This week shows where momentum is building in early 2026: AI infrastructure, silicon photonics, edge AI, memory R&D, and industrial policy.
The main signals are clear: rising demand for advanced compute, growing importance of optical interconnects, tighter links between research and manufacturing, and continued government support.
It also highlights how the industry is expanding beyond logic and memory, with progress in photonics, quantum, and key supply-chain areas like helium, all driven by AI and long-term capacity buildout.
1) Global Foundry Revenue Rises in 4Q25 on AI-Driven Demand
Advanced nodes remain the main growth engine
The world’s top ten foundries generated $46.3 billion in 4Q25 revenue, up 2.6% quarter on quarter, mainly driven by strong demand for AI GPUs, TPUs, and flagship smartphone processors.
TSMC remained far ahead with $33.7 billion in revenue and 70.4% market share, while Samsung Foundry grew to $3.4 billion, helped by early 2nm ramp and HBM4-related logic demand. SMIC and UMC also posted growth, supported by local demand and stable utilization.
Mature nodes stay important across AI and industrial demand
Demand for PMICs and MCUs kept 8-inch fabs busy, especially in servers and edge AI.
GlobalFoundries grew strongly on data-center peripheral demand, while Tower Semiconductor moved up in the rankings thanks to rising silicon photonics production. This shows that AI growth is not only helping leading-edge foundries, but also several specialty and mature-node players.
2) NVIDIA–Nebius Partnership Signals Next Phase of AI Cloud Infrastructure Scaling
$2 billion investment supports hyperscale AI expansion
NVIDIA announced a strategic partnership with Nebius, including a $2 billion commitment to help build more than 5 GW of NVIDIA-based AI compute capacity by 2030.
Nebius will gain early access to next-generation NVIDIA platforms, including Rubin GPUs, Vera CPUs, and BlueField systems, helping it scale faster as demand for AI infrastructure continues to rise.
AI cloud design becomes more vertically integrated
The partnership reflects a bigger industry shift toward full-stack AI cloud platforms, where silicon, networking, infrastructure, and software are designed together.
A key focus is inference efficiency, GPU utilization, and system-level optimization, which are becoming just as important as raw training capacity.
3) ABB Robotics Integrates NVIDIA Omniverse to Improve Industrial-Grade Physical AI
Simulation gets much closer to real-world robot behavior
ABB Robotics is integrating NVIDIA Omniverse into its RobotStudio platform to create a new system called RobotStudio HyperReality, due in the second half of 2026.
ABB says the platform can achieve up to 99% correlation between simulated and physical robot behavior, helping manufacturers reduce engineering time and deployment costs.
Important for electronics and precision manufacturing
The platform combines physics-based simulation, synthetic data generation, and photorealistic rendering. ABB says this could cut deployment costs by up to 40% and reduce time-to-market by around 50%.
It is especially relevant for electronics and semiconductor-related assembly, where positioning accuracy and low tolerance windows matter.
4) STMicroelectronics Moves Silicon Photonics Platform into High-Volume Production
PIC100 enters large-scale manufacturing
STMicroelectronics has started high-volume production of its PIC100 silicon photonics platform on 300mm lines.
The platform supports 800G and 1.6T optical transceivers, which are increasingly needed in hyperscale data centers and AI clusters to improve bandwidth and energy efficiency.
Roadmap extends toward co-packaged optics
ST plans to expand PIC100 output fourfold by 2027 and is also developing PIC100 TSV, which uses through-silicon vias to improve density and thermal performance.
This supports the long-term shift from pluggable optics toward near-packaged and co-packaged optics, as AI infrastructure pushes optical interconnects to the center of system design.
5) Texas Instruments Introduces MCU Families With Integrated NPUs
Edge AI moves into lower-cost devices
Texas Instruments launched two new MCU families with its TinyEngine neural processing unit, bringing AI acceleration into smaller and cheaper embedded systems.
