CPUs, GPUs, fabrication, and the silicon supply chain.
42 storiesLatest: Oct 31, 2026
Intel beat its own Q3 guidance as data center revenue rose 78%, then guided Q4 about 9% below consensus. Its shares fell 28.58% in one of their worst sessions.
TPW Editorial8 min read
Samsung's Q3 operating profit hit KRW 107.4 trillion, the first above KRW 100 trillion for a tech company, as memory earned 99% and devices lost money.
SK Hynix posted record third-quarter 2026 revenue of 104.62 trillion won and an 80% operating margin, with HBM4 set to outsell HBM3E in the fourth quarter.
Nvidia's DSX POD packages a Vera Rubin AI factory into partner-built containers that install in 14 days. What ships inside, and what the number leaves out.
Google open-sourced TorchTPU and Amazon opened TorchNeuron's beta, both putting PyTorch on their own chips. Porting costs fall; benchmarks remain vendor claims.
TSMC's Q3 2026 net income hit a record NT$763 billion on 15 October, up 68.7% year on year, as AI demand beat its own guidance and chip stocks fell anyway.
Taiwan plans to add AI servers and cutting-edge chips to its export-control list, making unauthorised shipments to China a crime under the Foreign Trade Act.
Nvidia's share of China's AI-accelerator market fell to 34 percent in the first half of 2026 as Huawei climbed to 28 percent, per IDC data reported by Reuters.
Microsoft unveiled the Maia 300, its third in-house AI accelerator, on 12 October 2026, with multi-gigawatt inference deployments planned from 2027 onward.
SpaceX is in early talks to raise about $40 billion to buy Nvidia AI chips, with reports describing $10 billion of bank loans and $30 billion of bonds.
Tencent has reportedly leased 100,000 AI chips in Oracle data centers in South-East Asia for five years and about $7 billion, a deal neither firm confirms.
Amazon is seeking outside investors for a vehicle holding about $8 billion of Nvidia chips, the Financial Times reported, as AI compute moves onto debt markets.
Micron reported record fiscal Q4 2026 revenue of $54.23 billion, up nearly fivefold from a year earlier, as AI demand for DRAM and HBM lifted margins.
DeepSeek open-sourced TileLang and a set of Ascend kernel libraries, a CUDA-style software stack for Huawei's AI chips that lowers the cost of leaving Nvidia.
AMD agreed to acquire World Labs, Fei-Fei Li's spatial-intelligence startup, for $8.2 billion in stock, naming her executive vice president and chief scientist.
A Piper Sandler research note says Amazon, Apple, AMD, Google, Tesla, Microsoft, Nvidia and Qualcomm are evaluating Intel's 14A process, with no commitments.
Qualcomm announced the Snapdragon 8 Elite Gen 6 and 8 Elite Extreme Gen 6 on September 22, 2026: two 2nm chips that run 30B-parameter AI models on a phone.
Alibaba's T-Head unit unveiled the Zhenwu V900 AI accelerator at Apsara 2026 and set a target of more than 20GW of global data-center capacity by 2032.
CXMT said on 20 September 2026 that its fifth-generation DRAM platform is in mass production, using quadruple patterning to reach an 11.95nm array half-pitch without EUV.
Meta will deploy its MTIA 450 inference chip, designed with Broadcom and built by TSMC, in data centers in the first half of 2027, as it scales custom silicon.
Intel CEO Lip-Bu Tan says the chipmaker fills only about half of CPU demand as AI inference strains its fabs, and the 14A production timeline is in dispute.
Samsung has begun 2nm trial runs for Tesla's next-generation AI5 processor at its Taylor, Texas fab, pulling production forward as TSMC capacity stays tight.
Winbond will pay $1.12 billion in cash for Infineon's NOR Flash and F-RAM business, reviving the Spansion name as a standalone US-headquartered memory maker.
Huawei pulled the Ascend 960DT to Q1 2027, showed the industry-first NPO Atlas 960E SuperPoD at Connect 2026 and set a yearly doubling cadence in Shanghai.
Nvidia's Vera Rubin NVL72 made its MLPerf Inference v6.1 preview debut, claiming up to 3.7x the throughput of GB300 as AMD and Intel also posted results.
Ambarella launched X7, its first standalone AI accelerator: a 2-5W CVflow co-processor bringing physical AI to existing Arm and x86 systems, sampling now.
SK Hynix is in exploratory talks with Intel about producing memory chips on US soil for the first time, with Intel's delayed Ohio complex as one option.
Apple's first 2nm iPhone chip tops Geekbench 7's single-core chart and beats desktop CPUs there, but it still trails far behind them on multi-core workloads.
Samsung has opened an advanced semiconductor packaging research hub in Yokohama, betting that Japanese materials and toolmakers hold the key to its AI memory ambitions.
TSMC is raising advanced-node capacity sharply, targeting a roughly 70% increase in 3nm wafer output over 18 months as AI chip demand outstrips supply.
Fujitsu says its made-in-Japan 2nm MONAKA Arm CPU and new server go on sale in November, claiming twice the AI inference throughput of rival processors.
Nvidia's CUDA Toolkit 13.4 reaches general availability with native Windows on Arm development, a preview of the Rubin GPU architecture, and a rebuilt MPS.
Intel says it has processed more than a million High-NA EUV wafers, roughly three-quarters of the world total, as TSMC and Samsung sign on to the mask standard it championed.
Amazon gets warrants worth about $4 billion as Qualcomm buys its way into AWS AI infrastructure, with a vesting ceiling of $60 billion in purchases.
Huawei has lifted its Ascend 950DT above 250,000 yuan and Cambricon is repricing its next chip, as a high-bandwidth memory crunch reshapes China's AI buildout.
TSMC's August 2026 revenue hit a record NT$514.8 billion, up 53.3% year on year, as AI demand keeps outrunning its expanding fab and packaging capacity.
Broadcom's custom AI chip business tripled to $16.7 billion last quarter as the company raised its two-year outlook, even as next-quarter guidance landed just below Wall Street's bar.
TSMC has completed development of its 1.6nm-class A16 process with Super Power Rail, targeting Q4 mass production for AI and HPC customers.
Extreme ultraviolet lithography uses 13.5 nm light from tin plasma to print the smallest chip layers. How EUV works, what it costs, and why ASML controls it.
DeepSeek repriced the frontier, Samsung hit golden yield on HBM4, Google put Gemini on-device, Databricks raised $5B and Microsoft patched 400 flaws.
Samsung's HBM4 yield has reportedly reached 80%, the industry's 'golden yield' threshold, unlocking volume supply for major Nvidia and AMD accelerators.
Smaller transistors no longer deliver big enough gains. How advanced packaging, chiplets, 2.5D/3D stacking, HBM, became the chip industry's battleground.
