Broadcom has begun shipping the industry's first 2nm custom compute SoC built on its 3.5D eXtreme Dimension System in Package (XDSiP) platform to Japan's Fujitsu, marking a concrete step from roadmap promise to commercial deployment in the AI infrastructure race.
The memory market is no longer just a component story — it is becoming a fault line running through the entire tech industry. As AI infrastructure buildout accelerates, cloud and data-center operators are consuming DRAM and NAND at a pace that is crowding out smartphone makers, distorting foundry economics, and forcing chipmakers to rethink how they secure supply. The consequences are rippling from factory floors in Asia to boardrooms in Silicon Valley.
Qualcomm has begun delivering rack-scale AI hardware and software systems for data centers, built around its AI 100 inference chip. The move signals a renewed push into a market where Nvidia and AMD currently set the standard.
When chipmakers spend big, it is usually the machines that cost the most. Process equipment — the lithography tools, deposition systems, and etch chambers that define the bleeding edge of semiconductor manufacturing — has historically commanded the largest share of TSMC's capital budget. Facilities and civil construction matter, but they have rarely led the bill. That conventional wisdom is now being tested.
A proposed US tariff and duty-exemption framework could force Korean memory suppliers to accelerate US fab investments or risk losing AI server market share, according to DIGITIMES analyst Luke Lin, speaking on a DIGITIMES podcast.
TSMC's decision to upgrade its second Kumamoto fab to 3nm reflects long-term customer demand and supply chain strategy rather than competitive pressure from Rapidus, according to Luke Lin, speaking on a DIGITIMES podcast.
Intel's 14A process node has become a focal point for the semiconductor industry, with market observers viewing it as a critical test of the company's manufacturing resurgence strategy. The advanced node has attracted significant attention amid speculation that major customers like Apple and Nvidia could adopt it for high-volume production. However, recent developments suggest these optimistic timelines may need recalibration.
Liquid cooling adoption has been led by Nvidia servers, with ASIC servers now accelerating the shift. DIGITIMES Research estimates liquid cooling will account for 50% of AI server thermal solutions in 2026, driven largely by ASIC platforms. Industry sources said cooling suppliers are not only targeting Nvidia GPU systems but are increasingly competing for ASIC server orders due to higher margins.
Blue Origin recently unveiled plans to deploy its TeraWave satellite constellation starting in the fourth quarter of 2027, signaling a strategic push into space communications beyond traditional launch services. According to DIGITIMES Research, TeraWave's core objective is not to compete in consumer broadband but to establish a space-based optical backhaul network, using satellites to handle high-capacity data transmission across regions and continents.
