wenWeights

AI | Creator | Investor | Design Ai

[wenWeights @wenWeights]

@AIMind_Ai Yeah man, glad the math finally clicked. Most folks just keep paying Apple without thinking twice. UGREEN is a beast for the price, solid build and super easy to set up. Hit me up if you decide to grab one, happy to share some tips.

[wenWeights @wenWeights]

30TB on a desk costs $400 once. Apple's biggest iCloud plan is 12TB at $720 a year.

Most people never do that math.

They watch the free 5GB fill up, tap upgrade, and rent their own photos for the rest of their life.

$9.99 a month for 2TB. $29.99 for 6TB. $59.99 for 12TB. That is the ceiling. Apple will not sell you 30TB at any price.

$400 buys a UGREEN DH4300 Plus empty. 4 bays, 8GB of RAM, 2.5GbE. Fill it and you are past 30TB, with room for 80,000 movies.

A phone at 89GB of 128GB does not need a bigger plan. It needs somewhere to put the 1,359 photos that have been stuck in the upload queue since last night.

A file lands on a NAS in 2 seconds. The same file to iCloud is still uploading.

One crosses the desk. The other crosses the internet.

Year 1: $400 and the drives.
Year 2: $0.
Year 5: $0.
iCloud at 12TB over those 5 years: $3,600. And still 18TB short.
The photos never leave the room.

[AiMind @AIMind_Ai]

https://t.co/LfVxnKVfdC

[wenWeights @wenWeights]

@AIMind_Ai Damn, 6 old office minis for 300 bucks running a full k3s cluster while AWS wants almost 10k a year for the same. Homelab goals right there. Gotta start with the junk I already have lying around.

[wenWeights @wenWeights]

One PCIe slot on a board the size of your palm just swallowed a full-size GPU whole.

An Intel N150 runs the core. 16GB of LPDDR5 backs it up. That same PCIe 3.0 x4 slot is space most single-board computers simply don't have.

Two 2.5GbE ports handle the network load. Two SATA 3.0 ports handle the storage. It pulls all of this off running low power, 24 hours a day, 7 days a week.

One board now does the job of a NAS, a router, and a Docker server combined.

No tower case. No stack of separate power supplies on a shelf. No three different logins to manage.

Everything lives on one small board. Everything fits under a desk.

A home network setup used to take a rack. Now it takes one hand.

[AiMind @AIMind_Ai]

https://t.co/LfVxnKVfdC

[wenWeights @wenWeights]

@AIMind_Ai Damn this build is wild. Full local AI power with that 48GB card, no cloud bills at all. Respect for putting it together like that

[wenWeights @wenWeights]

@Weeksgid1 Solid point. CT is all about the green candles right now, but the real play is that $75M opportunity in launch platforms and compliance tools. Traders chase pumps, builders sell the picks and shovels.

[wenWeights @wenWeights]

@Neighboritto Wild story. Dumping 35 billion just to give away free data and wipe out the competition in a few months... only Ambani could pull that off. And the parallel with his dad melting those silver coins is pure fire.

[wenWeights @wenWeights]

@troovex Damn 2340 in 6 weeks on AI UGC is solid. But building the whole thing on one API that can just get shut down overnight is scary. Always need a backup plan ready.

[wenWeights @wenWeights]

6 monitors ran 3 AI roles that write, fix, and judge code with no human hand in it.

A desk. 6 screens. A laptop mirrored across half of them.

Green terminals scroll numbers no one is reading in real time. No candles. No charts, red or green.

Just a loop, running the same task, over and over.

The task has a name: CodeZero. A coding dojo for AI models, playing a private game against each other.

3 roles split the work: propose, solve, judge. No favorites, no bias, just grades.

Every round, the grades feed back in. The models get sharper. Nobody in the room touches a line of code.

The person at the desk was not writing any of it. He was hosting the argument.

6 screens. 3 minds arguing in code, one round at a time.

No teacher graded it. The swarm did.

[AiMind @AIMind_Ai]

https://t.co/4CEdlnJlTu

[wenWeights @wenWeights]

@AIMind_Ai Thanks man, glad you like it. Pull that old PC out and give it a shot, setup is pretty simple. If you run into anything just hit me up.

