Apple’s New CEO Ternus Takes the Helm September 1, AI Becomes Top Priority
On August 30, tech reporter Mark Gurman reported that Apple’s hardware engineering chief, John Ternus, will officially step into the CEO role on September 1, with artificial intelligence taking center stage during his tenure. With 25 years of hardware experience under his belt—having overseen products like the Mac, iPad, AirPods, and iPhone—Ternus now faces a host of challenges, including Apple’s lagging AI efforts, slowing services growth, and rising supply chain costs. According to Gurman, Apple plans to roll out a revamped Siri AI across the board in September, while continuing to push new AI-powered devices, such as smart home gadgets that recognize users and serve up personalized content, camera-equipped AirPods with environmental awareness, and smart glasses. Ternus is also championing the use of AI to improve product development, hardware testing, and device reliability. Meanwhile, Apple’s first foldable iPhone is expected to make its debut at the September 9 launch event—marking Ternus’s first major product showcase as CEO.
Zhipu: H1 Total Revenue Around 954 Million Yuan, Up Nearly 400% YoY
Zhipu announced on August 31 that its first-half total revenue came in at roughly 954 million yuan (about $142 million), a year-over-year surge of nearly 400%. Of that, revenue from its open platform and API business reached approximately 825 million yuan (around $123 million), skyrocketing 2,735.7% compared to the same period last year.
OpenCSG Closes Hundreds of Millions in Pre-A Funding
Recently, OpenCSG announced it has wrapped up a Pre-A funding round worth hundreds of millions of yuan, pushing its post-investment valuation into unicorn territory. The round saw participation from Jing’an Capital, Shibei High-Tech, Xuhui Capital, Huanghai Financial Holdings, and others. With this fresh capital, the company plans to double down on open-source AI infrastructure, its Agentic AI platform, global developer ecosystems, and city-level AI infrastructure, helping to take China’s open-source AI tech and ecosystem to the world stage.
SK Hynix Considers Japan Joint Venture Plant to Meet AI Storage Demand
News on August 31: South Korea’s SK Hynix is studying the feasibility of setting up a joint-venture memory chip plant in Japan, aiming to keep pace with surging storage demand from the AI boom while also reining in production costs. SK Group Chairman Chey Tae-won said the company is scouting potential locations across Japan, with “ample electricity and water resources” as the top criteria—though he didn’t reveal any prospective partners or specific sites just yet.
Tencent Hunyuan: Hy4 Preview Usage Spikes, WorkBuddy Gets Emergency Expansion
On August 31, Tencent Hunyuan posted an update: Since the Hy4 preview was first integrated into WorkBuddy on August 28, its significantly upgraded Agent capabilities have drawn enthusiastic feedback and heavy usage from users and developers alike. On launch day, the model was already seeing queue times in WorkBuddy’s task pipeline, and we sincerely apologize for the inconvenience.To ensure a smooth experience, we’ve urgently expanded the Hy4 preview inference cluster and will keep dynamically reallocating resources. That said, given the limits of high-end compute capacity and peak-time concurrency, some queues may still pop up during certain windows.
OpenAI Reportedly Snaps Up Thousands of Apple Macs for AI Training
On August 31, reports emerged that OpenAI has purchased thousands of Apple Mac mini and Mac Studio units, specifically for reinforcement learning training and developing AI agents that can operate computers. Anthropic, meanwhile, has been renting Mac compute at scale through Amazon Web Services. Notably, OpenAI’s purchases are all Mac mini and Mac Studio models without displays or keyboards—no MacBooks here—and the company has reportedly pushed Apple to speed up deliveries multiple times due to overwhelming demand. Apple’s M-series chips feature a unified memory architecture, letting the CPU, GPU, and other components share the same memory pool, which cuts down on the performance drag caused by shuffling data around during AI training. Some high-end chips offer up to 512GB of unified memory, and the Mac mini and Mac Studio’s thermal designs are also better suited for marathon training sessions.
