The silicon arms race: NVIDIA’s challengers, custom ASICs, memory bottlenecks and export-control fallout.
459 articles
Google is developing a new 'Frozen' chip architecture designed to significantly enhance the operational efficiency of its Gemini AI models. This hardware innovation aims to optimize compute resources for large-scale model inference.
钛媒体 · 60d ago
VeriSilicon announced 6.413 billion RMB in new orders between April 30 and July 16, 2026. Over 90% of these orders are focused on AI computing power and data processing.
36氪 · 65d ago
Apple is actively scouting for semiconductor startups and engaging with bankers to acquire AI server chip technology. This move aims to address performance issues within their internal AI server infrastructure.
36氪 · 65d ago
Anthropic is in talks with Samsung to develop custom AI chips to secure computing sovereignty and optimize inference performance. This reflects a broader trend of large model companies moving upstream into hardware.
钛媒体 · 66d ago
This week's roundup covers DeepSeek's custom inference chips, OpenAI's IPO rumors, and significant price competition from Chinese AI models. It also highlights new model developments from MiniMax and xAI.
钛媒体 · 68d ago
Synopsys is shifting its engineering focus away from semiconductor manufacturing control software to prioritize the more profitable AI chip design market. This strategic pivot aims to capture higher margins in the AI hardware sector.
The Next Web (TNW) · 74d ago
Intel is reportedly upgrading its roadmap to include a '14A Gen2' process to compete with TSMC and Samsung's upcoming 1.4nm chip manufacturing technologies. This move emphasizes Intel's strategy to maintain competitiveness in the high-performance AI chip foundry market.
cnBeta (Full RSS) · 75d ago
The dominance of general-purpose GPUs is being challenged by specialized inference chips (ASICs) designed to optimize cost, latency, and power efficiency. Major AI firms are increasingly developing custom silicon to break free from GPU dependency and improve token economics.
虎嗅 · 77d ago
Anthropic, a leading AI research company, is reportedly shifting focus toward developing its own AI chips after 5 years of operation. This move aims to secure hardware sovereignty for their large language models.
钛媒体 · 78d ago
Samsung Electronics has reportedly secured a contract worth over 10 trillion KRW to manufacture Meta's next-generation AI chips. The partnership focuses on producing 'MTIA' accelerators using Samsung's cutting-edge 2nm process.
36氪 · 78d ago
Global supply chains are under extreme pressure due to geopolitical conflicts, trade restructuring, and new regulatory costs like CBAM. AI chip supply remains highly constrained, while China's domestic AI chip self-sufficiency is rising rapidly.
虎嗅 · 79d ago
Meituan successfully deployed a 50,000-card domestic AI chip cluster (likely Huawei Ascend) for its 1.6 trillion parameter model, LongCat-2.0. This marks a shift in the domestic supply chain from single-chip breakthroughs to full-system commercial delivery.
虎嗅 · 79d ago
Qualcomm is in advanced talks to acquire AI chip startup Modular in a deal valued at approximately $4 billion. This move signals Qualcomm's intent to strengthen its AI hardware and software stack capabilities.
Bloomberg Technology · 88d ago
Google is deepening its partnership with MediaTek to develop the 'Triggerfish' TPU v9 chip. The project focuses on advancing AI agents, reinforcement learning, and maximizing compute efficiency.
cnBeta (Full RSS) · 89d ago
A SemiAnalysis report reveals that the co-design of DeepSeek V4 and the Huawei Ascend 950DT AI accelerator has achieved a 75% reduction in inference costs. This highlights the efficiency gains possible through hardware-software co-optimization.
Pandaily · 96d ago
Xiaomi’s XRing O3 SoC has entered mass production, with MiMo running directly on-device. Its O100 AI accelerator and D100 intelligent-driving chip are targeting commercial use in the first half of 2027.
Pandaily · 10d ago
The sharp selloff in Moore Threads after its lock-up expiry shows that China’s GPU companies are being judged less by domestic-substitution narratives and more by profitability, cash flow, and dilution risk. Biren Technology plunged, while Iluvatar CoreX rose modestly, reflecting differences in earnings quality and investor confidence.
虎嗅 · 10d ago
Malaysia is leaning toward using Huawei AI chips for its sovereign AI initiative. The move could represent a significant advance for Chinese AI hardware despite opposition from the Donald Trump administration.
Bloomberg Technology · 12d ago
DeepSeek reportedly ordered 160,000 Huawei Ascend 950DT chips for a new 1GW data center in Inner Mongolia. The alleged purchase could strengthen Huawei’s position in China’s AI chip market and increase pressure on Nvidia’s China-specific offerings.
cnBeta (Full RSS) · 15d ago
DeepSeek is reportedly preparing a large order of Huawei AI chips for a new data centre in Inner Mongolia, according to Bloomberg. The chip model and order size remain undisclosed, and DeepSeek has not confirmed the plan.
The Next Web (TNW) · 15d ago
柬埔寨公司销售的“昊天AI”可在WhatsApp、Zoom、Teams等平台实时换脸，并被用于电信诈骗和绕过刷脸验证。文章还揭示了结合GPT、Gemini、批量账号和代理IP的一站式诈骗工具链。
虎嗅 · 10d ago
Hot Chips ‘26 highlighted a more diverse AI hardware market focused on reducing inference costs, power consumption, and dependence on GPUs. OpenAI’s Jalapeño inference chip reportedly delivered industry-leading results in SemiAnalysis testing, while Nvidia, Intel, and others introduced CPU and accelerator alternatives.
Computerworld · 19d ago
NVIDIA will invest $3.5 billion in MediaTek while expanding their chip collaboration through NVLink Fusion. MediaTek previously designed the GB10 system-on-chip used in NVIDIA’s RTX Spark consumer product and DGX Spark mini workstation.
cnBeta (Full RSS) · 19d ago
Sunrise, spun out of SenseTime’s chip business, reportedly raised RMB2 billion ($280 million) at a post-money valuation of about RMB20 billion. The company is developing chips focused on AI inference, and the financing would nearly double its valuation from the previous round.
TechNode · 19d ago
Nvidia expects to sell $20 billion worth of Vera Rubin systems in its third fiscal quarter as shipments begin. The company says the platform could represent 20% of data center revenue, making it the fastest-ramping AI platform in its history.
Tom's Hardware · 23d ago
Nvidia sold a small number of H200 AI chips to Chinese customers in the latest quarter. Shipments remained below the amount permitted under the US government’s approval.
Bloomberg Technology · 23d ago
Huawei has proposed building AI data centers in Egypt for military, surveillance, and other public-sector applications. The pitch highlights Huawei’s global AI-chip ambitions and has prompted concern in Washington, where officials are considering a counteroffer.
Bloomberg Technology · 24d ago
Huawei has proposed building AI data centers in Egypt for military, surveillance, and other public-sector applications. The effort highlights Huawei’s ambitions to expand its AI chip business globally and has raised concerns in Washington.
Bloomberg Technology · 24d ago
OpenAI says its new Jalapeno chips outperformed Nvidia’s current lineup in testing. The chips reportedly led in AI workload per unit of power and response speed.
Bloomberg Technology · 25d ago
OpenAI claims its new Jalapeno chip outperformed Nvidia processors in internal tests. The chip is intended to let customers choose between lower-cost inference and faster responses.
Bloomberg Technology · 25d ago