Stories about NVIDIA

Z.ai has released the open-source model GLM-5.3-Flash, which has 32 billion parameters and scores only three points lower than the larger GLM-5.3 on the Artificial Analysis Intelligence Index, but at one-seventh the cost. Notably, all inference traffic runs on Chinese AI chips rather than Nvidia hardware. The release of GLM-5.3-Flash marks a new breakthrough in the balance between cost and performance of AI models.

Nvidia is acquiring open-source AI platform Hugging Face for $12.9 billion, with the latter generating $150 million in annual revenue. The deal marks the latest move by Nvidia to invest billions of dollars in open-source AI models, while closed-source AI providers such as OpenAI and Anthropic are moving away from Nvidia hardware. The acquisition of Hugging Face is another move by Nvidia to expand its presence in the AI space.

Nvidia is acquiring open-source AI startup Hugging Face for $13 billion, a move that will help Nvidia shield its chip empire and get back into the cloud computing business. Hugging Face has a vast array of models and community resources that Nvidia can leverage to further its AI business.

In response to surging demand for data centers, Amazon has placed an additional order for 2 million graphics processing unit chips from Nvidia, a threefold increase from the previous order. This partnership will not only be limited to chip procurement but will also extend to other areas.

Training navigation strategies for robots is a crucial step towards achieving robot autonomy. Unlike simple motion control, navigation requires a robot to continuously locate itself, interpret changes in the surrounding environment, choose paths, and avoid obstacles. However, this process often requires a large amount of data, simulation resources, and robot interfaces. Recently, researchers have proposed a method for training cross-body robot navigation strategies using AI agents, aiming to improve the universality and adaptability of robot navigation strategies.

Alibaba has released the Qwen3.8-Flash-Next model weights as a preview of the upcoming Qwen4 architecture for developers to experiment with and evaluate. The model uses a multi-modal mixture of experts (MoE) architecture, with a total of 176 billion parameters, including 51 billion N-gram embedding parameters, and 6 billion parameters activated per token. Its native context window size is 262,144 tokens, and can be extended to 1 million tokens with YaRN.

Alibaba has released the model weights for Qwen3.8-Flash-Next as a preview of the upcoming Qwen4 architecture, allowing developers to experiment and evaluate it. The model adopts a multi-modal MoE architecture, with 12.5 billion main model parameters, supplemented by 5.1 billion N-gram embeddings, and 600 million activation parameters per token. It supports a context window of up to 262,144 tokens, which can be extended to a maximum of 1 million tokens.

OpenAI unveiled its first deep learning inference chip, Jalapeño, and benchmarked it at the Hot Chips conference. The results showed that Jalapeño outperformed Nvidia's Blackwell and Rubin in terms of throughput and energy efficiency. SemiAnalysis CEO Dylan Patel said it was a dark horse that beat Nvidia on its first chip.

Nvidia announced it has begun mass-producing its inference chip, the Groq 3 LPX, and achieved 3,400 tokens per second on the Gemma 4 31B, which is claimed to be four times faster than Cerebras. However, this number does not reveal the whole truth. Nvidia requires at least 64 accelerators to achieve this speed, while Cerebras only needs one or two accelerators.

An Nvidia senior executive is suspected of being involved in a plan to smuggle AI servers to China by Supermicro, a move that Jensen Huang publicly criticized.

Nvidia is in talks to invest in Perplexity at a valuation of over $30 billion, as the latter's annualized revenue has reached $750 million.

NVIDIA's AVO reaches 100% on ARC-AGI-3, demonstrating a frontier-level general-purpose architecture for long-horizon autonomous agents. This marks a significant breakthrough for NVIDIA in the field of autonomous agents. This means NVIDIA's competitiveness in the AI field has been further enhanced. Developers can leverage the AVO architecture to develop more intelligent autonomous agents.

NVIDIA's blog post discusses the role of security in an AI agent stack, highlighting the importance of integrating security into AI systems. The article notes that security is not just about adding a separate component, but rather requires a holistic approach to securing the entire system.