Compute & Infrastructure

Elon Musk stated via social media that his Terafab Project chipmaking venture will launch on March 22, 2026, describing it as a multi-billion dollar moonshot according to Tom's Hardware. No fabrication capacity details, technology node targets, manufacturing partners, or geographic location have been disclosed. The announcement comes as governments worldwide prioritise domestic semiconductor production to reduce dependence on Taiwan and South Korea for advanced chips.

The venture's viability depends entirely on details yet to be revealed. Building competitive fab capacity requires partnerships with equipment suppliers like ASML for extreme ultraviolet lithography tools, which have multi-year lead times and export restrictions. If Terafab aims to produce AI accelerators at leading nodes, it would compete directly with TSMC's established relationships with NVIDIA and other hyperscalers who have already reserved capacity years in advance.

Seven weeks after ASML announced 1,700 management position eliminations representing 4% of its 42,500-person global workforce, affected employees have received no clarity on their status, generating internal unrest according to Dutch broadcaster Omroep Brabant as reported by Tom's Hardware. The prolonged uncertainty at the monopoly supplier of extreme ultraviolet lithography equipment essential for sub-7nm chip production comes as semiconductor manufacturers race to expand capacity for AI accelerators.

ASML's tools are the sole enabler of advanced node production at TSMC, Samsung, and Intel fabs. Any operational disruption from workforce turbulence could ripple through the entire AI hardware supply chain. The company has not disclosed which divisions or geographies face cuts, leaving uncertainty about whether lithography tool development, manufacturing, or field support teams are affected.

A survey found 75% of Americans are aware of data centres but remain divided on approval, with most viewing them as environmentally negative according to The Register. The ambivalent public perception emerges as hyperscalers plan unprecedented data centre buildout to support AI training and inference workloads requiring gigawatts of power and massive water consumption for cooling.

The survey indicates grassroots environmental opposition could materialise into permitting delays, utility connection challenges, and local resistance that slows expansion. Microsoft, Google, Amazon, and Meta have each announced plans for dozens of new facilities over the next three years, but those timelines assume relatively frictionless local approval processes that this sentiment data suggests may not hold.

ASRock launched the H610M Combo II motherboard featuring one DDR4 slot and two DDR5 slots simultaneously, a hybrid design responding to ongoing RAM supply constraints as reported by Tom's Hardware. The unconventional architecture allows system builders to use whatever memory modules they can source, signalling continued volatility in DRAM markets despite industry transition from DDR4 to DDR5 that was supposed to be largely complete by 2026.

The persistence of memory supply issues complicates both consumer PC builds and enterprise server procurement at a time when AI inference servers require ever-larger memory pools. If leading manufacturers are resorting to hybrid architectures for low-end consumer boards, it suggests deeper supply-demand imbalances than industry roadmaps acknowledge.