Google will expand its AI infrastructure research and development center in Taipei's Shilin district, adding 60% more office space, as it deepens work with Taiwan's supply chain on next-generation computing. The move underscores Taiwan's growing role in global AI development, where chip, cooling, and data-center advances affect users worldwide.

The most consequential figures on HPE's fiscal third-quarter earnings call on September 2 were not the results but the commitments standing behind them. According to Bloomberg's transcript of the call, the company has more than doubled its networking purchase commitments in a single quarter, carried US$11.8 billion of inventory into the fourth quarter, and signed multi-year supply agreements locking component capacity it can reallocate every 90 days. With two newly disclosed anchor deals, that pre-buying is what underwrites the roughly US$4 billion added to the fiscal 2027 revenue framework over the past quarter.

Hewlett Packard Enterprise's fiscal third-quarter results, reported on September 2, are the point at which the Juniper Networks deal and the enterprise AI cycle stopped being a cost story and became a margin story. Revenue reached US$12.2 billion in the three months to July 31, up 34% from a year earlier, and GAAP operating margin widened to 11.4% from 2.7%. What now limits the business is neither demand nor integration. It is component supply.

Cooling specialist Auras Technology is expanding from GPU platforms to ASIC-related products as its next growth driver, amid the ongoing shift toward liquid cooling in AI servers, with chairman Steve Lin highlighting sustained progress in projects for major international cloud service providers (CSP) such as AWS and Meta.

I visited Automation Taipei 2026 in mid-August, where the show looked lively and occupied a footprint 7% larger than in 2025. Physical AI and vision-language-action (VLA) models dominated the venue. Yet most robots on display still repeated fixed motions behind barriers, highlighting how far robotics has to go to move from automation to autonomy.

SB Energy, the data center and power infrastructure company under SoftBank Group, filed for an IPO with the US Securities and Exchange Commission (SEC) on September 1, planning to list under the ticker "SBE" on Nasdaq and Nasdaq Texas. The deal could value the company at more than US$50 billion, while Nvidia's promised US$1.5 billion investment has emerged as the most prominent backing for the offering.

Thailand has launched its TH-AI Passport and core AiPASS platform, taking a different path from South Korea's national AI push by buying access to overseas commercial AI tools such as ChatGPT and Claude and offering them free to citizens. The move aims to expand AI access while addressing the country's widening AI usage gap.

The rise of physical AI is creating a new wave of demand for advanced sensors as robots move from controlled demonstrations toward practical deployment. Industry experts at SEMICON Taiwan 2026 said robots will need increasingly diverse sensing capabilities to perceive, interpret, and respond to complex real-world environments in real time.

A senior Lam Research executive told an industry audience this week that the next major leap in semiconductor manufacturing will not come from removing humans entirely from fabs, but from pairing collaborative robots with AI-driven predictive maintenance to close a long-standing gap in fab operations.

At a shared stage at Semicon Taiwan 2026, executives from Samsung Electronics, SK Hynix, and Micron agreed that traditional 2.5D high-bandwidth memory (HBM) using interposers is approaching a performance ceiling. Consequently, the industry is transitioning toward 3D vertical stacking architectures to minimize transmission power consumption and leverage wafer-level bonding technologies.

Four days after listing on Nasdaq, Pasqal came to SEMICON Taiwan with an argument aimed squarely at its competitors: the industry's assumption that quantum computing delivers nothing useful before fault tolerance arrives around 2030 is wrong, and the French company intends to sell against it.

Cloud AI capital expenditure continues to shatter market expectations, sparking growing concerns over whether the spending surge is overheating and heading toward a bubble. Semicon Taiwan 2026 may serve as the ultimate litmus test for this debate.

Cable manufacturer Wonderful Hi-Tech announced its financial results for the second quarter and first half of 2026 during an earnings call held on September 1. A ramp-up in high-speed transmission cable shipments for artificial intelligence (AI) applications, coupled with strong pull-in momentum from a major US low-Earth-orbit (LEO) satellite customer, drove second-quarter results. The company also said it is expanding capacity in Thailand and Vietnam to capture new opportunities in space data centers and physical AI.

