Can a Sparse-AI Hardware Architecture for Data Centers Work?
Weight‑pruning and activation sparsity can compress neural networks up to 50× and reduce compute by 2‑10×, creating a theoretical two‑order‑of‑magnitude MAC reduction. The Sparse Computing Core Technology (SCCT) architecture leverages a multibus design with private SRAM per lane, offering three modes—general‑purpose, SIMD, and very‑high‑throughput computing (VHTC). Cycle‑accurate simulations across 1,520 sparse matrices show SCCT converts 94.5% of MAC savings into wall‑clock speed, delivering a median 9.9× advantage over GPUs. The design targets up to 100× SIMD speedup and roughly 160× acceleration for softmax‑type operations, aiming to close the memory‑wall gap in data‑center AI workloads.
Keysight Enables End-to-End Electrical-Optical-Electrical Simulation for Data Center and Ethernet Design
Keysight Technologies launched an Electrical‑Optical‑Electrical (EOE) simulation capability in its ADS 2026 software, allowing engineers to model the complete signal chain—from electrical transmitter through photonic components to electrical receiver—in a single environment. The feature addresses the growing need for 800 Gbps...
Bull, Orange, Scaleway and Partners Unite on French AI Gigafactory Bid
A coalition of French and European tech leaders—including Bull, Orange, Scaleway, EDF, Capgemini, Artefact and Ardian—has formed the AION consortium to submit a bid for the EU’s AI Gigafactory program. The group aims to build a sovereign, high‑performance AI infrastructure...
Siemens EDA Expands AI and Advanced Packaging Collaboration with TSMC
Siemens EDA announced an expanded AI‑driven collaboration with TSMC at the 2026 Technology Symposium, targeting automated design rule checking, AI‑assisted physical verification, and intelligent design optimization via its Fuse AI System. The partnership includes certification of Siemens tools on TSMC’s...
Semidynamics Secures a Strategic Investment to Advance Memory-Centric AI Inference Chips
Semidynamics announced a strategic investment to accelerate its memory‑centric AI inference chips, aiming to overcome the industry’s “memory wall.” The funding will expand engineering teams, speed product development, and deepen ecosystem partnerships. By re‑architecting silicon around memory efficiency, the startup...
Amkor Secures More Land For U.S. Packaging Site
Amkor Technology, the world’s second‑largest OSAT, has acquired an additional 67‑acre parcel adjacent to its 104‑acre advanced packaging and test campus in Peoria, Arizona. The expanded 171‑acre site will host the first high‑volume advanced packaging OSAT facility in the United...
NVIDIA Vera: First In-House CPU Effort Lands with Leading AI Labs
NVIDIA has begun shipping its in‑house Vera CPU to four marquee AI customers—Anthropic, OpenAI, Oracle Cloud Infrastructure, and SpaceXAI. The processor packs 88 custom Olympus cores, up to 1.2 TB/s memory bandwidth and 3.4 TB/s fabric bandwidth, targeting the latency‑critical orchestration layer...
Amazon’s Trainium AI Chips Gain Traction with Developers as Software Matures and Nvidia GPUs Face Capacity Constraints
Amazon’s custom Trainium AI chips are gaining momentum as Anthropic and OpenAI secure large capacity commitments and developers migrate workloads thanks to maturing software. The AWS custom silicon portfolio now exceeds a $20 B annualized run rate, equivalent to roughly $50 B...
How Micron (MU) Is Targeting AI Data Centers With Higher-Capacity Server Memory
Micron Technology announced on May 12 that it has sampled a 256GB DDR5 registered DIMM built on its 1‑gamma DRAM process, delivering speeds up to 9,200 MT/s—about 40% faster than existing volume‑produced modules. The memory uses advanced 3D stacking and...
Firefly CSC2-N48SPK3 – A 2880 TOPS RISC-V AI Server with 48 SpacemiT K3 Nodes, 48 NVMe SSDs
Firefly has introduced the CSC2‑N48SPK3, a 2U rack‑mount server that packs 48 SpacemiT K3 RISC‑V compute nodes and a Rockchip RK3588 control node. Each K3 node delivers up to 60 TOPS of sparse AI performance, up to 32 GB LPDDR5 memory and...
Tata Electronics and ASML: India’s 300 Mm Fab Gets Its Lithography Backbone
Tata Electronics and ASML signed a memorandum of understanding on May 16, 2026 to back the construction and ramp‑up of India’s first commercial 300 mm semiconductor fab in Dholera, Gujarat. The partnership centers on delivering a suite of lithography tools, talent...
Pasqal and Aramco Launch QCaaS Platform for Saudi Quantum Computer
Aramco and French quantum‑computing firm Pasqal have inaugurated Saudi Arabia’s first quantum computer, a neutral‑atom processor with 200 qubits housed in Aramco’s Dhahran data center. The launch also introduces the Middle East’s first commercial Quantum Computing as a Service (QCaaS)...
KOSPI Recovers +0.31% From Intraday Sidecar Low: Foreign Capital Exits Semiconductor Majors Into Power Infrastructure and HBM Front-End
The KOSPI rebounded 0.31% after a 4.5% intraday dip, as foreign investors dumped roughly $1.7 billion of Samsung Electronics and SK Hynix shares and redeployed capital into power‑infrastructure and HBM front‑end equipment. JUSUNG Engineering rallied nearly 30% on a $29 million foreign...
More Q1 '26 Foundry/OSAT Earnings
The first‑quarter 2026 earnings for major foundry and OSAT players show mixed performance across the semiconductor supply chain. SMIC posted $2.505 billion in sales with $197 million profit and 93% fab utilization, while Hua Hong reported $660.9 million revenue, $20.9 million profit and near‑full...
CEO Interview with Adi Gelvan of Speedata
Speedata, led by veteran CEO Adi Gelvan, has launched the world’s first Analytics Processing Unit (APU), a silicon‑engineered chip that accelerates SQL‑heavy analytics and AI data‑preparation workloads. The APU promises up to 100× speed gains versus CPUs and GPUs and...