Infra
DeepSeek and Huawei Just Open-Sourced a Simpler Alternative to CUDA — the Language War Is the Real Chip War
DeepSeek and Huawei released open-source programming tools for Huawei's Ascend chips, with Peking University's TileLang language — which DeepSeek uses for its AGI work — as the core, plus a 128-chip Ascend 950 supernode optimization. It's China's coordinated answer to Nvidia's four-million-developer CUDA moat, and it lands the same week OpenAI's Jalapeño chip has analysts calling that moat 'potentially dead.'...
2026-10-01
Infra
Alibaba's Zhenwu V900: China's Response to Nvidia's AI Chip Dominance
Alibaba unveiled the Zhenwu V900 AI chip at Apsara Conference 2026, delivering 3x performance over predecessor, with plans for 500K-accelerator clusters and 20GW data center capacity by 2032....
2026-09-23
Infra
Huawei's Bold 2035 Bet: AI Agents Will Burn 100,000 Times More Tokens Than Today
Huawei predicts AI agents will consume 100,000 times more tokens by 2035, with 900 billion agents operating worldwide. The company unveiled a comprehensive roadmap focusing on SuperPoD computing, agent OS, and infrastructure for an agent-centered internet....
2026-09-20
Infra
Huawei's 9-Month Shortcut: How China's AI Chip Ambitions Just Accelerated
Huawei announced the Ascend 960DT will launch in Q1 2027 — nine months early with performance doubling. The accelerated timeline signals China's AI chip independence is no longer a future goal but a present reality....
2026-09-17
Infra
China's 9,800 EFLOPS Bet: How the World's Largest AI Compute Buildout Reshapes the Global Order
China's MIIT targets 9,800 EFLOPS of AI computing capacity by 2030 under the 15th Five-Year Plan, with RMB 3.8 trillion in infrastructure investment — a fourfold increase signaling AI as strategic sovereignty....
2026-09-12
Infra
Google's Air-Cooling U-Turn: The Physics Problem Behind India's 1-GW Data Centre Boom
Google agreed to air-cool its 1-GW Visakhapatnam data centre after local protests, but physicists say it's impractical. Meanwhile, TCS's HyperVault is building a liquid-cooled 1-GW campus in Hyderabad. The physics of cooling don't care about community politics....
2026-09-06
Infra
DeepSeek's $2 Billion Bet: 160,000 Huawei Chips and the End of Nvidia's China Reign
DeepSeek plans 160,000 Huawei Ascend 950DT chips for inference at a new 1GW data center in Inner Mongolia. Training stays on Nvidia, but the inference split marks China's transition from adapting to sanctions to building independent infrastructure....
2026-09-05
Infra
China's CXMT Makes First HBM3E Chips — But the Memory Gap With the West Is Still Massive
CXMT begins small-scale HBM3E production for Huawei, marking China's first domestic high-bandwidth memory for AI — but 25% yields and no EUV tools mean the technology gap with SK Hynix remains massive....
2026-09-01
Infra
OpenAI's First Custom Chip 'Jalapeño' Smokes Nvidia — The CUDA Moat Is Cracking
OpenAI's first custom chip "Jalapeño" beats Nvidia's Blackwell and Rubin in inference benchmarks, delivered 1.5x-1.9x more AI work per watt. Built with Broadcom in 16 months, the inference-only chip signals the end of CUDA monopoly and the rise of vertical integration in AI....
2026-08-27
Infra
Alibaba's T-Head Is Coming for Nvidia — With a Second-Gen Chip Ready This Year
Alibaba CEO Eddie Wu confirms second-generation T-Head chip will tape out in H2 2026. The Zhenwu M890 already serves 650+ customers with 3x performance vs predecessor. T-Head spun off as independent entity with 1B yuan capital....
2026-08-23
Infra
Apple's Quiet Gamble: Testing Chinese Memory Chips Amid U.S. Pressure and Global Shortage
Apple is testing Chinese CXMT memory chips for iPhones and MacBooks amid global supply shortages, but U.S. export controls and Senate pressure complicate any potential deal....
2026-08-11
Infra
AI's Energy Appetite Drives Nvidia and Amazon to Pour Billions Into Power Infrastructure
Nvidia invests $3B in power infrastructure Lancium while Amazon backs Texas gas plant as AI data center energy demand threatens to double by 2030, forcing tech giants to secure energy independence....
2026-08-10
Infra
Huawei's MateBook Pro S Proves China Can Build AI Laptops — Without Intel or AMD
Huawei launches MateBook Pro S — a 798g AI laptop powered by its own Kirin XE90 chip and HarmonyOS, marking a major step toward Chinese technological independence....
2026-08-08
Infra
AMD's Taalas Acquisition: Etching AI Models Directly Into Silicon to Challenge Nvidia's Inference Dominance
AMD acquires Toronto-based Taalas, a startup that etches AI model weights directly into silicon for order-of-magnitude inference performance improvements....
2026-08-07
Infra
Nvidia's Quiet Gambit: Seeking Chinese Partner for AI-Powered 6G Base Stations
Nvidia reportedly partnering with Shenzhen-based Jiaxian Communications to develop AI-powered 6G base stations for overseas markets, trial networks targeted for 2027-2028....
2026-08-07
Infra
Helsing's $1.8B Funding Round Proves Europe Can Build Defense AI — If It Wants To
Helsing raises $1.8B Series E at $18B valuation, marking Europe biggest defense AI funding round....
2026-08-06
Infra
HP, Asus, Acer Quietly Adopt Chinese Memory Chips as AI Squeeze Bites
HP, Asus, and Acer have begun adopting CXMT Chinese memory chips in laptops for non-US markets, driven by an unprecedented AI-induced DRAM shortage. CXMT prioritizes Huawei and Chinese customers....
2026-08-05
Infra
The CXMT Shock: A $489 Billion Chinese Chip IPO Just Rewrote the Memory Market
ChangXin Memory Technologies' shares surged 466% on its Shanghai debut, reaching a $489 billion valuation and shaking Nvidia, Micron, and SK Hynix as China builds a viable DRAM alternative....
2026-07-30
Infra
China's GPU IPO Frenzy Meets New York's Data Center Ban — The AI Infrastructure Crisis Nobody's Talking About
MetaX files for Hong Kong IPO amid a wave of Chinese GPU startups going public, while New York becomes the first US state to ban new hyperscale data center construction. Two sides of the same breaking point: China needs chips to bypass export controls, but the US can't power the data centers that would use them....
2026-07-24
Infra
AMD's Helios Just Declared War on Nvidia — And the AI World Should Care
AMD launches Helios, its first rack-scale AI system that beats Nvidia's Vera Rubin on paper across multiple metrics. Backed by $5B Anthropic deal, Microsoft deployment, and a customer list including OpenAI and Meta, Helios represents the most serious challenge to Nvidia's 95% datacenter GPU monopoly — powered by UALoE over Ethernet that could finally break Nvidia's hardware lock-in....
2026-07-24
