Investors name five leaders
A growing investor narrative places five companies — NVIDIA, Broadcom, Micron, TSMC, and ASML — at the center of the AI hardware supply chain, from GPUs to EUV lithography, and argues memory and Blackwell-era ramps will drive near-term growth. Traders have already pushed several semis and optical stocks up sharply as demand signals show through. (x.com) (x.com)
Five names keep coming up in the same trade because they sit on different steps of the same assembly line: NVIDIA designs the graphics processing units, Taiwan Semiconductor Manufacturing makes the chips, Micron supplies high-bandwidth memory, Broadcom sells custom accelerators and networking gear, and ASML makes the extreme ultraviolet machines used to print the most advanced chip features. (nvidia.com/)(tsmc.com/)(micron.com/)(broadcom.com/)(asml.com/) The reason investors bundle them together is simple: one artificial intelligence server is not one product. It is a stack of processors, memory, packaging, networking, and factory equipment, and a bottleneck in any one layer can slow the whole build. (tsmc.com/)(micron.com/)(broadcom.com/) NVIDIA is still the center of gravity because its data center business reached $62.3 billion in the quarter ended January 25, 2026, and total quarterly revenue hit $68.1 billion. Jensen Huang said on February 25 that Grace Blackwell systems are now driving inference, which is the step where trained models answer real user requests. Broadcom matters for a different reason: it helps cloud companies build their own artificial intelligence chips and the networks that connect them. On March 4, 2026, Broadcom said first-quarter artificial intelligence revenue was $8.4 billion, up 106% from a year earlier, and guided to $10.7 billion for the next quarter. Micron is in the group because advanced artificial intelligence chips are useless without very fast memory sitting beside them. Micron says its high-bandwidth memory products are built to feed the most data-hungry workloads, and its March 2026 newsroom updates said it had moved into high-volume production of HBM4 designed for NVIDIA Vera Rubin systems. Taiwan Semiconductor Manufacturing, usually called TSMC, is the foundry that turns chip designs into physical wafers. Its Chip on Wafer on Substrate packaging platform is built for artificial intelligence processors with high-bandwidth memory stacked next to logic chips, which is why packaging capacity has become almost as important as wafer capacity. TSMC is huge even before you get to packaging. The company says annual capacity across its manufacturing facilities exceeded 17 million 12-inch-equivalent wafers in 2025, and NVIDIA said in October 2025 that the first Blackwell wafer had been produced at TSMC’s Phoenix site as Blackwell reached volume production. ASML sits furthest upstream because its machines are used before any NVIDIA or Broadcom chip exists. On January 28, 2026, ASML said fourth-quarter net bookings were €13.2 billion, including €7.4 billion of extreme ultraviolet systems, and it ended 2025 with a €38.8 billion backlog while guiding 2026 sales to €34 billion to €39 billion. That is why memory and packaging suddenly get treated like headline items instead of boring components. If Blackwell systems ramp, every extra rack pulls on Micron memory, TSMC packaging, Broadcom switching silicon, and eventually ASML tool demand at the factories expanding advanced capacity. The market has started trading that chain as one story instead of five separate stocks. The bet is not just that artificial intelligence spending stays high, but that the money now reaches deeper into the supply chain, from the chip designer at the top all the way back to the Dutch company selling the light source that prints the chips in the first place.
