How the demand tile was computed

The demand score this run is +0.04. 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
Backlog / purchase commitments (nvidia)

Nvidia is expanding a vendor-financing/backstop program — backstopping GPU purchases in exchange for revenue sharing and agreeing to rent back unused GPUs at fixed rates — with early adopters Firmus (170,000 GPUs) and Sharon AI (40,000 GB300), about 210,000 GPUs combined, following CoreWeave ($6.3B) and Lambda ($1.5B) support deals.

  • Data Center Dynamics (2026-07-02, secondary source) - link
0.142+0.093
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 +0.067 for this cycle's findings. The tile above blends this with longer-horizon signals, so the headline score (+0.04) 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