Very few people understand this.
It won’t take long for companies building frontier AI models to start designing their own chips. Within the next five years, the idea of a purely “fabless” company may largely disappear.
Instead, there will be AI companies capable of designing virtually any chip they need, and foundries that manufacture those designs.
AI will likely take a bit longer to master the physical world, which gives semiconductor fabs some breathing room. At least in the near term, manufacturing will remain the harder moat.
$INTC $AMD $TSM $NVDA $SPCX
@austinsemis@AMD@BBCWorld $AMD has publicly committed 10 bln to its suppliers in Taiwan. This is only the start of what will become a more capex intensive model as the downstream supply chain passes the capital intensity back up to the fabless design houses.
The fabless model is breaking.
@hallilua@TheFederalist65@luo_yuehan I’m pretty sure her family has contributed far more to the U.S.—through taxes and other service—than you ever could.
@commonsenseplay The market has heard enough BS numbers Elon pulls out of his ass already. Remember all the promises he made to Tesla investors? SpaceX is the new cash cow.
$SPCX
@imnotharsh Some unique cards $INTC got, some $AMD
But intel can potentially do everything AMD does in future, that’s possible. Other way around is not possible.
I’m 44 years old and used to work at JPMorgan Chase. My monthly income is $110,000.
My July advice 27th:
$SNDK (SanDisk)— Don’t buy
$LITE (Lumentum) — Don’t buy
$MU (Micron Technologys) — Don’t buy
$INTC (Intel) — Buy at $85–$89
$TSLA (Tesla) — Buy at $300–$308
$NVDA (Nvidia) — Buy at $190–$196
$MSFT (Microsoft) — Buy at $380–$387
People ask, Why don’t you charge?
I’ve made enough. Sharing is my passion ,that’s why I post for fre.
Same day: Micron -6%, SK Hynix -9%, SanDisk -12%. Why? Reports of China producing domestic DUV lithography machines just as $CXMT flexes its scale.
$MU
The real story isn’t today’s market share (~8-10% of global DRAM). It’s the trajectory: monthly wafer capacity nearly doubling by 2028.
But here’s the catch bulls are ignoring: CXMT’s revenue surge is pricing-driven, not volume-driven. Capacity shipped grew just 11% QoQ while ASPs rose 57%.
And in HBM — the actual AI money-maker — Samsung/SK Hynix/Micron still control 99%+ of supply. CXMT’s HBM3E target isn’t until 2027. DUV can’t touch leading-edge AI memory.
Commodity DRAM gap = closing fast.
AI/HBM moat = still intact.