AI chip and semiconductor news
Compute decides how fast AI can move. This page tracks accelerator launches, foundry capacity at TSMC and Samsung, HBM memory supply, export controls and the data-centre build-outs that follow — with strong coverage of the Asian supply chain where most of it happens.
Latest stories
Rep. Chip Roy doesn't want to regulate AI, but says Congress should have an oversight role
In an appearance on CNBC's "Squawk Box," Rep. Chip Roy, R-Texas, said he supports congressional hearings on AI, but wasn't keen on additional regulations.
CNBC Technology
heise+ | GPUs partitionieren mit Nvidia MIG
Nvidias Multi-Instance GPU teilt physische GPU in getrennte Recheninstanzen auf. Ressourcenschonend, aber setzt das Netzwerk Sicherheitsgrenzen ebenfalls durch?
heise online – KI
Introducing Amazon SageMaker HyperPod Inference Gateway
Amazon SageMaker HyperPod Inference Gateway is a Kubernetes-native, GPU-aware routing add-on for Amazon EKS. It uses real-time GPU signals to send each inference request to the best-suited pod, cutting first-token latency by up to 82% with no changes to your model servers or client applications.
AWS Machine Learning
SoftBank adds $21 billion to AI firepower in new debt deals
The Japanese conglomerate increased its margin loan backed by shares of its chip unit Arm Holdings Plc by $5 billion to $25 billion, people familiar with the matter said on Friday. And it recently secured an additional $450 million to an existing credit line, bringing the total to $6.5 billion, according to people familiar with that deal.
Economic Times Tech
Sources: SK Hynix subsidiary Solidigm weighs building a NAND flash memory factory in the US; SK Hynix is also in talks with Intel for a separate project in Ohio (Reuters)
Reuters : Sources: SK Hynix subsidiary Solidigm weighs building a NAND flash memory factory in the US; SK Hynix is also in talks with Intel for a separate project in Ohio — Chipmaker SK Hynix's (000660.KS) subsidiary Solidigm is considering building a NAND flash memory chip factory in the US …
Techmeme
India has a once-in-a-generation semiconductor demand window: Infosys SVP
Speaking at SEMICON India 2026, Meghal said the traditional semiconductor pipeline is breaking amid massive shifts driven by artificial intelligence (AI), geopolitical supply chain challenges, and a transition from Moore's Law towards system-level scaling.
Economic Times Tech
Sources: Chinese DRAM leader CXMT is preparing to expand into flash memory, competing with Samsung and domestic rival YMTC amid an AI-driven memory shortage (Reuters)
Reuters : Sources: Chinese DRAM leader CXMT is preparing to expand into flash memory, competing with Samsung and domestic rival YMTC amid an AI-driven memory shortage — Chinese chipmaker CXMT (688825.SS) is preparing to enter the booming flash memory chip market dominated by Samsung Electronics …
Techmeme
Analysis: China's 15th five-year plan spotlights diamond for AI chip thermal management
China's Ministry of Industry and Information Technology (MIIT) and National Development and Reform Commission (NDRC) have highlighted ultra-wide bandgap (UWBG) semiconductor materials, including gallium oxide (Ga₂O₃) and synthetic diamond, in their draft development plan for the electronic information manufacturing sector.
DIGITIMES
AI 晶片封裝突破光罩極限，CoWoS-L 穩居先進封裝主流至 2028 年
根據 TrendForce 最新半導體先進封裝產業研究，隨著 GPU 業者、超大型雲端服務供應商（CSP）持續 […]
TechNews (TW)
Chip foundries better insulated in an AI slowdown than Asia-Pacific tech peers, S&P says
As market fears mount over waning Big Tech spending and calls to slow down frontier artificial intelligence development, Asia-Pacific semiconductor foundries are better positioned to withstand a potential downturn in AI investment than other tech hardware firms in the region, according to S&P Global Ratings. In a report published on Thursday, the rating agency said it had stress tested four key Asia-Pacific sectors – foundries, memory manufacturers, cooling component suppliers and original...
SCMP Tech
China's AI data center token costs hinge on access to advanced GPUs
China's AI data centers are confronting a severe chip access bottleneck, with the cost of producing tokens shaped less by model design than by geopolitics, according to a recent Caijing analysis. The report said restricted access to high-end GPUs has become the most critical cost burden in China's AI era.
DIGITIMES
Nvidia CEO Pushes Back on Broad AI Rules, Says Regulators Should Target Products, Not Technology
Huang said AI is not a consumer product like social media but, a technology that powers many different applications.
