The more I look at AI, the less I think this is just a chip trade.
Power and cooling might be the real bottleneck.
GPUs get the headline.
Infrastructure decides how far the cycle can actually scale.
AI is not one trade anymore.
I’m mapping it by bottlenecks:
Compute: NVDA, AMD
Memory: MU
Networking: AVGO, MRVL, COHR
Packaging: TSM, AMKR
Power/Cooling: ETN, GEV, VRT
Physical AI: TSLA, QCOM
The edge is not chasing “AI.”
It’s finding which layer breaks first.
Traders need to separate “level” from ���signal.”
$BTC under $63k is just the headline.
The real signal is how price reacts to visible liquidity:
Hold + absorption = possible bottom.
Flush through = trap.
The level is not the trade.
The reaction is.
Without a doubt, if $BTC fills these bids, it's the macro bottom.
These types of bids show up at every major low.
Either the market front runs them, or they get filled and mark the bottom.
Same story.
Every.
Single.
Cycle.
We're close, boys. Don't get shaken out now.
“AI stock” is becoming a lazy label.
The real trade is AI capacity.
Nebius didn’t run 500% because people like buzzwords.
It ran because compute, power, backlog, and cloud contracts are becoming scarce assets.
The market is paying for bottlenecks now.
@AutismCapital $SPCX being red today doesn’t matter.
This is still one of the cleanest access trades in the market.
People are not buying rockets.
They are buying the front door to orbit.
$SPCX is not being valued like a rocket company.
It is being valued like an access layer to orbit.
Cheaper infrastructure comes first.
New demand shows up later.
The trade is access.
The funny part: cheaper AI probably doesn’t mean companies spend less on AI.
It means they shove AI into more meetings, more docs, more workflows, more nonsense.
Great for usage.
Messy for model margins.
BREAKING: Microsoft exploring DeepSeek over OpenAI and Anthropic as Copilot Cowork moves to usage-based pricing
“We have users who do hundreds of tasks a week… the consequence is the costs can go very high...”
Jevons paradox