Topic Archive

Agentic AI refers to AI systems that can understand objectives, plan a sequence of steps and independently use tools to complete tasks. Recent developments span model delegation to reduce costs, enterprise adoption and multi-window collaboration, as well as cloud infrastructure, identity and access management, payments and crypto asset management. At the same time, coding success rates remain below 50%, while defenses against malicious agents have emerged as another concern. Performance, cost, security and governance will directly determine whether businesses can move AI agents from experiments into reliable day-to-day operations, making this a topic worth watching closely.

AI chips include GPUs, ASICs and HBM used for model training and inference, along with the related advanced manufacturing and packaging technologies. Rising demand has recently prompted TSMC, ASML and SK hynix to expand capacity or raise their outlooks, while Powertech Technology and Broadcom are also investing in advanced packaging. At the same time, higher materials and equipment prices, memory costs, export controls and concerns about excess capacity are emerging. Developments in AI chips affect semiconductor supply chains, cloud infrastructure, international trade and corporate capital spending, making them a key area to watch.

Stablecoins are crypto assets designed to maintain a stable value against fiat currencies such as the U.S. dollar or other assets. Their use is gradually expanding beyond trading into cross-border payments, institutional yield products and payments by AI agents. Final rules under the U.S. GENIUS Act have yet to be implemented, Europe is accelerating a shift from USDT to compliant USDC, and companies including Visa and Stripe are competing to build the next generation of payments infrastructure. Potential bank deposit outflows, anti-money laundering concerns and geopolitical implications are also drawing scrutiny, making regulatory progress, real-world adoption and financial risks key areas to watch.

Large language models (LLMs) are AI models trained on vast datasets to understand and generate text and code. Recent areas of focus include the delay of Gemini 3.5 Pro, competition from Kimi K3 and enterprise use in software development and advertising review. Model jailbreaking, cybersecurity abuse, copyright disputes and content controls imposed by authoritarian governments are also drawing scrutiny. As LLMs become more deeply embedded in software, media and the public information environment, advances in their capabilities, governance, computing costs and social impact warrant close attention.

AI data centers provide the infrastructure for model training, inference and cloud computing, with the industry’s focus expanding beyond chip supply to power, cooling, water use and financing. Taiwan recently amended its regulations to require major energy users to add power generation or storage capacity, while Taipower is considering dedicated direct-supply lines. Overseas operators are stepping up investment in power infrastructure and seeking capital through public listings, even as they face rising server costs and local opposition. Power usage effectiveness, water usage effectiveness, reclaimed water and low-carbon energy solutions will shape construction costs, environmental impact and the pace of computing-capacity expansion.

NVIDIA is a leading global supplier of AI chips and computing platforms, with developments at the company affecting data centers, HBM, power equipment and the physical AI supply chain. CEO Jensen Huang recently visited Japan, working with companies including Toyota and Kawasaki Heavy Industries to advance smart cities, AI-powered shipbuilding and next-generation data centers. Japan is also planning chip purchases to build sovereign AI capabilities. Meanwhile, H200 shipments to China, a Samsung patent investigation and Michael Burry's short position against AI semiconductor stocks highlight policy, legal and valuation risks that warrant close attention.