The August 28 archive covers NVIDIA's reported $12.9 billion agreement to acquire Hugging Face, Google expanding AI Mode with travel-agent capabilities, and a 100-plus-company coalition urging action against rogue AI. In China, Zhipu shipped GLM-5.3 and the lightweight GLM-5.3-Flash on domestic compute, Yunzhisheng's H1 results showed agent revenue of 478 million yuan, and RayNeo's iO smart glasses sold strongly from ¥1,996. The throughline: agentic AI is moving from answers to actions and transactions, while open-source ecosystems and domestic compute grow more tightly coupled.
Zhipu AI released GLM-5.3 on Hugging Face on August 28, a bilingual conversational text-generation model built on transformers with safetensors weights.The registry entry carries the glm_moe_dsa tag, indicating a Mixture-of-Experts-style architecture, and references arXiv paper 2602.15763.
Zhipu's lightweight GLM-5.3-Flash model has officially launched, with SenseTime's platform providing domestic compute support.The Flash tier targets high-frequency, low-latency workloads suited to large-scale applications that need fast responses.The notable signal is at the compute layer: a homegrown heterogeneous computing platform is now carrying public-facing service for a leading Chinese foundation model, tying model capability more closely to the domestic supply chain and pointing to more controllable supply and affordable costs for developers.
Google is expanding AI Mode in Search with travel-agent capabilities, letting users track flight prices and get help booking hotels.The update moves AI Mode beyond returning links and answers into actually handling parts of trip planning and booking.It pairs generative search with action-oriented features, putting AI Mode in more direct competition with dedicated travel booking platforms and other AI travel agents.The open question is how far the booking funnel goes and how quickly the features roll out to more users and regions.
SenseTime's computing platform and HiDream.ai completed a full domestic-compute adaptation for video generation, reaching a 93% multi-card parallel speedup for DiT models on domestic chips through step and CFG distillation plus multiple parallelization strategies.The teams also tightened feature consistency across multi-card inference, and built a unified hardware abstraction layer enabling zero-cost migration across more than 10 heterogeneous chip types.HiDream's image models now support large-scale online traffic and short-video creation, validating a reusable path for moving compute-heavy generative AI into domestic-chip production.
The Information reports that NVIDIA has agreed to acquire Hugging Face for $12.9 billion, a deal not yet formally finalized, after Microsoft's own acquisition talks fell through.The platform, known as the GitHub of AI, hosts more than 3 million open models and 1 million datasets; NVIDIA was an investor in its 2023 $235 million Series D.Owning the distribution hub of open-source AI would help NVIDIA keep training and inference workloads on its hardware as closed labs build their own chips, and give it an outlet for unused cloud capacity.
Yunzhisheng, the first AGI-focused company listed in Hong Kong, posted 562 million yuan in first-half revenue, up 38.7%, with its agent business contributing 478 million yuan, or 85.1% of the total.Token revenue reached nearly 30 million yuan with Q2 up over 500% quarter over quarter at a gross margin above 60%, while recurring revenue exceeded 60% of sales and the loss narrowed 20.
RayNeo published first-sale results for its iO smart glasses on August 27, reporting strong early sales for the 34-gram, all-day AI wearable priced from 1,996 yuan, and framing the debut as a signal that AI glasses are moving toward mass adoption.The strong start validates the combination of light weight, always-on AI, and an accessible price point, putting an always-wearable AI entry point squarely in the mainstream price band.Weight, battery life, and price decide how fast the category scales, so rivals are likely to adjust pricing strategies if the band is validated.The key question is whether initial heat converts into sustained shipping volumes.
OpenAI and Thailand's Ministry of Higher Education, Science, Research and Innovation (MHESI) launched an eight-week accelerator for Thai AI startups, welcoming an inaugural cohort of 10 companies in health, wellness, and education with the goal of turning AI prototypes into trusted, production-ready products.For OpenAI, the move deepens its Southeast Asia footprint; for Thailand, it gives local developers a formal channel of support from a leading model maker.The sector focus is notable since these are among the fastest areas of generative AI adoption but also among the most regulation-sensitive.The key things to watch are what the cohort ships and whether the model expands to other markets.
TIME magazine's global AI 100 list has put a spotlight on the reclusive leader of Zhiyuan Robotics, the humanoid robotics company founded by Peng Zhihui ("Zhi Hui Jun").QbitAI reports that the company's behind-the-scenes helmsman, who has long avoided the public eye, made the list — a sign that the company's weight is extending from its products to the global AI conversation.Zhiyuan sits at the center of an increasingly heated humanoid robot race, and in a sector where founder personal brands shape narratives, the public image of a core leader is itself becoming part of the brand.
Barret Zoph, a Thinking Machines co-founder who briefly returned to OpenAI earlier this year, has joined Google as vice president of research.A Google spokesperson told the Wall Street Journal the company looks forward to Zoph bringing his reinforcement learning and post-training expertise to Gemini.Zoph spent only about five months at OpenAI, where he headed AI enterprise sales, after leaving Thinking Machines in January with co-founder Luke Metz.The move comes amid heavy turnover at OpenAI, which has lost critical staff from its COO to a top data center executive as it prepares for an IPO.
Sandhya Devanathan, a Meta executive, is leaving the social media giant to join OpenAI, where she will oversee some of its operations across Southeast Asia and Australia, TechCrunch reported.The move comes as Meta faces growing scrutiny in India, adding a talent dimension to a period of heightened regulatory attention.For OpenAI, the hire brings a seasoned operator to its fast-expanding international business in two of the fastest-growing regions for AI adoption.The move continues a broader pattern of senior talent moving from big tech to frontier AI companies.
OpenAI, Anthropic, Google, and more than 100 other tech companies jointly called for action to defend against rogue AI, criticizing the current state of cybersecurity and promoting a new solution they say can counter a new generation of cyber threats.The report does not disclose technical details, but frames the threat landscape as broadening beyond traditional hackers to compromised or maliciously used AI systems.The rare united front signals AI safety is moving from lab discussions to collective industry action, and could drive new security standards or regulatory moves.
