Hugging Face Hosts Alibaba AI, NVIDIA Rubin GPU

The artificial intelligence landscape continues to evolve rapidly, with significant developments in model efficiency, hardware, and application across various sectors. Alibaba has introduced Qwen3-Next, an 80-billion-parameter AI model that is ten times cheaper to train and significantly faster than its predecessor, utilizing innovations like a hybrid attention mechanism and a sparse Mixture of Experts (MoE) architecture. This model is now open-source on Hugging Face, aiming to democratize access to advanced AI. In hardware, NVIDIA unveiled its Rubin CPX GPU for large context inference, while South Korean startup FuriosaAI demonstrated its RNGD chip powering OpenAI's GPT-OSS 120B model, highlighting efforts to reduce AI power costs and improve efficiency. The demand for semiconductors is surging, with investments expected to exceed $1.5 trillion by 2030, driven by AI and electric vehicles. AI's growing capabilities are also evident in its improved memory functions, with chatbots like ChatGPT, Claude, and Gemini becoming better at recalling user details and preferences, a feature OpenAI CEO Sam Altman sees as crucial for personalized AI assistants. Beyond core AI development, applications are expanding. The HR Tech 2025 conference will feature an AI Playground and numerous new product launches focused on AI integration in human resources. In drug discovery, despite technological progress, the industry faces funding challenges due to macroeconomic and regulatory factors. Meanwhile, authorities are warning against the misuse of AI, specifically the sharing of fake AI-generated images of a shooting suspect that can misdirect investigations. Efforts to power AI sustainably are also emerging, with plans to use existing infrastructure for AI data centers, though environmental concerns remain. Companies like Forvis Mazars are offering services to help organizations strategize and integrate AI responsibly, while AI recorders like the Plaud Note Pro are enhancing note-taking with AI-powered summarization and highlighting features.

Key Takeaways

• Alibaba's new Qwen3-Next AI model, with 80 billion parameters, is ten times cheaper to train and faster than previous versions, and is available open-source on Hugging Face.
• NVIDIA has unveiled its Rubin CPX GPU designed for large context inference, while FuriosaAI's RNGD chip demonstrated efficient powering of OpenAI's GPT-OSS 120B model.
• Global semiconductor investment is projected to surpass $1.5 trillion between 2024 and 2030, fueled by demand from AI and electric vehicles.
• AI chatbots like ChatGPT, Claude, and Gemini are enhancing their ability to remember user details and past interactions across conversations.
• The HR Tech 2025 conference will showcase over 60 new AI-driven products and services for human resources.
• Authorities are cautioning against sharing fake AI-generated images of suspects, as they can hinder investigations.
• AI drug discovery, a $18 billion industry, is experiencing a slowdown in venture capital funding despite technological advancements.
• A company plans to repurpose a former coal-fired power plant on Cayuga Lake to power a new AI data center, sparking environmental debates.
• The Plaud Note Pro AI recorder features a 'Press to Highlight' button and uses AI models like GPT, Claude, and Gemini for summaries.
• Forvis Mazars offers services to guide organizations in developing AI strategies and ensuring responsible, ethical integration.

Alibaba's Qwen3-Next AI model offers efficiency and power

Alibaba has launched Qwen3-Next, a new AI model architecture designed for better understanding of long texts, large scale, and high efficiency. It uses innovations like a hybrid attention mechanism and a sparse Mixture of Expert (MoE) architecture. The Qwen3-Next-80B-A3B-Base model, with 80 billion parameters, activates only 3 billion during use. It is now available open source on Hugging Face and Alibaba Cloud. This new model performs better than the previous Qwen3-32B model while using less training cost and offering much higher speed for long texts.

Alibaba's new Qwen3 model boosts AI power and cuts costs

Alibaba has released its latest AI model, Qwen3-Next, built on a new architecture that makes it 10 times more powerful and 10 times cheaper to build than its predecessor. This 80-billion-parameter model uses advanced techniques like hybrid attention and a sparse Mixture of Experts (MoE) architecture to achieve high performance with less computational cost. The model is now open-sourced on Hugging Face, allowing developers worldwide to use and modify it. Alibaba aims to narrow the gap with US AI firms through this open-source approach, making advanced AI more accessible.

Alibaba's Qwen3-Next AI model slashes training costs

Alibaba Cloud has released its new open-source AI model, Qwen3-Next-80B-A3B, which costs ten times less to train and is ten times faster than previous models. This 80-billion-parameter model achieves performance comparable to Alibaba's larger flagship model but uses significantly fewer resources. The company has made the model open-source on platforms like GitHub and Hugging Face, encouraging global developer innovation. This release highlights a shift in AI development towards efficiency and innovation over sheer size, making advanced AI more accessible on consumer-grade hardware.

AI news roundup: Algolia, C3 AI, Expert.ai, NVIDIA and more

This week's AI news includes updates from several companies. Algolia launched an Intelligent Data Kit for easier data preparation. C3 AI introduced agentic process automation for enterprise AI. Sphinx raised $9.5 million for its Copilot platform. CoreWeave Ventures will invest in AI infrastructure startups. CTERA and Digital Realty are hosting events and opening labs focused on AI and cloud. Expert.ai enhanced its GenAI solutions, while Gigamon and Hitachi Digital Services launched AI-powered security and agent systems. Honeycomb and Kestra released new observability and orchestration platforms. NVIDIA unveiled its Rubin CPX GPU for large context inference, and Perforce added AI features for software development. Progress Software launched a SaaS RAG platform, and Ontra improved its legal document analysis with AI.

