agentic AI safety and security.
-
ASML says AI-driven customer demand is now set to exceed its planned chip-tool supply in 2027 and 2028, with some customers discussing orders for 2029 and 2030.
-
Wikimedia says AI agents it believes were operated by OpenAI made unauthorized wiki edits, probed Etherpad, and generated heavy traffic—without evidence of a system compromise.
-
MEGMEET is showing a 1 MW 800 VDC power rack and liquid-cooled power hardware for next-generation AI data centers at OCP Global Summit 2026.
-
NVIDIA says its October DLSS 5 update doubles neural-rendering model performance while reducing model size and 4K VRAM use.
-
NVIDIA says its new Nemotron 3.5 Bolt 74B model will arrive on October 15 for local AI agents running on RTX Spark PCs. In the company’s October 13 software update, NVIDIA describes the model as a 74-billion-parameter mixture-of-experts system with 9 billion active parameters, compressed into 24GB using 2-bit quantization. NVIDIA says developers will be…
