AI supply chain beneficiaries mapping: analyst identifies key component suppliers across custom silicon, memory, networking, thermal, and infrastructure for Google's $80B AI capex expansion
2026 AI capex ~$185B (2x 2025); memory ~60% of capex; TPU replacement thesis for Nvidia; specific roles: AVGO TPU, TSM TPU foundry, MRVL inference, MU/Samsung/SK HBM, SNDK NAND, optical (LITE R64, COHR transceivers), thermal (VRT cooling), power (ETN), fiber (GLW), datacenters (WULF, HUT, CIFR)
$GOOGL AI Supply Chain ETF: Since they're raising $80B to finance AI infrastructure expansion. Total 2026 AI Capex: ~$185B [2x 2025] Custom silicon / memory: - $AVGO: TPU - $TSM: TPU foundry + advanced packaging - $MRVL: apparently in talks for inference TPU + MPU. - Samsung: HBM - SK Hynix: HBM3E - $MU: third source HBM - $SNDK: NAND Networking & optical: - $LITE: R64 optical circuit switch, co-developed w/ Google + in production - $COHR: 400G/800G/1.6T transceiver - $CRDO: AECs linking TPU clusters - $FN: contract manufacturer for transceivers Google cloud compute: - $NVDA: GPUs - $AMD: GPUs, smaller exposure Server ODM & power: - $CLS: primary TPU board + rack-level integrator - $VRT: CDUs, liquid cooling + rack power architecture - $ETN: Switchgear, busways, UPS, electrification - $BE: fuel cells - $GLW: subsea fiber Real estate / colocation: - $WULF: hosts Google-backstopped AI capacity + Google holds warrants - $HUT: 245 MW lease w/ Google financial backstop - $CIFR: Fluidstack/Google HPC hosting Just for a top level overview of potential direct beneficiaries. There are ofc plenty of other names located in Taiwan/Korea like Wiwynn (6669) for server ODM flows. But given the ~$460B cloud backlog, a large chunk of the raise will probably go to more memory since it's ~60% of capex. Also TPUs since Google's thesis is replacing Nvidia with TPUs e.g. Broadcom.
| Horizon | Raw move | vs SPY |
|---|---|---|
| 1 day | +0.1% | -0.3% |
| 1 week | -11.4% | -7.5% |
| 1 month | -15.1% | -14.7% |
| 3 months | — | — |
| 6 months | — | — |