Semicon Briefing

TSMC (from 2030) and Samsung (from 2028) have committed jointly with Intel to use ASML's High-NA EUV machines at $400M each – all three leading foundries now stand behind the same lithography platform. In parallel, all four companies founded the Large Mask Consortium to upgrade the photomask standard from 6 to 12 inches, thereby eliminating a 30% throughput loss.

Intel is the only manufacturer to have already processed over one million wafers on High-NA EUV systems, including production layers for the upcoming Panther Lake chip – putting Intel years ahead of Samsung's 2028 and TSMC's 2030 targets. This manufacturing maturity technology lead could warrant a reassessment of Intel's foundry strategy.

Qualcomm and Amazon have announced a multi-generation partnership for custom AI inference chips and optical connectivity up to 1.6 Tbps – a signal that hyperscalers are increasingly turning to custom silicon instead of Nvidia. The deal deepens Amazon's diversification strategy away from pure GPU clusters.

The US Department of Justice has launched an antitrust review of Nvidia's planned acquisition of Groq, burdening the deal with significant regulatory risk. Given Nvidia's already dominant position in the AI chip market, the review could result in a blockade or forced concessions.

Nvidia has agreed to acquire open-source AI platform Hugging Face for $12.9B, thereby combining hardware dominance with the leading model and community platform. This would be a tectonic step toward verticalizing the entire AI stack – from chips through frameworks to models.

Most-discussed export post of the week with 54 likes and 14 retweets: critics decry that the US, despite concerns about Chinese AI superiority, continues to export chips and manufacturing equipment – while Beijing pivots to domestic chips with a $532B plan. The structural effectiveness of US export controls is thereby openly called into question.