Semiconductor foundry company headquartered in Taiwan. Ultrathink has published 8 stories on TSMC. Latest, on : Microsoft Maia 300: 40% More Tokens per Megawatt, Ships 2027.
Microsoft's Maia 300 inference chip targets 40% more tokens per megawatt than leading GPUs and 7x Maia 200 rack throughput. 96 chips per 167 kW rack; production in 2027, scaling to multi-gigawatt.
An analytical deep dive into the Google Pixel 11 Gemini AI architecture, detailing Tensor G6 performance, latency drops, and the battle with Apple.
Samsung next-generation memory architectures use advanced wafer-bonding to solve critical thermal and density bottlenecks in modern AI data centers.
Samsung's new zHBM AI memory technology uses advanced wafer bonding to bypass the data center memory wall and outpace traditional HBM4 architectures.
The new Nvidia Vera CPU uses custom Olympus cores to accelerate agentic AI workloads, locking cloud providers into Nvidia's complete data center stack.
GlobalFoundries silicon photonics receives $300 million from the US to break AI hardware bottlenecks and secure domestic optical interconnect manufacturing.
The deepening AI chip stock sell-off reflects growing Wall Street anxiety over the massive capital expenditure required for generative AI infrastructure.
OpenAI infrastructure spending is projected to hit $750 billion by 2030, a macroeconomic shift that will reshape global energy grids and sovereign alliances.