Alibaba Unveils Zhenwu V900 AI Chip, Claiming Triple Predecessor's Speed

Alibaba has unveiled its Zhenwu V900 AI accelerator, described as China's most powerful AI chip, alongside a commitment of more than US$53.00 billion over three years to expand data center capacity for Alibaba Cloud by 2032. The push into homegrown AI infrastructure positions Alibaba as a key domestic alternative to foreign chip suppliers and could materially influence how it competes in China's rapidly evolving cloud and AI market. The spending carries rising financial stakes: Alibaba's Cloud and AI segment grew revenue 45% year over year, yet EBITDA margins compressed and free cash flow swung to a roughly RMB 44.7 billion outflow. Alibaba's narrative projects CN¥1487.6 billion revenue and CN¥176.3 billion earnings by 2029, requiring 12.5% yearly revenue growth and about a CN¥103 billion earnings increase from CN¥73.3 billion today. The most optimistic analysts already expected Alibaba to reach about CN¥1,687.4 billion in revenue and CN¥362.0 billion in earnings by 2029, so the new AI capex push could either support that upbeat view or expose how aggressive those assumptions were.

Google met South Korean officials on September 9 to discuss supplying its tensor processing units for the country's sovereign-AI buildout, according to a September 22 report from Seoul Economic Daily, putting Alphabet into a market where governments have leaned heavily on NVIDIA GPUs. Alphabet has spent years using TPUs to lower the cost of running its own models and Google Cloud workloads, and supplying them directly to universities, research organizations or on-premise sovereign installations would widen the addressable market and weaken one of Nvidia's strongest structural advantages outside hyperscaler clouds. Google reportedly argues TPUs can lower AI-compute costs materially, though those claims depend on workload, software support and utilization, and Nvidia's CUDA, networking and systems software remain a deeply entrenched platform that governments may still prefer. A Korean TPU deployment would not mean Nvidia has been displaced, but it could establish a template for splitting national clusters across competing architectures. Insider Monkey's database showed 275 hedge funds holding Alphabet at the end of Q2 2026, up from 265 in Q1, with Berkshire Hathaway increasing its Class A Alphabet position about 45% during the quarter, while Nvidia's holder count rose to 285 from 275 and Fisher Asset Management increased its stake about 3%.

AT&S announced an expanded long-term collaboration with Marvell Technology to increase advanced IC substrate capacity for next-generation AI and cloud infrastructure. Marvell Technology is the additional customer identified in AT&S's previously announced expansion of its Kulim manufacturing site in Malaysia, which includes the fit-out of Plant 2 and the construction of a new manufacturing facility for IC substrate cores and advanced packaging. The agreement builds on the companies' existing relationship and is supported by long-term customer commitments that AT&S said provide visibility for future growth while supporting a disciplined investment strategy. AT&S CEO Michael Mertin called the deal an important milestone and said the two companies are well positioned to capture opportunities created by the global build-out of AI infrastructure, while Marvell Senior Vice President and Chief Supply Chain Officer Vinay Krishna said the expanded collaboration strengthens Marvell's supply chain foundation as its data center business expands. AT&S, which employs around 15,000 people and operates production sites in Austria, China, Malaysia and India, said the expansion responds to industry shifts including the move from monolithic chips to chiplet-based designs and rising demand for larger substrate formats, higher integration density and increased layer counts.