Decade of Compute: The Chip Layer
I believe AI computing keeps growing for a decade, and the chip designers and manufacturers behind it, along with the memory and networking that ship alongside them, will show it in their results.
What’s inside
- NVDA NVIDIA: The single clearest earnings proof of AI compute demand, quarter after quarter.
- TSM TSMC: Manufactures nearly every leading AI accelerator; AI megatrend drives its record capex.
- AVGO Broadcom: Custom AI accelerators for hyperscalers plus switching; AI revenue growing triple digits.
- AMD AMD: The second merchant accelerator source if AI compute demand outruns one supplier.
- MU Micron: HBM sold out and priced up; memory is the tightest link in the AI compute chain.
- ANET Arista Networks: Back-end AI networking scales with cluster size, not just with chip counts.
- MRVL Marvell Technology: Custom silicon and optical interconnect sold into the same AI data-center budgets.
- ALAB Astera Labs: Connectivity silicon inside racks; pure leverage to accelerator attach rates.
- ARM Arm Holdings: Royalties on CPU cores increasingly paired with accelerators in AI servers.
- QCOM Qualcomm: Design IP and inference silicon extending AI compute beyond the data center.
Why it could work
If AI infrastructure spending keeps compounding toward the trillions, the companies designing, fabricating, remembering and connecting those chips capture the revenue first.
What could break it
A single large hyperscaler cutting capex, or a memory glut after new capacity lands, could reprice the whole chain fast.
For learning only. Not investment advice.