SuriAlpha @surchang - Twitter Profile

surchang retweeted

29 days ago

If you are taking investment advice on memory stocks from Bloomberg, you deserve to lose money (Save this). Bloomberg Intelligence just published that the memory shortage peaked in Q2 2026, eases rapidly into 2H26/1H27 and the industry could go into oversupply by 2028. Now here is why Bloomberg's oversupply thesis falls apart when you look at the actual fab buildout timeline. This chart shows every major new fab that Micron, Samsung, and SK Hynix have planned and not a single one produces meaningful wafer output before late 2027 at the earliest. Most are 2028–2029, New York is 2030, SK Hynix's flagship Yongin fab doesn't even start cleanroom construction until 2027 and that is construction, not output. Bloomberg sees capital expenditure going up and assumes supply is coming and that is completely wrong. JPMorgan ran the actual numbers, capex in the memory industry is surging 74% to $144 billion in 2027 but due to HBM chip area loss, increased process steps and equipment lead times stretching 12+ months, actual supply growth will be only around 13%. You cannot spend your way to fast supply in this industry because the physics don't allow it. By the end of 2027, analysts estimate supply will meet only about 60% of HBM demand and the industry would need to grow output 12% annually to close the gap, Counterpoint projects actual growth closer to 7.5%. And here is the thing Bloomberg's model completely misses, demand is not standing still while supply is being built. Token consumption is projected to grow 40x by 2030. Every new Nvidia GPU generation requires more HBM, HBM5 is already being unveiled while HBM3E is still in shortage. Each generation stacks more dies, consumes more wafer area, and deepens the supply hole. The 2028 oversupply thesis requires AI demand to plateau right when every hyperscaler on earth is in the middle of the biggest infrastructure buildout in history. Google, Amazon, Microsoft, and Meta have collectively committed hundreds of billions in capex, almost all of it memory-hungry. Long Micron and our Milk Road Pro subscribers are up massively on Micron, come join us just for a dollar to see rest of our trades using the link below!

MilkRoadAI's tweet photo. If you are taking investment advice on memory stocks from Bloomberg, you deserve to lose money (Save this).

Bloomberg Intelligence just published that the memory shortage peaked in Q2 2026, eases rapidly into 2H26/1H27 and the industry could go into oversupply by 2028.

Now here is why Bloomberg's oversupply thesis falls apart when you look at the actual fab buildout timeline.

This chart shows every major new fab that Micron, Samsung, and SK Hynix have planned and not a single one produces meaningful wafer output before late 2027 at the earliest.

Most are 2028–2029, New York is 2030, SK Hynix's flagship Yongin fab doesn't even start cleanroom construction until 2027 and that is construction, not output.

Bloomberg sees capital expenditure going up and assumes supply is coming and that is completely wrong.

JPMorgan ran the actual numbers, capex in the memory industry is surging 74% to $144 billion in 2027 but due to HBM chip area loss, increased process steps and equipment lead times stretching 12+ months, actual supply growth will be only around 13%.

You cannot spend your way to fast supply in this industry because the physics don't allow it.

By the end of 2027, analysts estimate supply will meet only about 60% of HBM demand and the industry would need to grow output 12% annually to close the gap, Counterpoint projects actual growth closer to 7.5%.

And here is the thing Bloomberg's model completely misses, demand is not standing still while supply is being built.

Token consumption is projected to grow 40x by 2030.

Every new Nvidia GPU generation requires more HBM, HBM5 is already being unveiled while HBM3E is still in shortage.

Each generation stacks more dies, consumes more wafer area, and deepens the supply hole.

The 2028 oversupply thesis requires AI demand to plateau right when every hyperscaler on earth is in the middle of the biggest infrastructure buildout in history. Google, Amazon, Microsoft, and Meta have collectively committed hundreds of billions in capex, almost all of it memory-hungry.

Long Micron and our Milk Road Pro subscribers are up massively on Micron, come join us just for a dollar to see rest of our trades using the link below!

12

184

43

131

37K

SuriAlpha

@surchang

Who to follow

Last Seen Users on Sotwe

Trends for you

Most Popular Users