How the supply tile was computed

The supply score this run is -0.03. It lands on this five-word scale:

contractingsoftening● flatfirmaccelerating

What pulled it (weight × direction × strength / 3, from this cycle's findings)

WhatWeightStrength (1-3)Pull
Whole-chain inventory (nvidia)

Nvidia AI-server rack component costs are inflating rapidly — running about 2% to 3% per week, sitting 10% to 15% above baseline, and swinging as much as 40% in a single week — driven by memory and fast storage drives scarcity.

  • BigGo Finance (2026-07-04, secondary source) - link
0.082-0.053
Alternative supply (nvidia)

Meituan's LongCat-2.0, a 1.6-trillion-parameter MoE model, was pretrained entirely on over 50,000 domestic (non-Nvidia) AI custom AI chip superpods across more than 35 trillion tokens, demonstrating frontier-scale training on alternative hardware.

  • Meituan LongCat (2026-06-30, secondary source) - link
0.042+0.027

These pulls add up to -0.027 for this cycle's findings. The tile above blends this with longer-horizon signals, so the headline score (-0.03) can differ.

Every row expands to the finding behind it and each piece of evidence: who published it, when, and whether it is a primary source.

Back to the page