TI says the NPU can deliver up to 90× lower inference latency and 120× lower energy per inference, making it useful for wearables, appliances, and small industrial devices.
AI and motor control now run together on one chip
The new AM13Ex family combines real-time control and AI in a single MCU, allowing functions like predictive maintenance and adaptive motor control without needing extra processors.
TI says this can cut bill-of-materials costs by up to 30%, which is important for cost-sensitive industrial and embedded applications.
6) Intel Introduces Core Series 2 for Deterministic Edge Computing
Built for real-time industrial workloads
Intel launched its Core Series 2 processors with P-cores for industrial and embedded edge applications, focusing on deterministic performance rather than just peak compute.
Intel says the chips offer lower PCIe latency, more predictable response time, and better multi-threaded performance than a comparable AMD desktop chip in similar power range.
Healthcare AI suite expands edge use cases
Intel also introduced a preview of its Health & Life Sciences Edge AI Suite, aimed at workloads such as ECG analysis, contactless vital-sign monitoring, and anonymous 3D patient tracking.
Together, the new processors and software show Intel is pushing a broader edge AI strategy across manufacturing, healthcare, and automation.
7) Ayar Labs and Wiwynn Bring Co-Packaged Optics to Rack-Scale AI Systems
Optical interconnects move from chip level to rack level
Ayar Labs and Wiwynn announced a partnership to develop rack-scale AI infrastructure built around co-packaged optics.
The system combines Ayar’s optical engines and remote light source with Wiwynn’s rack-level architecture, aiming to solve the bandwidth and power limits of copper interconnects in large AI clusters.
Designed for hyperscale deployment
The architecture is built for more than 1,024 AI accelerators per rack system, with heavy focus on manufacturability, serviceability, fiber routing, cooling, and power delivery.
This signals that co-packaged optics is moving closer to real deployment in hyperscale AI infrastructure.
8) Applied Materials Expands U.S. Memory R&D Collaborations
Micron and SK hynix partnerships target next-gen memory
Applied Materials announced two major R&D partnerships, one with Micron and one with SK hynix, both focused on DRAM, HBM, NAND, and advanced packaging for AI memory systems.
The goal is to speed up development of new memory architectures as AI workloads continue to widen the gap between processor speed and memory bandwidth.
Company also raises dividend
Applied also increased its quarterly dividend by 15%, marking its ninth consecutive annual increase.
That reflects confidence in long-term cash flow, while the memory partnerships show where the company sees the next big manufacturing challenges.
9) IBM and Lam Research Expand Work on Sub-1nm Logic Scaling
New five-year program targets future logic nodes
IBM and Lam Research launched a new five-year collaboration focused on sub-1nm logic scaling, including new materials, etch and deposition flows, and High-NA EUV process integration.
The work builds on earlier joint efforts around 7nm, nanosheet devices, and IBM’s 2nm demonstration.
High-NA EUV still needs full process integration
A major challenge is not just printing smaller patterns, but reliably transferring those patterns into device layers with good yield.
The partnership aims to build production-ready process flows for future nanosheet and nanostack devices, including backside power delivery.
10) IBM Introduces Quantum-Centric Supercomputing Architecture
New model combines QPUs with CPUs and GPUs
IBM published what it calls the first reference architecture for quantum-centric supercomputing, showing how quantum processors can work alongside classical HPC systems.
The architecture combines quantum hardware, large CPU/GPU clusters, and orchestration software to run hybrid workloads in chemistry, materials science, and optimization.
Early demonstrations show practical scientific value
IBM and research partners have already used the approach on molecular and materials simulations, including protein structures and electronic-structure problems.
For the semiconductor industry, this matters because hybrid quantum-classical systems could eventually help with physics-heavy workloads such as materials discovery and advanced device modeling.
11) Intel Demonstrates 3nm FHE Accelerator With Major Speedups
Encrypted computing gets a dedicated chip architecture
Intel demonstrated Heracles, a custom fully homomorphic encryption accelerator built on its 3nm process and paired with 48 GB of HBM.
