This week’s semiconductor updates highlight major investments, product launches, and policy changes shaping the global chip industry.
1. Intel Expands AI Accelerator Portfolio with New Data Center GPU ‘Crescent Island’
Intel unveiled its new data centre GPU, Crescent Island, at the 2025 OCP Global Summit. Built on the Xe3P architecture, it features 160 GB of LPDDR5X memory and is optimised for AI inference workloads. Designed for air-cooled enterprise servers, it supports a wide range of data types for token-based inference and complements Intel’s Xeon 6 processors to deliver scalable AI performance from edge to cloud.
The GPU focuses on efficiency and scalability, addressing low-latency, real-time AI applications. Customer sampling is expected in the second half of 2026. Alongside this, Intel also showcased its Xeon 6+ Clearwater Forest CPU built on the 18A node, further strengthening its data-centre roadmap for AI-driven workloads.
2. NVIDIA and TSMC Mark Milestone with First U.S.-Made Blackwell Wafer
NVIDIA and TSMC celebrated the first production of the Blackwell wafer on U.S. soil at TSMC’s Phoenix facility. The wafer marks the beginning of domestic AI chip manufacturing for NVIDIA, underscoring a major step toward building resilient U.S. semiconductor supply chains.
The collaboration strengthens local AI infrastructure, reducing dependence on overseas fabs. TSMC’s Arizona plant supports 2 nm to 4 nm processes, while NVIDIA’s Blackwell GPUs — key to next-gen AI data centres — deliver high compute density and power efficiency. The companies will also integrate robotics and digital-twin technology into new facilities to optimise operations.
3. Infineon Accelerates Transition to 800 V AI Data Center Power Systems
Infineon introduced advanced solutions supporting the 800 V direct-current (VDC) power architecture proposed by NVIDIA for AI data centres. By combining SiC, GaN, and silicon technologies, Infineon aims to cut power losses and improve efficiency for megawatt-scale infrastructure.
Its CoolSiC™ JFET and hot-swap controller technologies enable safe board exchange while maintaining uptime — achieving up to 98% efficiency per conversion stage. Collaborations with STMicroelectronics and Navitas Semiconductor further advance 800 V DC solutions, setting a new benchmark for sustainable, high-density data-centre design.
4. TSMC Reports Strong Q3 2025 Results with NT$17.44 Earnings per Share
TSMC reported Q3 2025 revenue of NT$989.92 billion (US$33.1 billion) and net income of NT$452.3 billion, up 39% year-on-year. Advanced nodes (7 nm and below) made up 74% of total wafer revenue, led by strong demand for AI and HPC applications.
For Q4, TSMC expects revenue between US$32.2–33.4 billion and stable margins above 59%. The company anticipates continued AI-driven growth through 2026, reinforcing its role as the backbone of global semiconductor manufacturing.
5. ASML Reports €7.5 Billion Q3 Revenue and Confirms 2025 Growth Outlook
ASML achieved €7.5 billion in Q3 revenue, with a 51.6% gross margin and €2.1 billion in net income. Bookings reached €5.4 billion, including €3.6 billion from EUV systems. The company expects around 15% full-year revenue growth, supported by the ramp-up of EUV and High-NA EUV adoption.
ASML also shipped its new TWINSCAN XT:260 i-line scanner, offering 4× higher throughput for advanced packaging. Partnering with Mistral AI, ASML continues to integrate AI-driven system optimisation. The company confirmed a €1.60 interim dividend and progress on its €12 billion share buyback programme.
6. Oracle and AMD Extend Collaboration to Power Next-Generation AI Superclusters
Oracle and AMD announced an expanded partnership to deploy a massive AI supercluster featuring 50,000 AMD Instinct MI450 GPUs on Oracle Cloud Infrastructure (OCI) starting in 2026. The project will scale to over 131,000 GPUs per cluster, powering advanced AI training and inference workloads.
Built on AMD’s “Helios” architecture with EPYC CPUs and Pensando networking, the system uses liquid-cooled racks for high-efficiency performance. The open-source ROCm software stack ensures compatibility across AI frameworks, making the platform highly flexible for enterprise AI adoption.
7. Nscale and Microsoft Sign Landmark Agreement for 200,000 NVIDIA GB300 GPUs
Nscale and Microsoft announced a record AI infrastructure deal to deploy around 200,000 NVIDIA GB300 GPUs across the U.S. and Europe. Key sites include Texas (104,000 GPUs), Portugal, the UK, and Norway, supporting over 1.2 GW of combined capacity.
The collaboration strengthens sustainable AI infrastructure aligned with the UK-US Technology Partnership. By leveraging renewable power and high-efficiency design, the partnership boosts Microsoft Azure’s global compute availability and cements Nscale’s role in hyperscale GPU deployments.
8. KAIST Introduces ‘PIMBA’, a Processing-in-Memory AI Semiconductor Brain
KAIST researchers, alongside Georgia Tech and Uppsala University, unveiled PIMBA, a Processing-in-Memory AI semiconductor that combines Transformer and Mamba architectures. The chip delivers 4× faster inference and 2.2× lower power consumption than traditional GPU systems.
By executing computations directly within memory arrays, PIMBA reduces data transfer and heat, improving efficiency in large-language model processing. The research — recognised with the Gold Prize at the Samsung Humantech Paper Awards — will be presented at MICRO 2025 in Seoul.
9. GlobalFoundries and Partners Join Imec Automotive Chiplet Program
Imec announced that GlobalFoundries, Infineon, Silicon Box, STATS ChipPAC, and TIER IV have joined its Automotive Chiplet Program (ACP). The initiative focuses on scalable chiplet architectures for next-generation vehicles, including autonomous and in-vehicle computing systems.
Chiplet-based designs offer modularity, lower costs, and greater flexibility compared to traditional SoCs. This collaboration strengthens the ecosystem for automotive-grade chiplet integration and accelerates the development of safer, software-defined vehicle platforms.
10. SEMI Foundation and NSF Launch National RFP for Microelectronics Workforce Development
The SEMI Foundation, in partnership with the U.S. National Science Foundation, launched a national Request for Proposals (RFP) under the National Network for Microelectronics Education. The program will establish Regional Nodes across the U.S. to expand semiconductor talent pipelines and bridge industry-academic collaboration.
The initiative aims to create consistent training pathways, promote diversity in STEM, and support regional semiconductor ecosystems aligned with the CHIPS and Science Act. This marks another major step in addressing the industry’s long-term workforce challenge.
