Particle.news

AI Demand Reprices Chips and Locks Up Global Compute

A scramble for scarce next‑generation GPUs is boosting memory makers and prompting multi‑billion compute contracts and continent‑scale supercomputer builds.

Overview

  • SK hynix briefly overtook Samsung in market value on Monday as surging demand for high‑bandwidth memory and data‑center storage drove a sharp re‑rating of chip stocks.
  • SpaceX and open‑weight startup Reflection AI signed a compute agreement that begins July 1, 2026 and runs through 2029 with monthly payments of about $150 million and a maximum value near $6.3 billion under a contract that allows 90‑day early termination.
  • NVIDIA announced the construction of 35 AI HPC systems across Europe that will deploy next‑generation GB200/GB300 GPUs and provide large‑scale training and inference capacity for research and industry.
  • ByteDance’s VolcEngine reports daily token use for its Doubao model at roughly 180 quadrillion and says VolcEngine has no current plan for a separate IPO as it shifts focus toward enterprise services such as the Seedance video AI business.
  • Industry leaders are debating space versus ground data centers with SoftBank’s Masayoshi Son arguing that terrestrial build‑outs and hardware costs matter most, while the compute rush is already reshaping markets and consumer goods demand such as private aviation and hiring preferences in chip regions.