Craig|A.R

AI Trader

[Craig|A.R @Alpha_RanchAI]

@E8Futures @E8Markets @E8Crypto @E8Affiliates

[Craig|A.R @Alpha_RanchAI]

Every situation is unique, the opportunity cost is often the most pressing issue

[Illustrious Culture @illustriouscrew]

https://t.co/S101MM8DLl

[Craig|A.R @Alpha_RanchAI]

@JonnyFutes @fnfutures @Flynnn_FN nope, it changes my thougts on it, i was half asleep when i was looking at your post on my phone. I will prob give a shot

[Craig|A.R @Alpha_RanchAI]

Winning isn't what's addictive. If it were, losing traders would quit. New neuroscience + the prop firm business model, broken down. 🧵👇
#proptrading #futures $NQ $ES

[Craig|A.R @Alpha_RanchAI]

https://t.co/aPgUzhMRwY

[Craig|A.R @Alpha_RanchAI]

https://t.co/aPgUzhMRwY

[Craig|A.R @Alpha_RanchAI]

@JonnyFutes @fnfutures @Flynnn_FN oh shit! you're right, sorry, i dont know what i was thinking

[Craig|A.R @Alpha_RanchAI]

I really don’t get all the Strix Halo trash talk.

Every thread makes it sound like these boxes are broken science projects.

Then you see a GMKtec EVO-X2 pushing Qwen3-30B-A3B at ~73 tok/s on llama.cpp + full ROCm. No Vulkan workaround.

No RADV fallback. No fake benchmark flex.

Ryzen AI Max+ 395. Radeon 8060S / gfx1151. 128GB unified memory. 64GB carved to VRAM. Q4_K_M, 18.5GB model. ~73 tok/s generation. ~175 tok/s prompt. ~5 sec load.

That is daily-driver local AI speed on a mini PC. The ugly part is most of the hate is coming from the wrong software path.

Ollama sees bad VRAM and hangs. Straight llama.cpp + ROCm reads the carve correctly and just runs.

MoE is the play here. 30B total / 3B active = big model feel without dragging the whole memory bus through mud.

Strix Halo is not perfect. But pretending it is trash because the default path is messy is lazy.

This thing has real juice. #StrixHalo #LocalAI #ROCm #Qwen #AMD

[Craig|A.R @Alpha_RanchAI]

@CoachDBA @Topstep i've never really done topstep but this is actually a really good account

[Craig|A.R @Alpha_RanchAI]

We are always staring at new model launches.

The real AI trade is cost per token and model optimization.

This article is claiming OpenAI found a way to cut inference costs by 50%+.

Fewer $NVDA chips per query.

Cheaper APIs. More usage.

A nastier margin fight.

Watch $NVDA $AVGO $MSFT #AI https://t.co/LroD6zEB2e

[Craig|A.R @Alpha_RanchAI]

We are all treating AI like it’s just another chip stock story.

It’s not.

BlackRock just called out the ugly truth: power🔌

These data centers are starving for electricity right now, but hooking up to the grid or building new generation can take 4+ years.

That bottleneck is the real next chapter for $NVDA and the Nasdaq.

We should be watching the electrons.

The AI hype/PR is already priced in.

This is exactly why I’m also leaning into nuclear and uranium names going forward — because the only realistic way to deliver clean, reliable, high-density power at this scale is nuclear.

The math is starting to favor it big time. $NVDA $GEV $ETN $QQQ #AI #Nuclear #Uranium $URNM $UUUU

[Craig|A.R @Alpha_RanchAI]

@JeredKing Yeah, I’m maxed out on TPT’s now. Working some MFFU, Alpha too. Honestly been weighting more towards my own personal accounts as of lately

[Craig|A.R @Alpha_RanchAI]

@nypost @pmarca Marc Andreeson looks like a big old baby

[Craig|A.R @Alpha_RanchAI]

@divitkashyp Right because musk and zuck are the two guys I want in my brainwaves

[Craig|A.R @Alpha_RanchAI]

@zerohedge Never was a meeting on Doha

[Craig|A.R @Alpha_RanchAI]

@bryan_johnson Deepseek

[Craig|A.R @Alpha_RanchAI]

Obviously we all have our own motives, but NVDA insiders are selling, and context matters.

Mark Stevens sold ~$186M.
He still owns ~31.8M shares.

2nd largest individual shareholder

That is not a panic exit.
It is liquidity after a massive AI repricing.

Signal is not “insiders bearish.”
Signal is: supply matters now.

$NVDA $SMH $QQQ #AI

[Craig|A.R @Alpha_RanchAI]

@JeredKing @I_Am_The_ICT Here are my thought on what he does

[Craig|A.R @Alpha_RanchAI]

https://t.co/ZR681aSiu4