As AI and high-performance computing (HPC) continue to drive semiconductor advances, industry competition is shifting beyond chip performance toward packaging, cooling, assembly, and power delivery. With GPUs, high-bandwidth memory (HBM), and advanced packaging pushing power density higher, chipmakers now face challenges that extend well beyond silicon manufacturing, including managing thermal expansion, warpage, heat dissipation, and reliability across heterogeneously integrated materials.

Elon Musk has restructured the SpaceX leadership team tasked with constructing the company's AI data centers, replacing several leaders with executives drawn from SpaceX's rocket and satellite internet divisions. The reshuffle follows mounting civil engineering concerns and persistent operational reliability issues at existing data center facilities in Tennessee and Mississippi. During its initial push to establish AI infrastructure rapidly, the company operated clusters for extended periods without backup cooling or power systems, significantly elevating the risk of system outages, sources told The Information.

Dell Technologies' quarterly results showed record AI server orders and backlog, but executives used the earnings call's question-and-answer session on September 1 to disclose a more detailed point not in the headline numbers: the pace at which enterprise customers, rather than just large cloud and sovereign buyers, are adopting its AI infrastructure.

Dell Technologies closed its fiscal second quarter with revenue of US$47.0 billion, up 58% year over year, but the more telling number is what happened below the top line: operating income more than tripled, and the operating margin nearly doubled to 11.5% from 6.0% a year earlier. AI infrastructure did not just add revenue this quarter — it pulled the entire company's profitability up with it.

DIGITIMES observed that as AI systems continued to scale, interconnect requirements inside AI data centers had expanded from chips, packages, and boards to racks, clusters, and eventually inter-data-center connections.

Anthropic has reportedly signed a cloud-computing contract worth US$35 billion with Lambda, an Nvidia-backed cloud provider, under which Nvidia itself holds the data center lease, The Wall Street Journal reported, citing people familiar with the deal. The site is being developed by Hut 8, a bitcoin miner turned data-center builder, in Nueces County, Texas, and Nvidia signed an agreement with Hut 8 a few weeks ago to secure capacity, according to the Journal.

For global cloud operators and network vendors, Cisco said co-packaged optics (CPO) is moving from lab promise to commercial necessity as AI data centers strain power budgets and bandwidth limits. The company said pluggable modules, near-packaged optics, and co-packaged optics will coexist, but the shift could reshape how hyperscale infrastructure is built worldwide.

Humanoid robot scaling is still constrained by price, and Chinese humanoid robots — which command the largest global market share — show a sharp gap between consumer and industrial models. Industrial versions demand high load, high precision, and strong stability, making several precision transmission parts a key battleground for Chinese players and Taiwan suppliers.

Nvidia's US$3.5 billion investment in MediaTek's overseas convertible bonds marks more than a financial bet. It underscores how the chip leader is trying to anchor its AI infrastructure strategy as cloud giants and model developers expand their use of custom silicon and rack-scale systems.

Nvidia and MediaTek are widening a global AI partnership that could reshape data center, edge, and PC hardware markets. MediaTek's US$3.9 billion overseas convertible bond sale, largely bought by Nvidia, signals a deeper strategic tie that may affect supply chains, customer choices, and competition worldwide.

Nidec Chaun-Choung Technology plans to establish a dedicated liquid-cooling team and deepen cooperation with major cloud service providers, citing rising thermal-management demand from AI servers. The company also plans to expand cooling capacity beyond its existing operations in Kunshan, China, to Hanoi, Vietnam, with mass production and shipments expected to begin in 2028.

Capex by the world's nine largest cloud service providers (CSPs) is expected to rise about 90% year over year in 2026, while AI servers are growing faster than conventional servers, increasing demand for power and energy-management solutions.