FHE allows computation on encrypted data, but it is usually extremely slow. Intel says Heracles speeds up key operations by 1,000× to more than 5,000× compared with CPU-based execution.
Privacy-preserving AI becomes more realistic
The chip is aimed at large-scale encrypted computing workloads, which could become important in private AI inference, secure cloud services, and confidential search systems.
Intel has not yet announced commercial plans, but the demo shows hardware acceleration for FHE is becoming a real part of the future compute roadmap.
12) UMC Expands Thin-Film Lithium Niobate Manufacturing
TFLN photonics moves toward foundry-scale production
HyperLight, UMC, and Wavetek announced a manufacturing partnership to bring thin-film lithium niobate photonics into high-volume foundry production on 6-inch and 8-inch wafers.
The target markets include datacom, coherent optics, telecom modules, and co-packaged optics for AI systems.
Jabil added to support deployment at scale
A second announcement added Jabil to the chain, bringing packaging and system-integration support.
This matters because TFLN is being positioned as a promising material for 1.6T+ optical interconnects, with lower drive voltages and better power efficiency.
13) TSMC Posts Strong Year-on-Year Growth Despite February Dip
Seasonal decline does not change broader growth trend
TSMC reported NT$317.66 billion in February 2026 revenue, down 20.8% month on month, but still up 22.2% year on year.
The monthly drop looks seasonal, while the yearly gain shows continued demand for advanced-node logic tied to AI, HPC, and 5G.
First two months of 2026 remain strong
Revenue for January and February combined reached NT$718.91 billion, up 29.9% compared with the same period in 2025.
That points to a much stronger start to the year than the industry saw during the earlier inventory correction cycle.
14) Scintil Photonics Introduces DWDM Laser Source Evaluation Kit
DWDM optical architectures gain ground in AI scale-up networks
Scintil Photonics launched its LEAF Light evaluation kit at OFC 2026, allowing customers to test its single-chip DWDM laser architecture for AI networking.
The platform is designed for GPU scale-up networks, where copper interconnects struggle with power, bandwidth, and reach.
Step toward production-ready embedded light sources
The kit includes up to 16 wavelengths, temperature control, tuning features, and monitoring tools that help customers test production-like behaviour before full module deployment.
This is an important step toward more scalable DWDM-based co-packaged optics in large AI clusters.
15) Imec Launches European University Consortium for CMOS 2.0 Research
26 academic groups join next-generation chip research effort
Imec created a new European university consortium focused on CMOS 2.0, bringing together 26 academic groups to work on post-CMOS design, architecture, and heterogeneous 3D integration.
The consortium is linked to imec’s NanoIC pilot line, giving researchers access to early PDKs and advanced process technologies.
Europe tries to strengthen long-term talent and technology base
The program funds 26 PhD positions and aims to connect academic research more directly with future industrial chip design challenges.
The broader goal is to help Europe stay competitive in next-generation compute platforms built around 3D stacking and function-partitioned architectures.
16) IonQ Expands UK Quantum Commercialization Through Cambridge Partnership
New quantum center planned at Cambridge
IonQ and the University of Cambridge announced a new Quantum Innovation Centre, which will include deployment of IonQ’s 256-qubit chip-based system.
The center is meant to accelerate commercialization of quantum technologies across computing, networking, sensing, and security.
Aligned with the UK’s national quantum strategy
The collaboration links directly with the UK’s broader quantum programs and is intended to support both workforce development and industry-facing research.
It gives companies and researchers earlier access to commercially relevant quantum hardware and cloud tools.
17) Japan Expands Strategic Investment Push Toward AI, Robotics, and Semiconductors
61 technologies identified as national priorities
Japan has officially listed 61 products and technologies as investment priorities across 17 strategic sectors, including semiconductors, AI, robotics, quantum, and advanced materials.
