The AI chip race is a supply-chain race. An AI model runs on a GPU or custom accelerator that needs HBM memory stacked beside it, a leading-edge logic chip bui…
The public arrival of large language models pushed training compute, and later inference demand, up sharply, and made the AI data centre a central capital-spending prior…
Generative-AI investment accelerated and demand for Nvidia's AI accelerators ran well past readily available supply. Cloud providers competed for allocation, and AI star…
Nvidia introduced and ramped Blackwell, spanning the GPU, HBM, NVLink networking and full rack-scale server systems. The generation made explicit that a modern AI produc…
As Blackwell ramped, the scarce resources shifted from GPU wafers to HBM supply and to TSMC's CoWoS advanced-packaging capacity. TSMC began a multi-year push to expand C…
After 2025 restrictions on Nvidia's China-market H20 and AMD's MI308, an unusual arrangement had the two companies remit 15 percent of China AI-chip sales revenue to the…
Memory makers can point a wafer at high-margin AI memory or at PC, phone and consumer DRAM, and HBM earns an estimated three to five times more revenue per wafer than st…
SK Hynix said it completed HBM4 development and finished mass-production preparations; it and Samsung delivered final qualification samples to Nvidia, and Micron reporte…
On 26 August 2026 Nvidia reported quarterly revenue of $96.2 billion, up 106 percent year over year, with Data Center revenue of $89.0 billion, up 117 percent. It guided…
Nvidia's roadmap places Rubin Ultra in the second half of 2027 on an annual cadence, with Feynman announced for 2028. Expect HBM4 volume across all three memory makers,…
Read the complete story, FAQs and key facts.