AI Chips & Semis (Google News)
China's memory push gains ground across LPDDR6, advanced NAND, with HBM still the test
AI is reshaping the global memory industry, tightening supply while creating an opening for China's domestic suppliers. With Samsung Electronics, SK Hynix, and Micron prioritizing high-bandwidth memory (HBM) and high-end server memory, ChangXin Memory Technologies (CXMT) and Yangtze Memory Technologies (YMTC) are gaining ground in conventional DRAM, mobile memory, and NAND Flash.
DIGITIMES
Korea & Taiwan AI Markets (Google News)
AI Global WireHow Your AI Questions Are Answered: Inside Modern Data Centers
Asking AI a question triggers neural networks with trillions of parameters. Read how specialized chips process your data and deliver instant answers.
Korea & Taiwan AI Markets (Google News)
China to see major shift to Huawei for AI model training in 2027: rotating chair
Huawei Technologies expects a major domestic shift towards its artificial intelligence computing infrastructure for model training next year, as its Ascend processors gain ground on Nvidia in China despite supply constraints, according to rotating chairman Eric Xu Zhijun. “I believe that starting from next year, a lot of the AI model training will be based on the SuperPoD or SuperCluster based on Ascend 950DT,” Xu told reporters during the Huawei Connect 2026 conference in Shanghai on...
SCMP Tech
High-end GPUs may be slowing China's AI compute expansion
China's wide gap in token costs stems largely from domestic companies' difficulty in accessing Nvidia high-end GPUs, forcing them to squeeze more value from each token to recoup costs. But supply-chain sources say the scale of China and US AI compute investment may actually be roughly comparable, despite US hyperscalers' far larger 2025 capex.
DIGITIMES
Chip equipment makers face waits of up to 40 months for critical parts
The semiconductor industry's capacity crunch is moving deeper into the equipment supply chain, with shortages of specialized components stretching delivery times as chipmakers accelerate fab investment.
DIGITIMES
Samsung deepens Euclyd ties as startup reportedly reaches test-chip stage
Samsung Electronics is deepening its relationship with Dutch chip startup Euclyd after the company reportedly reached the test-chip stage at Samsung Foundry, adding a manufacturing dimension to Samsung's recent investment in the two-year-old company.
DIGITIMES
King Charles tests AI's safety consensus as Nvidia keeps its foot on the accelerator
King Charles III has stepped into the intensifying debate over artificial intelligence, urging executives from Nvidia, Google DeepMind, OpenAI, and Anthropic to keep AI under human control and ensure the technology serves people and the planet.
DIGITIMES
TechNews (TW)
AI Global WireAI 基建燒錢 CoreWeave 發行可轉債籌 30 億美元
AI 基礎建設瘋狂燒錢，背後有重量級靠山輝達（NVIDIA）的 AI 雲端運算商 CoreWeave 17 日 […]
TechNews (TW)
AI Chips & Semis (Google News)
AI Global WireNvidia and Google form AI energy management alliance — here's why it matters
Nvidia and Google want to make AI data centers easier on the power grid. But the companies building them will need a reason to go along with the plan.
AI Chips & Semis (Google News)
AI Chips & Semis (Google News)
AI Global WireNvidia vs. Broadcom: Whose AI revenue actually grows faster from here?
Broadcom has higher AI revenue growth projections, but Nvidia's larger and more diversified business could make its growth more durable.
AI Chips & Semis (Google News)
Mobile SoC battle for 2026 centers on NPUs
More technical details on 2026 mobile systems on a chip (SoCs) from major vendors are emerging, and the biggest changes in chips such as Apple's A20 Pro and MediaTek's Dimensity 9600 Pro are clearly concentrated in the neural processing unit (NPU). Qualcomm also published a technical article on NPU development on September 10, saying its next-generation NPU will expand shared memory capacity to improve access efficiency and speed for model data, helping mobile AI maintain fast response times.
DIGITIMES
TechNews (TW)
AI Global Wire假道東南亞、暗藏空殼公司，美報告揭輝達 AI 晶片輸中黑幕
美國非營利監控機構 C4ADS 於 17 日發布最新報告，揭露儘管美國嚴管輝達（Nvidia）高階 AI 晶片 […]
TechNews (TW)
TechNews (TW)
AI Global WireHBM 不再是唯一解方！AI 推理重組記憶體階層
隨著 Test-time Scaling 展開，AI 產業重心正從訓練（Training）快速轉向推理（Inf […]
TechNews (TW)
Nikkei Asia
AI Global WireThailand must invest in front-end chip production, industry body says
Nikkei Asia
