OpenAI has reportedly started leasing Google's TPU chips to handle inference for ChatGPT and other services. The move marks its first major compute shift beyond Nvidia hardware, aiming to rein in spiraling inference costs and diversify a single-vendor supply chain.
NVIDIA just released early developer previews of Isaac Sim 5.0 and Isaac Lab 2.2 on GitHub. The updates give robotics builders physics-based simulation tools to train and validate robots, adding open-source access, synthetic data generation extensions, and new robot models.
Nvidia just teamed up with Deutsche Telekom to launch Germany’s largest-ever AI deployment—an industrial AI cloud packing 10,000 Blackwell GPUs in its first phase. Operated by Deutsche Telekom, the setup will power cognitive robotics and simulation-heavy digital twins for European manufacturers, marking a crucial step toward European sovereign AI.
NVIDIA and Stability AI just introduced TensorRT FP8 optimizations for Stable Diffusion 3.5, slashing Large model VRAM needs from over 18GB down to 11GB while boosting performance by up to 2.3x. Developers can also grab the new standalone TensorRT for RTX SDK with JIT compilation. This basically puts heavyweight image generation within reach for standard consumer RTX rigs.
Cisco is rallying behind new president Jeetu Patel, whose aggressive product overhaul and deep Nvidia AI partnerships have pushed shares to 52-week highs. After years of bloated portfolios and sales-first culture, insiders see Patel's technical streamlining as the exact catalyst Cisco needs to modernize for the AI era.
Mistral AI just unveiled Mistral Compute, a private full-stack AI infrastructure platform spanning GPUs, orchestration, and APIs from bare metal to fully managed PaaS. Partnering with NVIDIA, it aims to offer an independent, sovereign European alternative to major US and Chinese cloud providers.
DeepL just deployed Nvidia's DGX SuperPOD packed with B200 Grace Blackwell chips, cutting the time needed to translate the entire internet from 194 days to just 18. The $2 billion startup plans to leverage the massive compute boost to train next-gen translation models and power tools like Clarify.
Cisco just overhauled its networking portfolio across data centers and enterprise campuses to address the crushing bandwidth demands of modern AI workloads. The rollout includes a unified Nexus Dashboard equipped with a troubleshooting AI assistant, new 400G optical modules, and SiliconOne-powered campus switches. Cisco is positioning itself as the complete plumbing provider for the agentic AI era.
AMD just acquired Sacramento-based photonics startup Enosemi to bolster its in-house AI hardware development, with terms kept under wraps. The move centers on co-packaged optics (CPO), which fuses optical transceivers directly into switch silicon to kill off bulky pluggable transceivers. Why it matters: Optical interconnects are essential for scaling next-gen AI clusters, letting AMD better take on Nvidia's unified rack systems.
NVIDIA just released Llama Nemotron Nano 4B, an open reasoning model tailored for edge agents. It delivers higher accuracy and 50% faster throughput than other sub-8B open models across coding, math, and function calling. It runs directly on Jetson and RTX GPUs for low-cost local reasoning.
Google just introduced SynthID Detector, a web portal that scans text, audio, images, and video to identify SynthID watermarks and pinpoint modified segments. Over 10 billion pieces of AI content already carry the mark. The tool is rolling out first to waitlisted journalists, media pros, and researchers.
NVIDIA CEO Jensen Huang just took the stage at COMPUTEX to declare that computing centers are officially turning into token-producing "AI factories." Backing up predictions of a multi-trillion-dollar infrastructure boom, NVIDIA unveiled the semi-custom NVLink Fusion architecture, DGX Spark desktop supercomputers, and a physics engine built with DeepMind and Disney. It is a full-stack push to cement NVIDIA as the bedrock of both digital and physical AI.
Nvidia just showcased broad partner adoption for its single-chip Grace CPU C1 at COMPUTEX, drawing support from Foxconn, Supermicro, and Quanta. The processor claims twice the energy efficiency of traditional CPUs, targeting power-tight setups across edge computing, storage, and 5G AI-RAN. Why it matters: Nvidia is pushing Arm-based efficiency beyond hyperscaler data centers into distributed physical infrastructure.
NVIDIA just rallied major storage makers and ODMs around its AI Data Platform reference design to power agentic AI. By embedding GPUs and DPUs closer to massive enterprise data stores, the new hardware helps agents index, retrieve, and reason across private docs in real time without bottlenecking.
Nvidia just announced Isaac GR00T N1.5 and a synthetic motion blueprint called GR00T-Dreams at Computex. Taking a single image as input, it generates simulation videos to train robot behaviors in just 36 hours instead of three months of manual recording. Why it matters: Tackling the data shortage makes scaling humanoids in factories and homes far more practical.
Nvidia just introduced NVLink Fusion, a new silicon interconnect that lets hyperscalers hook up custom ASICs and CPUs directly into Nvidia's AI rack architecture. Heavyweights like MediaTek, Marvell, and Qualcomm are already on board. Why it matters: Big tech no longer has to pick between internal custom silicon and Nvidia's computing fabric—they can now stitch both together in the same AI factory.
LM Studio just rolled out version 0.3.15, bringing CUDA 12.8 support and serious performance upgrades for NVIDIA GeForce RTX users. With CUDA graphs and Flash Attention kernels baked into llama.cpp, local model throughput jumps by up to 35% across RTX 20 Series through Blackwell GPUs. The update also adds granular tool_choice controls and an overhauled system prompt editor for running offline agents.
ServiceNow and Nvidia just dropped Apriel Nemotron 15B, a compact open-source reasoning model built to run enterprise AI agents at lower latency and inference costs. Delivered as an Nvidia NIM microservice, it pairs with a new joint data flywheel architecture that feeds real enterprise workflow data back into model training.
Nvidia just launched Parakeet-TDT-0.6B-v2, an open-source speech recognition model that can transcribe an hour of audio in just one second on GPU hardware. The 600M-parameter model tops the Hugging Face Open ASR leaderboard with a 6.05% word error rate, approaching proprietary rivals. Best of all, it is released under a permissive CC-BY-4.0 commercial license.
Nvidia is designing yet another round of compliant AI processors for China after Washington effectively barred its H20 GPUs, forcing a $5.5B inventory charge. The chipmaker has already briefed Alibaba, Tencent, and ByteDance on its new roadmap. Why it matters: With Huawei ramping up mass shipments of its Ascend 910C to local tech giants, Nvidia is racing against time to prevent its China data center revenue from crashing to zero.
