\n\n
Key developments: Chinese tech firms have purchased over one million Nvidia gaming and workstation GPUs this year, turning to chips not originally intended for AI workloads. At the same time, the cost to rent H100‑class AI compute has jumped 60% over the past twelve months, reflecting soaring demand across the globe. Industry insiders estimate that while domestic production may reach 3.5 million AI‑grade chips (including about one million made abroad), total demand exceeds five million, creating a pronounced supply gap.
Why it matters: The influx of consumer‑grade Nvidia silicon into Chinese data centers underscores how critical GPU capacity has become for AI development, even prompting nations to relax import restrictions. The sharp rent‑price increase signals tighter margins for cloud providers and higher operating costs for companies building large‑scale models, potentially slowing AI‑driven innovation or shifting investment toward alternative hardware solutions.
What to watch next: • Whether Nvidia reallocates dies from consumer GPUs (e.g., RTX 5090) to data‑center and professional lines, further tightening consumer supply. • Responses from rival makers such as AMD and Intel, including any new product launches aimed at alleviating the gap. • Government policies in China and the U.S. that could affect cross‑border chip flows, especially as Beijing seeks to build a domestic AI supply chain. [[1]]