Memory supply remains one of the clearest constraints on AIartificial intelligence expansion. SK hynix expects DRAM1. dynamic random access memoryMemory technology or system for temporarily or permanently storing digital information.2. dynamic random-access memoryDRAM is a volatile memory technology and HBM1. human body modelESD test model that simulates a person discharging static electricity into a component.2. High Bandwidth MemoryHBM is a stacked DRAM architecture designed to provide very high memory bandwidth. volumes to stay effectively sold out through 2029, while Samsung sees tight supply through 2028 and says its 2027 HBM1. human body modelESD test model that simulates a person discharging static electricity into a component.2. High Bandwidth MemoryHBM is a stacked DRAM architecture designed to provide very high memory bandwidth. supply is already committed. Samsung points to long fab build times, extended tool lead times, larger dies, higher process complexity in next-generation HBM1. human body modelESD test model that simulates a person discharging static electricity into a component.2. High Bandwidth MemoryHBM is a stacked DRAM architecture designed to provide very high memory bandwidth., and the continued migration of DRAM1. dynamic random access memoryMemory technology or system for temporarily or permanently storing digital information.2. dynamic random-access memoryDRAM is a volatile memory technology capacity into HBM1. human body modelESD test model that simulates a person discharging static electricity into a component.2. High Bandwidth MemoryHBM is a stacked DRAM architecture designed to provide very high memory bandwidth..
Capacity pressure is also showing up in advanced packaging. ASEAdvanced Semiconductor EngineeringASE is a semiconductor assembly and packaging company involved in the joint research program. expects its advanced-packaging revenue to double in 2027, and Amkor is ramping new high-density fan-out and 2.5D programs. TSMCTaiwan Semiconductor Manufacturing CompanyTSMC is named as a founding partner of the EPIC Center. is reportedly weighing a 20% price increase for CoWoS as AIartificial intelligence demand keeps packaging lines constrained. Applied Materials also plans to raise its stake in Besi to about 20%, extending a partnership that now reaches beyond hybrid bonding into new interconnect architectures and other advanced packaging technologies for AIartificial intelligence.
Several manufacturing moves underline the broader capital cycle. Intel Foundry reported improving execution in the third quarter, citing better yields and cycle times on 18A. The company also released its 14A 0.9 PDKprocess design kitA PDK provides the technology files and design rules needed to design chips for a specific process. in October and continues to engage prospective external customers on 18A-P and 14A, while advancing EMIB-T advanced packaging. Infineon plans to invest €1.1 billion through 2034 at Villach to build Europe’s first high-volume manufacturing site for next-generation GaNgallium nitrideGaN is another material platform mentioned for use in different power delivery roles. power semiconductors on 300 mm wafers. UMC approved an additional $3.2 billion in capital spending for capacity expansion in Singapore and Tainan.
Lam Research and imec expanded their collaboration on advanced 3D integration, next-generation memory, and high-NAnumerical apertureOptical measure of the light-gathering angle of a lens, related to resolution. EUVextreme ultravioletExtremely short-wavelength ultraviolet radiation; EUV lithography uses about 13.5 nm light. patterning. The work covers deposition, etch, and plasma dicing for advanced packaging, along with Lam’s Aether dry resist technology.
On the AIartificial intelligence infrastructure side, Cadence introduced domain-specialized AIartificial intelligence models for agentic chip and system design, while Arm defined a new CPUcentral processing unitMain processor that executes program instructions and performs data operations. benchmarking methodology for agentic AIartificial intelligence workloads. Google is expanding its custom accelerators, with TPUthermal processing unitThermal process in which a device or material undergoes controlled heating.s moving into broader deployment and Oracle planning to offer them through its cloud infrastructure from 2027, alongside Arm’s AGI CPUcentral processing unitMain processor that executes program instructions and performs data operations.s in fall 2027.
Funding and acquisitions also remain active. Intel plans to acquire Classiq to add automated quantum algorithm synthesis software to its silicon spin-qubit technology and manufacturing. FuriosaAI raised $620 million to scale production of its AIartificial intelligence inference accelerator, while Anthropic said HBM1. human body modelESD test model that simulates a person discharging static electricity into a component.2. High Bandwidth MemoryHBM is a stacked DRAM architecture designed to provide very high memory bandwidth. shortages and limited TSMCTaiwan Semiconductor Manufacturing CompanyTSMC is named as a founding partner of the EPIC Center. capacity could constrain its own AIartificial intelligence expansion.
Quarterly results from several companies reflect the same demand mix. Samsung reported revenue of KRW 195.3 trillion, up 127%, with a major HBM4 ramp. SK hynix posted record revenue of KRW 104.6 trillion, up 328%. Amkor, ASEAdvanced Semiconductor EngineeringASE is a semiconductor assembly and packaging company involved in the joint research program., Cadence, Intel, Rambus, Teradyne and UMC also reported strong year-over-year growth, with tester demand, backlog, utilization and foundry revenue all pointing to continued pressure across the semiconductor supply chain.