CONNECT WITH US
Sep 29, 17:29
Anthropic's US$2 trillion IPO bet comes with 80 pages of warnings
Confidential details of Anthropic's initial public offering (IPO) prospectus, reviewed by Reuters, reveal an enterprise operating at an unprecedented scale and making a historic wager that artificial intelligence will transform the global economy far more profoundly than industrialization, electricity, or the internet. To capture that future, the startup is targeting a public valuation of more than US$2 trillion.
Amazon will take an equity stake in Gold Circuit Electronics (GCE), a Taiwanese maker of printed circuit boards (PCBs) for AI servers and network switches. GCE's board approved the pricing of a NT$1.589 billion private share placement on September 29, with Amazon as the buyer. According to industry observers, it is rare for a major cloud provider to invest directly in a Taiwanese PCB maker. The deal turns a customer-supplier relationship into one where the customer also owns part of the company.
RoboTechnik completed a Hong Kong listing on September 29 and began trading on the Hong Kong stock exchange, according to its German photonics automation subsidiary ficonTEC. The new capital will be used for further investment in ficonTEC, which is expanding its role in silicon photonics and optical interconnect manufacturing for AI and high-performance computing infrastructure.
Cerebras Systems is expanding its push into AI inference infrastructure through a collaboration with Gimlet Labs, as demand grows for computing systems capable of delivering faster responses from increasingly complex AI models.

GlobalFoundries (GF) is preparing a major global capacity expansion as TSMC is set to gradually exit gallium nitride (GaN) foundry services in 2027. GF said it was the first foundry to proactively approach TSMC and secure a GaN technology license, positioning itself to give existing customers a "seamless migration" path to its own platform.

MIPS, a GlobalFoundries (GF) unit, said it has completed its integration with the ARC unit acquired from Synopsys earlier this year and is using the move to position RISC-V for physical AI, while expanding from AIoT into automotive and data center applications. MIPS said the strategy lowers the barrier for global customers to adopt GF platforms across process technologies and has pushed GF and MIPS to become the world's second-largest processor IP provider.
Qualcomm Technologies held its Snapdragon Summit 2026 in Maui, Hawaii, from September 22 to 24. During the keynote on the second day, Qualcomm president and CEO Cristiano Amon spoke with co-founder and CEO of Liquid AI Ramin Hasani.

Chinese President Xi Jinping and his wife, Peng Liyuan, arrived at the White House on the evening of September 24 US time to attend a state dinner hosted by US President Donald Trump and First Lady Melania Trump.

Nvidia expanded its open-source CUDA-Q platform with CUDA-Q Logical, an orchestration layer aimed at helping researchers build software for fault-tolerant quantum computers. The company said the platform already integrates with about 75% of publicly available quantum processing units (QPUs), giving developers a common environment across multiple hardware types.
LG AI Research is moving into Kazakhstan and the Middle East to capture demand for sovereign AI in emerging markets. The South Korea-based unit is using its model development and commercialization experience to pursue overseas deals as US and Chinese firms continue to dominate the global AI market.
Meta has introduced a camera-free smart glasses model while Samsung Electronics prepares to enter the AI glasses market in 2026, as privacy concerns and product design trade-offs reshape the category. The shift comes amid growing scrutiny of unauthorized filming, changing regulatory expectations, and rapid expansion in China.

AI agents are moving into enterprise workflows and driving a sharp rise in demand for AI infrastructure, with a new report from IDC and IEIT Systems warning that the global compute supply-demand gap will continue widening through 2030. The report, released at the AICC2026 Artificial Intelligence Computing Conference, said high-bandwidth memory (HBM) was the biggest bottleneck.