AMD and Micron have both ripped on AI data-center demand, but the video argues MU is the better buy now because it sells memory to every winner while AMD is priced for perfection with >45x forward earnings and up to 16% dilution from OpenAI and Meta warrants. The AI buildout is shifting from GPU-heavy racks toward more CPUs and far more DRAM/HBM as agents increase CPU tool calls, database lookups, and KV-cache storage. MU’s business looks tighter for longer — contracts, price floors, sold-out HBM, and supply additions that lag until mid-2027 to late-2028 — while AMD’s upside is offset by valuation and dilution risk. The final call is to wait on AMD until a pullback or a major earnings beat, and to keep buying MU until DRAM contract prices flatten or fall.
$10,000 in AMD one year ago would be over $35,000 today; $10,000 in MU would be almost $60,000.
Both AMD and MU surged because AI data centers can’t get enough chips; MU is cited as a 10-bagger in 2 years.
AI agents are expected to shrink the GPU-to-CPU ratio from 8:4 to 2:1 as tool calls, code execution, and database lookups move onto CPUs.
AMD’s Lisa Su said server CPU supply has been tight since the start of the year because demand was not forecast in advance; Intel’s CEO said Intel can serve only about half its customers.
One week after Intel’s comment, AMD rose almost 10% in a day and hit a $1 trillion market cap for the first time.
Micron says memory stays tight beyond 2027; Samsung says the shortage lasts through 2028 with no sign of supply catching demand.
MU says it can meet only about half to two-thirds of demand from major customers; DRAM prices jumped more than 60% this spring, then less than 20% over the summer.
AMD has 46% of the x86 server market by revenue today, up steadily since before the AI era.
In NVDA Vera Rubin racks, 36 superchips include CPUs linked to two GPUs each, with 288 GB of HBM per GPU and up to 1.5 TB of DRAM per CPU.
The rack totals almost 21 TB of GPU memory plus up to 54 TB of CPU memory, showing why AI agents increase both CPU and memory demand.
Market US says global AI will almost 19x over the next 9 years, implying 38.5% CAGR through 2034.
MU’s data-center segment brought in $34 billion last quarter, described as almost 3x AMD and Intel’s data-center revenues combined and counting all of AMD’s GPUs.
Both AMD and MU are around $1 trillion companies, but Wall Street expects both to roughly double earnings next year; AMD trades above 45x forward earnings while MU trades below 7x.
AMD’s latest quarter posted $11.5 billion revenue, up 50% y/y; data-center revenue more than doubled on Instinct GPU ramp, and EPYC server CPU sales rose more than 70%.
AMD guided to about $13 billion next quarter and Lisa Su expects server revenue to grow more than 80% in 2H this year and more than 70% in 2027.
Two AMD deals give OpenAI and Meta warrants for up to 160 million shares each at $0.01; if exercised, they could own up to 16% combined and dilute holders by around 16% once AMD stock hits $600, a level it is already above.
MU reported 54.2 billion in revenue, about 8% above its own forecast and almost 5x a year ago, with 87% gross margins and guidance for 61.5 billion next quarter.
MU signed 26 long-term deals covering more than a third of revenue through 2030, backed by $32 billion in customer commitments, and most HBM is sold out through 2027 at much higher prices.
MU spent $27 billion on fabs and equipment last fiscal year and plans about $25 billion more over the next 6 months, while new Idaho and Japan fabs don’t start until mid-2027 to late-2028.
Samsung’s HBM share jumped from 21% to 33% in a quarter while MU’s fell from 21% to 18%; MU still says margins should stay meaningfully above prior-cycle peaks.
The buy thesis is MU, but only until DRAM contract prices flatten or fall; AMD is a long-run winner but only after a big pullback or after dilution hits and the stock drops, or after AMD crushes earnings enough to justify the multiple.