Since hostilities resumed, vessel transits through the Strait of Hormuz have plummeted, pushing transportation costs higher.

According to Bitget market data, Hong Kong-listed "the two leading large model players" advanced, with Zhipu AI surging more than 8% at one point and MINIMAX-W rising over 5%.

According to on-chain analyst Ai Yi (@ai_9684xtpa), whale address 0x604…0b21d previously shorted BTC worth $45.17 million between August 24 and 25, exiting the trade with a total loss of $831,000. Early this morning, the address switched to a long position, opening a 12x leveraged long order for 554.71 BTC, with the position valued at roughly $43.72 million — ranking as the 8th-largest BTC holding on Hyperliquid. Its average entry price stands at $80,140.6, and it is currently facing an unrealized loss of approximately $748,000.

NVIDIA unveils NVHBM technology for next-generation AI infrastructure. NVHBM is a custom HBM base die co-designed and validated with leading memory chip suppliers, and will be offered via NVLink Fusion to hyperscale cloud service providers and AI-native enterprises developing custom AI accelerators (XPUs). Compared to standard HBM4e, each NVHBM stack delivers up to 30% additional memory bandwidth, cuts HBM power consumption by up to 15%, and frees as much as 25% of the XPU’s compute die area. NVIDIA says these improvements will help custom accelerators support larger AI models, read KV caches faster, and enhance training and large-scale inference efficiency. NVHBM achieves this by redesigning the physical memory interface, moving the memory controller into the 3D HBM stack and integrating a custom PHY, which reduces the area of the PHY and its supporting regions by up to 67% while simplifying interposer routing. The freed chip area can be used to add matrix engines, caches, networking, and other workload-specific features. At the rack level, NVLink Fusion chiplets connect custom XPUs to the NVLink network and link to CPUs via NVLink-C2C. NVIDIA adds that when combined with NVLink Fusion, NVHBM boosts each XPU’s end-to-end performance by around 30%, while enabling custom XPUs to form heterogeneous computing systems with NVIDIA GPUs. The company has not disclosed specific commercialization timelines in the announcement.

According to people familiar with the matter, Japanese memory chip maker Kioxia is building a new manufacturing plant at its production base in Iwate Prefecture, northern Japan, to ramp up production capacity and meet surging demand for storage devices in the artificial intelligence (AI) sector. The project is expected to exceed 1 trillion yen (about $6.27 billion). The new plant, the third facility at the base, will produce Kioxia’s latest high-density 3D flash memory chips, designed to handle the massive workloads generated by AI services. Kioxia started shipping its stacked NAND chips, the 10th-generation BiCS flash memory, last month. Against the backdrop of Japan’s ongoing financial support for firms including TSMC, Sony Group and Micron Technology, reports say Kioxia and SanDisk will apply for subsidies from the Japanese government. A source familiar with the matter noted Kioxia plans to announce the expansion plan on Thursday evening. (Source: Jinshi)

Video conferencing giant Zoom projects adjusted earnings per share (EPS) of $1.46 to $1.48 for its third quarter, below the average analyst estimate of $1.50 compiled by LSEG, indicating that while the company continues to add AI features, competition in the video conferencing market is pressuring its profitability. Following the earnings release, Zoom’s shares fell 3.8% in after-hours trading. Zoom forecasts third-quarter revenue of $1.275 billion to $1.28 billion, matching the average analyst estimate of $1.28 billion. The company is currently rolling out AI Companion assistants, meeting reception tools, and its Zoom AI Services suite for enterprise clients to attract more users, though it still faces fierce competition from Microsoft Teams and Alphabet’s Google Meet. Zoom’s stake in Anthropic has risen in value to $3.13 billion. According to the company’s disclosures, this investment generated $1.61 billion in paper gains during the three months ending in July. Zoom first invested in Anthropic in early 2023, when the two parties established a partnership to integrate Anthropic’s Claude model into Zoom’s products. Additionally, Zoom revealed that it made an additional $301 million investment in Anthropic over the six months ending in July this year.

Amazon (AMZN) and Nvidia (NVDA) announced Wednesday local time that Amazon will deploy approximately 2 million of Nvidia’s high-end chips in 2027 and 2028. The chips are designed for training and running artificial intelligence (AI) models. Separately, Amazon stated in March this year that it would begin installing 1 million Nvidia chips in Amazon Web Services (AWS) data centers starting this year. The latest deal covers Nvidia’s Blackwell Ultra, Rubin, and Rubin Ultra graphics processing units (GPUs). Each of these products is priced at at least tens of thousands of dollars, though large clients are eligible for discounts.