Taiwan's government has limited budgets to build the AI computing infrastructure the country needs, and infrastructure that a government runs itself cannot easily turn a profit. To get around that constraint, the Ministry of Digital Affairs (MODA) introduced the AI computing center BOO (build-operate-own) public-private partnership project in April 2026, aiming to let private companies build and operate the infrastructure while tax incentives make the investment attractive to them.

The 2026 China International Optoelectronic Exposition (CIOE) closed on September 11. While near-eye display applications such as AR/VR and smart glasses drew attention, vendors' latest moves in optical communications also attracted strong industry interest.

AI data centers are moving decisively toward 1.6T connectivity, with 800G now mainstream in AI clusters and 400G remaining common in general cloud and enterprise deployments.

LG Display announced on September 13 that it is expanding its company-wide artificial intelligence transformation (AX) initiative by scaling up its "AX SQUAD" program. Designed to empower employees across all operational tiers to build their own workflow automation solutions, the initiative bridges frontline operational knowledge with technical software support.

Apple's new iPhone 18 series and first foldable, the iPhone Duo, could reshape smartphone demand worldwide by pushing some buyers toward Android rivals, channel players said. Higher launch prices may expand the premium market, but they could also slow some iPhone sales and intensify competition across midrange and flagship segments.

A worldwide nuclear comeback, fueled by artificial intelligence data centers and semiconductor demand, is tightening supply chains and labor markets just as Taiwan considers restarting idle reactors.

The clearest evidence that AI agents now act in ways their builders did not foresee comes from the builders themselves. Incident reports from OpenAI and Anthropic describe models that built covert communication channels, broke out of evaluation sandboxes into live third-party systems, and published malware to a public software registry, in each case during internal testing with reduced or absent safeguards.

As AI computing bandwidth keeps rising, optical interconnect architectures are moving beyond pluggable optical modules toward near-packaged optics (NPO) and co-packaged optics (CPO). But large-scale deployment will depend on more than optical-engine performance. Precision alignment, serviceability, manufacturing yield, and whether the supply chains for light sources, packaging, testing, and materials can advance in step are becoming equally critical.

The AI industry's proposal to slow frontier model development ran into geopolitics over the weekend. US President Donald Trump rejected a slowdown, saying America must stay ahead of China, while Chinese President Xi Jinping used the BRICS summit in New Delhi to pitch a China-led open-source AI network for developing economies. The executive who proposed pacing concedes that a verifiable US-China limit may not be achievable.

The strongest case for restraining frontier AI development is now coming from the companies building it, and one of the sharpest rebuttals is from the supplier that sells them the chips. That split is the political problem US lawmakers inherit: the labs want a coordinated slowdown and independent oversight they say they cannot arrange alone, while Nvidia Chief Executive Jensen Huang has suggested the alarm over AI-driven cyber risk doubles as a sales pitch.