NVIDIA to Invest $150 Billion Per Year in Taiwan—Behind the Push to Make America the AI Hub
It has been reported that NVIDIA plans to invest roughly $150 billion per year (about 22 trillion yen) in building out AI infrastructure in Taiwan. According to multiple media reports, NVIDIA CEO Jensen Huang positions Taiwan as "the center of the AI revolution," and the company intends to dramatically expand its investment in the country, a linchpin of the semiconductor supply chain.
The move stands in curious contrast to the Trump administration's push to make the United States an AI hub. According to Ars Technica, while policy is designed to encourage investment within the US, it highlights the reality that the AI industry simply cannot function without Taiwan's manufacturing infrastructure.
The concentration of AI semiconductor supply in Taiwan carries inherent geopolitical risks, and it is drawing attention in the context of US-China tensions. NVIDIA's massive investment suggests that reliance on Taiwan's manufacturing prowess—led by TSMC—will only deepen going forward.
Apple to Fully Overhaul Siri in iOS 27—Screenshots of New AI Features Leak
Bloomberg reporting has revealed that Apple plans a complete overhaul of Siri in its next OS, "iOS 27." The leaked screenshots show a redesigned Siri interface along with a number of new AI capabilities, including an AI-powered camera app called "Pro Camera."
Apple has rolled out AI features incrementally through WWDC announcements, but this Siri refresh is seen as signaling a shift from the traditional voice assistant toward a more autonomous AI agent. AI integration into the Photos and Camera apps also appears to be advancing, with the potential to change the very core of the user experience.
The story is also gaining traction on Hacker News, a sign of how closely watchers are following whether Apple's AI strategy is finally getting serious.
Tesla's AI Trainers Don't Trust Their Own Autopilot Technology—Reuters Investigation
A Reuters investigative report has revealed that employees working on AI training inside Tesla do not trust the company's autonomous driving technology or its safety statistics.
According to the report, the AI trainers responsible for data labeling and model evaluation have concerns about how safety metrics are calculated, and they point out that the published safety data may not accurately reflect reality. Because the safety of autonomous driving technology directly affects consumers' lives, voices of distrust from within the company carry serious weight.
The story has drawn significant attention—including 8 points on Hacker News—and could influence transparency and safety regulation across the autonomous driving industry as a whole.
SpaceX Is Building Its Own AI Training Stack in C
Elon Musk revealed on X (formerly Twitter) that SpaceX has nearly finished v1.0 of its own AI training stack, written in C.
The decision to build from scratch in C, rather than relying on existing Python-based ML frameworks such as PyTorch, appears aimed at optimizing performance and hardware control. SpaceX's rocket control systems demand extremely rigorous real-time performance, so eliminating the overhead of general-purpose frameworks is a sensible call.
As AI adoption accelerates in the space industry, this approach of building from the infrastructure layer up could influence other hardware-intensive industries as well.
First Evaluation of Utah's AI Prescription Pilot—A Regulatory Experiment in Medical AI
The first evaluation results of Utah's pilot program for AI-driven prescription refills have been released as a PDF. The pilot, which uses an AI system called Doctronic, has drawn attention as a pioneering case of regulators formally allowing AI to take on part of the practice of medicine.
AI-driven medical judgment is a difficult area when it comes to assigning accountability and guaranteeing safety, and regulators have proceeded cautiously. Utah's initiative amounts to a state-level regulatory sandbox experiment, and it may offer lessons for medical AI policy in other states and countries.
MONET—An Open Image and Text Dataset with Over 100 Million Entries
Jasper AI has released "MONET," a dataset containing 104.9 million high-quality images with text captions, under the Apache 2.0 license. Curated from a pool of 2.9 billion images, the dataset can be used to train text-to-image (T2I) generation models.
Alongside the dataset, Jasper AI has also open-sourced a UMAP-based tool for visualizing data distributions, a text and image search tool, and a T2I training codebase built on MONET—a major contribution to the open AI research community. Releasing a large-scale, high-quality image dataset in the open will further democratize image generation AI.
---