HR Tech 2025 to feature AI playground and new product launches

The HR Tech 2025 conference in Las Vegas will focus on the growing impact of AI and automation in human resources. Attendees can explore a new AI Playground for hands-on experience with various solutions. Over 60 new products and services will launch, covering areas like AI-driven recruitment and employee experience platforms. The event will also feature discussions on wellbeing and culture, and networking opportunities for senior leaders. This conference highlights the shift from simply finding new tools to effectively integrating them into HR strategies.

Police warn against sharing fake AI photos of shooting suspect

Authorities are warning the public and law enforcement not to share fake, AI-generated photos of a suspect in the shooting death of activist Charlie Kirk. Some people have used AI to 'enhance' blurry photos, creating images with fabricated details like distorted flags or extra accessories. These fake images have even fooled some law enforcement agencies, potentially hindering investigations. Experts emphasize that AI does not simply clean up photos but creates new ones, leading to misidentification and wasted resources. Sharing such images can misdirect investigations and cause confusion.

AI data center planned for Cayuga Lake power plant

A company called TeraWulf plans to use the infrastructure of an old coal-fired power plant on Cayuga Lake to power a new data center for artificial intelligence. TeraWulf claims this project will be a greener way to provide energy for AI. However, local officials and environmental advocates have expressed doubts about the environmental impact. The growing demand for power from AI is leading to proposals like this, sparking debate about sustainability in the tech industry.

AI chatbots are getting better at remembering user details

Popular AI chatbots like ChatGPT, Claude, and Gemini are rapidly improving their ability to remember information about users across conversations. Initially, chatbots had limited memory within a single chat session. Now, features allow them to store specific user preferences and recall past interactions, even in new chats. OpenAI CEO Sam Altman sees this as a key feature for creating personalized AI assistants. This development raises questions about how much personalization users want and the long-term implications of AI knowing so much about individuals.

AI drug discovery faces funding challenges despite tech progress

The AI drug development industry, now worth $18 billion, is facing a slowdown in venture capital funding. While AI has shown potential in drug discovery, with many compounds entering clinical trials, investor confidence has decreased due to macroeconomic and regulatory issues. Companies like Recursion and BenevolentAI have experienced difficulties, with some shelving drugs or delisting. Experts suggest that the challenges are more related to funding environments and biological complexities than the AI technology itself. The next year will be crucial for AI drug development companies to prove their success with emerging data.

Semiconductor investment surges for AI and electric vehicles

Global investment in semiconductors is expected to exceed $1.5 trillion from 2024 to 2030, driven by the high demand from AI applications and the growth of electric vehicles (EVs). Government initiatives like the US CHIPS Act are fueling significant investments in new fabrication facilities. AI workloads are reshaping the server market, with AI accelerators becoming a major driver of demand. In the automotive sector, EVs and autonomous driving features are increasing the need for advanced semiconductors, with silicon-carbide and gallium-nitride materials becoming more prevalent.

Plaud Note Pro AI recorder adds 'Press to Highlight' button

Plaud has launched the Note Pro, a credit card-sized AI audio recorder that aims to improve note-taking accuracy. A new 'Press to Highlight' button allows users to mark important moments during conversations, which the AI then prioritizes for summaries and action items. The device features four microphones, a week-long battery life, and a small AMOLED display. Its updated app will offer multimodal capabilities, combining audio with text and images, and will dynamically use different AI models like GPT, Claude, and Gemini for summaries. The Note Pro is priced at $179 and is available for pre-order.

FuriosaAI chip powers OpenAI model, tackling AI power costs

At the OpenAI Korea launch, South Korean startup FuriosaAI demonstrated its second-generation RNGD chip running OpenAI's open-source GPT-OSS 120B model. This showcase highlighted the chip's ability to efficiently power a large language model with minimal hardware, addressing the significant power and cost bottlenecks in AI. FuriosaAI's technology aims to make AI more sustainable and accessible by optimizing power-to-performance ratios. This demonstration positions Korea as a key player in the global AI semiconductor market, focusing on efficiency and cost-effectiveness.

Forvis Mazars offers AI strategy and integration services

Forvis Mazars provides services to help organizations implement Artificial Intelligence effectively and responsibly. They focus on aligning AI initiatives with business goals, identifying practical use cases, and guiding build vs. buy decisions. The company offers support across the entire AI transformation journey, from strategy development and readiness assessments to deployment, change management, and risk governance. Their services include data preparation, AI model development, and establishing AI Centers of Excellence to ensure safe and ethical AI integration at scale.

Sources

• Qwen3-Next: A New Generation of Ultra-Efficient Model Architecture Unveiled
• Alibaba releases next-generation Qwen3 model as AI rivalry intensifies
• Alibaba’s Qwen3-Next Slashes AI Training Costs Tenfold
• Artificial Intelligence News for the Week of September 12; Updates from Denodo, Progress, Insight Jam & More
• What's on tap at HR Tech? AI playground, fresh research, new products and more
• Stop sharing fake, AI-generated photos of Charlie Kirk shooting suspect
• Backstory: An AI data center on Cayuga Lake
• How Much Do We Want AI to Remember About Us?
• The AI drug breakthrough is taking a long time to arrive for reasons that may have little to do with the technology’s limits
• PwC: How is Investment in Semiconductors Driving AI & EVs?
• Plaud’s Note Pro Reinvents AI Note-Taking With One Button
• FuriosaAI Showcases RNGD at OpenAI Korea Launch — A Step Toward Solving AI’s Power and Cost Bottleneck - KoreaTechDesk | Korean Startup and Technology News
• AI Strategy & Integration