Semicon Briefing

Marvell has signed a custom AI chip deal with Google that could reach a volume of up to $120 billion by 2033 – revenue will be substantial only from 2029 onwards according to the CEO. The deal is one of the largest semiconductor contracts ever disclosed and confirms the trend toward high-volume custom ASIC partnerships between hyperscalers and chip design houses.

Intel's advanced packaging technology EMIB has achieved strong yields according to multiple industry sources and is being officially qualified by Google and MediaTek as an alternative to TSMC's CoWoS – a structural shift in the AI chip packaging supply chain. This is the first confirmed customer qualification that threatens TSMC with direct volume loss in its most profitable segment.

Both TSMC and Samsung have confirmed they will deploy ASML's High-NA EUV machines (0.55 NA) only at the 1-nm node – a milestone that according to Bloomberg is achievable at the earliest in 2030. Intel thus remains the sole customer in production and gains a time-limited technology advantage, while ASML must factor in revenue risks from delayed large orders.

Infineon completed the acquisition of ams OSRAM's analog sensor portfolio on July 1, 2026 for €570 million and thereby immediately expanded its automotive and industrial sensor business in an EPS-accretive manner. The transaction accelerates Infineon's pivot toward AI datacenter power supply and strengthens European semiconductor consolidation.

At ASPS 2026, Samsung's packaging lead confirmed adherence to a 415×510 mm panel for FOPLP targeting 2028, while TSMC's competing CoPoS format starts smaller – a direct format war for dominance in next-generation AI chip packaging. The format decision locks in customers long-term and determines market share in the most lucrative growth segment.

Anthropic has signed its largest infrastructure contract to date worth $45 billion for compute capacity with British cloud provider Nscale – in the same week the company won a legal dispute with the U.S. Department of Defense. The deal significantly increases structural demand for AI chips and datacenter capacity and puts further pressure on foundry and memory capacity.