The government also reaffirmed two major goals: capturing more than 30% of the global AI robotics market by 2040 and raising domestic semiconductor revenue to ¥40 trillion by 2040.
Could reshape capital flows across Japanese industry
The plan is likely to influence investment across semiconductor, automation, materials, and tool ecosystems.
At the same time, questions remain about funding, energy infrastructure, and whether such a broad list of priorities risks spreading resources too thinly.
18) Industrialization of 300mm Silicon Photonics Advances Optical Interconnect Roadmap
CEA-Leti and NcodiN push nanolaser-enabled optical interposers
CEA-Leti and startup NcodiN announced a partnership to industrialize nanolaser-based optical interposers on a 300mm silicon photonics platform.
The technology aims to address one of AI hardware’s biggest bottlenecks: moving massive amounts of data without exceeding power and thermal limits.
300mm manufacturing is key for scaling
Moving to a CMOS-compatible 300mm platform is critical if this approach is to reach the yield, uniformity, and cost needed for real AI hardware deployment.
This is another sign that optical interconnects are moving from lab concepts toward manufacturable infrastructure technology.
19) Helium Supply Instability Becomes a Critical Constraint for Semiconductor Manufacturing
Middle East disruptions highlight a major materials risk
The semiconductor industry is facing renewed concern over helium supply, as geopolitical disruption in the Middle East affects access to a gas that is essential in advanced semiconductor manufacturing.
Helium is used for wafer cooling, temperature stability, vacuum systems, and leak testing, and there is no large-scale substitute.
From pricing problem to continuity risk
Because Qatar accounts for roughly a third of global helium output, prolonged disruption could create shortages lasting months, not weeks.
This makes helium a strategic bottleneck for fabs, especially as AI-related manufacturing demand keeps rising.
20) China’s Semiconductor Export Surge Reflects Strong Domestic Capacity Growth
Chip exports rise sharply in early 2026
China’s semiconductor exports reached $43.3 billion in the first two months of 2026, up 72.6% year on year, while export volume rose 13.7%.
This reflects both domestic capacity growth and strong global AI-related demand.
Domestic output expands, but high-end AI constraints remain
SMIC and Hua Hong both increased output, and the overall Chinese semiconductor market is expected to grow strongly in 2026.
At the same time, access to top-end AI accelerators remains constrained by U.S. policy, which means China continues to push harder on domestic alternatives and state coordination.
21) FAMES Launches 2026 Open-Access Call
Europe expands access to advanced pilot-line technologies
The FAMES Pilot Line opened its 2026 call, adding new PDKs for RF filters, RF switches, magnetics-on-silicon, and early 10nm FD-SOI pathfinding.
The initiative is designed to give European companies and research groups access to advanced technologies that support RF, power management, NVM, and 3D integration.
Important piece of Europe’s semiconductor roadmap
FAMES brings together major European research institutes and universities under the Chips Joint Undertaking.
It gives Europe a shared platform for experimentation and early silicon work across several strategic technologies.
22) Sivers Semiconductors Expands Photonics Footprint in LiDAR
Customer production ramp expected in Q4 2026
Sivers Semiconductors announced that a strategic LiDAR customer has integrated its DFB lasers and optical amplifiers across multiple sensing platforms, with production ramp planned for Q4 2026.
The company sees this as part of a larger growth opportunity in automotive and industrial LiDAR.
Photonics business continues to diversify
The announcement points to possible multi-year revenue across future LiDAR platform generations.
It also shows that Sivers is building photonics exposure not only in AI datacenters, but also in automotive sensing, robotics, and industrial automation.
23) Europe Advances Industrial-Scale Photonic Chip Manufacturing
New 6-inch InP photonic chip factory launched in the Netherlands
The Netherlands has started building what is described as the world’s first industrial facility dedicated to 6-inch indium phosphide photonic chips.
The project is led by TNO and High Tech Campus Eindhoven and is meant to bridge the gap between photonics R&D and scalable manufacturing.
