Google has added two new products, TPU 8tx and TPU 8i Max, to its eighth-generation lineup of TPU AI chips. Because they are configuration variants built on existing chips, Google says it cut the release timeline to a matter of months. That brings the total to four TPU systems released within one year.
Nvidia CEO Jensen Huang admitted he had an inappropriate relationship with a woman suspected of being a Chinese intelligence informant. The board has launched an independent investigation. The scandal is likely to weaken Huang's lobbying to ease AI chip export restrictions.
Nvidia announced plans to invest in Korean startups at an AI startup forum it held with the Ministry of Science and ICT. More than 20 Korean startups took part in the forum. Nvidia has already invested in Point2 and said it plans to invest further in Motif Technologies, RLWRLD, and Upstage.
Nvidia is helping AI move into machines and processes in the real world, building on Korea's strengths in manufacturing and engineering. Samsung Electronics, Kakao Mobility, and LG Electronics are each expanding their collaboration in their own fields.
As AI agents (artificial intelligence programs that handle tasks on their own) become more common, technology to manage them safely is becoming important. Nvidia's blog introduced how Korean companies are releasing security tools and AI models to keep up with this trend.
Nvidia is expanding its cooperation with South Korea's Ministry of Science and ICT. The work covers many areas, from chips and data centers to talent development, startups, AI models, and security.
NVIDIA has released deployment settings that let four open AI models made by Korean companies run quickly across multiple GPUs (graphics processing units). The settings are optimized for NVIDIA's open-source tool Dynamo.
The daily users of DeepSeek, a Chinese AI company, have passed 1.54 million. On the same day, it was reported that Nvidia's chief executive officer (CEO) is under investigation. You can check the full details in the original article.
Germany's Siemens, which makes industrial products and automation equipment, has released a video showing how it uses NVIDIA AI tools to speed up production planning and quality control. It covers using AI agents, programs that divide up and handle tasks on their own, together with digital twins, which are virtual copies of real equipment, to help run factories.
NVIDIA has explained how to run the FLUX 3 Image Klein image generation model faster on Blackwell GPUs. By combining several optimization techniques, it greatly reduced the time needed to create images.
AWS now lets a single GPU be split into several pieces on SageMaker real-time endpoints, so multiple models can share it. This uses resources more efficiently than giving an entire GPU to one model. However, it does not benefit every workload.
In reinforcement learning post-training, the latest model weights must be continuously moved from training GPUs to inference GPUs. This transfer is called refit, and as models have grown, the amount of data that must be moved at once has increased sharply. NVIDIA introduced Reshard, a new communication method that moves only the necessary data.
NVIDIA's released wireless software and open-source tools are being used in 6G network research. Researchers are testing ways to help robots react faster and receive signals better.
AI agents handle tasks like searching, querying databases, and writing and running code, and each of these steps needs a CPU (the central processing unit of a computer). NVIDIA says its new Vera CPU delivers both high speed and high throughput per unit of power in a single chip.
NVIDIA has released four AI models that make virtual worlds nearly real-time. It has opened them so anyone can download the model weights (the values an AI gets through training). They are mainly tools used by robotics and self-driving researchers.
NVIDIA has introduced DSX OS, a set of open-source software for setting up, verifying, and running Kubernetes clusters (tools that group multiple computers together and manage programs across them) that use GPUs. Even after a cluster is first certified, its condition keeps changing as hardware is replaced or drivers are upgraded, so these tools help run repeated checks at several stages.
Chris Alexiuk, an NVIDIA engineer, works to spread awareness of open-source AI that anyone can use. He teaches AI development on YouTube and shares his knowledge with many people.
In data centers that run AI services, how usefully companies use high-bandwidth memory (HBM, fast memory attached next to the GPU) has become important, because it is fast but expensive and scarce. NVIDIA says it is solving this problem by designing chips, software, CPUs, and storage together.