Largan Precision will hold its earnings call on October 8, when the optical lens maker is due to report third-quarter results and outline its fourth-quarter outlook. With September revenue recovering and the company stepping up property acquisitions, investors are likely to focus on three issues: progress in co-packaged optics (CPO), capacity expansion and the pace of fourth-quarter customer orders.

At IROS 2026, one of the world's top robotics conferences, held in Pittsburgh at the end of September, a robot startup's booth featured a dual-arm robot catching, in midair, sticks dropped at random from above. The demonstration is one of many this year showing robots dancing, doing parkour, and even running half marathons—but the real question is what those performances actually prove about physical AI.

Kioxia helped create the NAND flash industry, but AI is changing what leadership in memory looks like. The Japanese supplier is benefiting from record NAND demand and higher prices, yet rivals are capturing more value from enterprise solid-state drives (SSDs) for AI servers. Kioxia's challenge is therefore less about making advanced NAND than turning it into qualified products, hyperscaler relationships, and committed volumes before today's AI infrastructure boom hardens the supplier hierarchy.

Demand for agentic AI in commercial and consumer markets was rising rapidly, and the market appeal of consumer services such as Meta Muse was expanding. In agentic AI tasks, large language model (LLM) inference was mainly handled by GPUs or application-specific integrated circuits (ASICs), while many other functions relied on CPUs, which was expected to boost CPU demand. Consumer-facing harness applications were expected to adopt enhanced general-purpose servers, while enterprise core automation would use high-density CPU racks deployed inside AI data centers.

The ferocious push toward artificial intelligence is gathering momentum amid growing industry demand for higher AI inference and increasingly compute-intensive models.

Semiconductors and AI data centers share a striking number of similarities.

The recent summit between US President Donald Trump and Chinese President Xi Jinping concluded without progress on automotive issues—namely, whether Chinese automakers might enter the US market or build manufacturing plants on American soil. While the prospect had previously sparked anxiety among American, European, and Japanese carmakers, the topic yielded no tangible breakthrough during the talks.

The rapid iteration of AI-capable hardware, following an industry-wide push toward integrating AI-processing capabilities into PCs, is bringing renewed attention to what AI could offer as its presence extends further into the notebook market. But beneath the expanding promise of these machines lies a less straightforward question: what is actually compelling buyers to bring them into their businesses and homes?

Data center AI accelerator package power has climbed from about 500W in the A100 generation to roughly 3,500–4,000W for Rubin Ultra-class products expected in 2027, pushing supply current into the thousands of amperes. In conventional frontside power delivery, current must cross the full metal interconnect stack before reaching the transistors, while power and signal lines compete for routing resources. Intel Foundry estimates that raising a single compute package from 1kW to 5kW increases power delivery network I²R losses from 89W to 2,222W, cutting the share of input power reaching the transistors from 66.0% to 41.8%.

When the underlying economics of infrastructure commitments meet the financing muscle of the companies supplying foundational compute, the traditional boundary between supplier and customer begins to blur.

New AI models arrive every few weeks, and the question of which is better often comes down to the leaderboard. Behind those rankings, evaluation can look like an academic exercise in setting rules, running tests, and assigning scores. However, public benchmarks—and strong placement on them—have long been tied to a lucrative business, with money flowing to different players in different eras.

Advanced Micro Devices (AMD) has agreed to acquire World Labs in an all-stock transaction valued at approximately US$8.2 billion. Following the closing, World Labs will operate as a dedicated frontier research organization for large language models (LLMs) and spatial intelligence, with founder Fei-Fei Li joining AMD as executive vice president and chief scientist, reporting directly to Chair and CEO Lisa Su.

China's rapid expansion of AI applications is driving a fresh wave of data center construction. SemiAnalysis estimates that the country has already delivered more than 24GW of data center capacity and will remain above that level by the end of 2026, making China the world's second-largest data center market after the US.

Nvidia's new AI agent security platform puts hardware at the centre of agent control, but it also forces enterprises to weigh stronger protection against higher infrastructure costs and deeper dependence on Nvidia's stack.

TSMC said the next phase of AI performance growth will depend less on shrinking a single chip and more on combining multiple chips, memory, and advanced packaging inside larger systems. The shift could reshape semiconductor competition, as demand for compute, bandwidth, power delivery, and cooling rises in tandem. The company outlined the trend at its heterogeneous integration forum during SEMICON Taiwan 2026, saying AI computing transistors in a single CoWoS package are set to rise more than 48-fold by 2029 from 2024 levels, while total HBM bandwidth is expected to increase more than 34-fold.