← Home
AI × Semi Thesis

Why are chip stocks ripping — and which to buy?

The AI race forced Big Tech to open its wallet. That money flows downstream to the semiconductor supply chain. Follow the capital and you'll see who actually makes the money.

DATA-BACKEDLIVE

Chip News Signals

How chip news actually moves the stocks — measured in α, not raw %.

See all signals
Recent signals
May 4, '26
SKHX
SKHXα +7.40%
Mar 17, '26
MU
MUα +2.87%
Jan 29, '26
SMSN
SMSNα -2.02%
PART 1

How the money flows

01 · The Spark
100M+
ChatGPT users in just 2 months

ChatGPT's late-2022 launch dragged every Big Tech firm into the AI race. The fear of being left behind opened every wallet at once.

02 · The Surge
$1T+
Combined annual AI infrastructure capex
MSFT
MSFT
~$80B
META
META
~$60B
AMZN
AMZN
~$75B
GOOGL
GOOGL
~$75B
Annual capex guidance

What MSFT, META, AMZN, and GOOGL are pouring into data centers, GPUs, and power. $20–30B per quarter each — over $1T per year combined.

03 · Where it lands

Most of it lands in semis

Semis
~50%
GPU · HBM · Foundry · Equipment
Power · Cooling · Real estate
~25%
Network · Storage
~15%
Other (labor · SW)
~10%

Building an AI data center is, fundamentally, about buying chips. About half of Big Tech capex flows directly to GPUs, memory, foundry, and equipment. The rest — power, cooling, network — is the adjacent infrastructure that exists to support those chips. The fiercer the AI race, the thicker this flow.

04 · The Real Winners

Who actually takes that slice

GPU
NVDA
NVDA#1
AMD
AMD
HBM
SKHX
SKHX#1
SMSN
SMSN
MU
MU
Foundry
TSM
TSM#1
Equipment
ASML
ASML#1
AMAT
AMAT

In each sub-category, the #1 takes the largest share: NVDA in GPUs, SK Hynix in HBM, TSMC in foundry, ASML in equipment. These category leaders are the core beneficiaries. Section 3 below has the full list per category with live prices and one-tap trade buttons.

PART 2

Semis at a glance

Key elements at a glance (tap to jump)
ComputeMemoryManufacturing

GPU vs CPU

CPU
Few cores · serial
GPU
Thousands of cores · parallel

A CPU is a small number of smart cores. A GPU is thousands of simple cores in parallel. AI repeats the same math billions of times, so GPUs crush CPUs. That's why AI = surging GPU demand.

HBM — AI-grade memory

Regular DRAM
slow
GPU
HBM stack
wide pipe
GPU

Regular DRAM can't feed GPUs fast enough. HBM stacks memory chips vertically next to the GPU to widen the pipe — purpose-built for AI. It costs 5–10× more than DRAM, so HBM leadership drives memory-company earnings.

DRAM vs NAND

DRAM
Working memory
Power off → wipes
Direct AI winner
NAND
Storage memory
Power off → keeps
Supply gap winner

DRAM is short-term working memory (wipes on power-off); NAND is long-term storage (persists). AI demand pushes DRAM — especially HBM — straight up. NAND wins from a different angle: every memory maker is shifting capex into HBM, starving NAND supply. NAND pure-plays (SNDK, Kioxia) ride the supply gap upward as the field empties out.

Foundry — design and manufacture are split

Design
NVDA
NVDA
AMD
AMD
Manufacture
TSM
TSM
NVDA/AMD design the chips; TSMC actually makes them

Chip design (NVDA, AMD, AAPL) and chip manufacturing (TSM, Samsung Foundry) are separated. At 3nm/2nm leading-edge nodes, TSMC is effectively a monopoly. Even NVDA can't ship chips without TSMC.

Equipment — why ASML is a monopoly

Every leading foundry
TSM
TSM
SMSN
SMSN
+
Intel
ASML monopoly
ASML
ASMLEUV lithography
~$300M each · dozens shipped per year

Making leading-edge chips requires EUV lithography machines, and ASML is the only company on earth that builds them. Each costs hundreds of millions; only dozens ship per year. Smaller, faster AI chips mean more ASML machines, period.

PART 3

Key names by category

NAND (supply gap)

Ranked by market share

Everyone shifted capex to HBM — NAND pure-plays mop up share

Foundry

Ranked by market share

The undisputed king of chip manufacturing