The AI chip boom

The world will need more AI computing power.

By Alex Chen

What’s inside

• NVDA NVIDIA: AI accelerators and computing platforms; Huang says demand runs past $1 trillion through 2027
• AVGO Broadcom: Custom chips and data-center networking for hyperscalers building their own compute
• AMD AMD: Compute accelerators and processors; the second source buyers turn to when capacity is short
• MU Micron: Memory used alongside AI processors; the inference surge is tightening HBM supply
• TSM TSMC: Advanced chip manufacturing; the single chokepoint for every AI accelerator
• ASML ASML: Lithography equipment for chip production; no new leading-edge capacity without it
• AMAT Applied Materials: Chip manufacturing equipment bought whenever fabs add AI capacity
• KLAC KLA: Chip process control and inspection as advanced-node output scales up
• VRT Vertiv: Cooling and power kit for every new AI hall; a direct read on the pace of the buildout
• CEG Constellation Energy: Firm generation sold to the electricity-intensive AI demand the grid cannot yet serve
• CRWV CoreWeave: Sells GPU capacity to buyers who cannot wait; a pure read on compute scarcity

Why it could work

The world will need more AI computing power. The basket combines direct participants and broader suppliers.

What could break it

Adoption could disappoint, competition could weaken margins, or high valuations could already price in the trend. Some holdings are broad, indirect proxies.

For learning only. Not investment advice.