TileRT's persistent engine on NVIDIA GPUs achieves up to 500 tokens/s/user on the InferenceX GLM5 FP8 744B benchmark on a single B200 decode server, approximately 3× faster than GB300 NVL72 running tr…
SpaceX aims to build and deliver 6-8GW of AI datacenter capacity in 2027, potentially exceeding 10GW, which could generate $300-500B in capex and $300B in annual recurring revenue for SpaceX, accordin…
Google announced on August 5th a complete overhaul of DeepMind leadership, with co-founder Demis Hassabis stepping back from day-to-day operations and former Google Chief Scientist Jeff Dean leaving t…
Moonshot AI released Kimi K3, an open frontier model that swept leaderboards, featuring a hybrid attention mechanism with Kimi Delta Attention (KDA), a linear attention layer derived from DeltaNet and…
Modular construction has become the default playbook for building datacenters, with SemiAnalysis tracking over 61GW of modular capacity across 1,000+ sites and estimating modular penetration will reac…
AMD has a great chance of breaking Nvidia's CUDA software moat in AI accelerators, according to SemiAnalysis, which upgraded its outlook from zero to non-zero to now a strong probability of success, c…
Nvidia's Vera Rubin NVL72 delivers 5.4x performance per megawatt and 5x performance per dollar over GB200 NVL72 on DeepSeek R1 inference, according to early engineering samples from CoreWeave. The sec…
Meta Platforms Inc.'s infrastructure team requires a cultural reset due to bloated management, over-engineered solutions, and political decision-making that have led to expensive missteps, including a…
Meta's AI division, Meta Superintelligence (MSL), has made significant progress in the past year following a rebuild after the Llama 4 failure, with advantages in data, talent, and compute that positi…
Anthropic reported a profit of over $1 billion in the third quarter of 2026, driven by strong adoption of its Claude Code product in software development. The company confidentially filed for an IPO o…
Nvidia's GPU debt backstop is enabling a multi-trillion-dollar AI credit market, with over $7 trillion in debt projected by 2029 to finance AI compute buildouts. The AI Project Trinity—capital, offtak…
Intel, TSMC, SK Hynix, Samsung, Micron, Marvell, Lightmatter, and Microsoft disclosed advanced packaging technologies at ECTC 2026, including Intel's EMIB-T with 36/35 µm bump pitch and plans for 25 µ…
SemiAnalysis surveyed over 50 enterprises and found that widespread token overspending, as reported at Meta and Uber, is overstated and not representative of most organizations. Companies are now adop…
The US grid cannot add capacity fast enough to meet surging datacenter power demand from AI labs and hyperscalers, driving behind-the-meter (BTM) solutions to power over half of new US datacenters by …
Co-packaged optics (CPO) are set to become the main driver of bandwidth increases in AI scale-up networking by the latter part of this decade, as copper-based links face range and scaling limitations.…
China's CXMT, the country's leading DRAM manufacturer, is set to go public on the STAR Market in what could be the largest semiconductor IPO in China. Founded by Zhu Yiming, the company has leveraged …
A viral claim that half of 2026 US datacenter capacity is canceled is based on flawed data and uninformed forecasts, according to SemiAnalysis. The firm's own tracking shows only ~1% movement in hyper…
Anthropic CEO Dario Amodei said reinforcement learning shows the same log-linear scaling as pre-training, but RL system efficiency is critical to afford enough training. Experiments on open models sho…
DeepSeek v4, an open-source model from a Chinese lab, achieved a 100x performance improvement on AMD hardware within 26 days of its Day 0 release, driven by the AMD SGLang engineering team. The Infere…
Chinese robotics firm Unitree is on track to ship its 10,000th humanoid robot in the coming weeks, tripling revenues year-over-year on 60% gross margins while pricing its flagship G1 humanoid at $27,3…