ByteDance is secretly training a 10 TRILLION parameter AI model — 3x bigger than Kimi K3, gunning straight for Anthropic's Mythos.

ByteDance doesn't need your money or an IPO. But their supply chain? That's where the alpha is.

Here's the real play:

$NVDA — H200/Blackwell clusters powering the whole thing
$AVGO — Custom AI chip partner + high-speed networking backbone
$TSM — Fabbing everything, capacity squeezed to the max
$MU — HBM memory supplier, bottleneck = pricing power
$SMCI — AI server assembly beast, orders surging with every GPU deployment
$VRT — Liquid cooling for monster clusters, heat = their revenue
$ANET — AI datacenter switches, GPU interconnect king
$MRVL — High-speed connectivity + custom ASIC specialist
601138 (Industrial Fulian) — Server ODM heavyweight
300308 (Zhongji Xuchuang) — 800G/1.6T optical modules flying off shelves
000977 (Inspur) — Domestic AI server lead
$AMAT — Chipmaking equipment upstream, TSMC's lifeline

Chips and memory are ALWAYS the chokepoint. Follow the iron, not the hype.