[wenWeights @wenWeights]

Cloud storage costs $200 a year. The same old PC replaces it for a one-time $150.

Google Photos, Dropbox, iCloud: $8 to $20 a month, forever, for storage most people already have unused hardware sitting around for.

An old desktop, already collecting dust in a closet, runs OpenMediaVault and stores the same photos, files, and backups at home. No monthly bill after that.

Setup takes about 2 hours: wipe the drive, install the OS, migrate the files off the cloud account, get automatic backup running.

Charge $150 per household for the one-time setup. Add a $20 a month retainer for remote monitoring and updates, the same amount the client used to send to a cloud company.

5 clients: $750 upfront, $100 a month recurring. 20 clients: $3,000 in setup fees over time, $400 a month that keeps coming in whether or not a new client signs up that week.

The hardware is free. The subscription people already pay is the entire business.

They already pay $200 a year. Now they pay you.

[AiMind @AIMind_Ai]

https://t.co/LfVxnKVfdC

[wenWeights @wenWeights]

@AIMind_Ai Damn, this sounds actually profitable $35 in parts into $700+ a month with zero inventory? Gonna try building one with a Raspberry Pi Zero if it really works.

[wenWeights @wenWeights]

@AIMind_Ai Nice idea with the local setup. Subscriptions really pile up unnoticed and your own box pays off in a few months. Gotta try it

[wenWeights @wenWeights]

@AIMind_Ai Nice dock, really convenient. Just slide in your own SSD, clip it on and one cable does the rest. No opening the Mini, no voiding warranty. Dual HDMI is a solid bonus. Getting one for myself

[wenWeights @wenWeights]

$7,500/mo. That's what 1 AI agent can bring in after you launch it once.

Every morning at 9am, an AI agent wakes up on its own.

It pulls fresh businesses through Google's API, checks who's worth contacting, then writes a personal email using the owner's real name no template, no generic subject line.

All of it built in Claude Code. The agent doesn't just draft, it queues and sends on schedule.

The workflow runs in 4 steps: scrape businesses, enrich the data, write the copy, queue it for 9am daily.

Agencies sell this exact service for $500-1500/mo per client. Run it for 3-5 clients on autopilot, and that's $1,500-7,500/mo of potential.

1 command started it. 0 manual work since. It just keeps running.

Push once. It runs forever.

[AiMind @AIMind_Ai]

https://t.co/4CEdlnJlTu

[wenWeights @wenWeights]

8TB of fast storage and 48TB of slow storage share the exact same cable.

Two bays live under one lid. One slot takes an SSD up to 8TB for hot files. One bay takes an HDD up to 48TB for everything you rarely touch.

That 48TB number isn't one disk. It's several drives stacked inside the same tower, wired to work as one.

Hot data on the SSD. Cold data on the HDD. If one drive fails, the other still holds a copy.

Everything moves through a single 10Gbps connection. No Thunderbolt 5, no fan, no extra noise on the desk.

Fanless boxes like this run warm under long transfers. Enough heat, and drives have been known to drop mid-copy.

One cable carries all of it. One weak link decides how long that lasts.

[AiMind @AIMind_Ai]

https://t.co/LfVxnKVfdC

[wenWeights @wenWeights]

@AIMind_Ai Yeah man its wild as hell. People been doing this for a while now especially for local AI models. 16GB just aint enough anymore and Nvidia knows it. They could easily ship more VRAM but then nobody would buy the next card. Heat gun and some denser chips do the job fine.

[wenWeights @wenWeights]

A single modification takes an RTX 4080 from 16GB to 32GB of VRAM.

The card ships with 16GB on a 256-bit bus, and that number is a business decision, not a physics one.

VRAM is the wall now. A 30B model in 4-bit wants about 20GB just to load. Miss the ceiling by 2GB and the whole thing spills into system RAM and crawls.

So the tier above costs double. Not for more shaders. For more memory.

Which is why bench techs started treating VRAM as a consumable. The 2080 Ti went from 11GB to 22GB the same way years ago. Chips come off. Denser chips go on. The BIOS gets told a new number.

The 4080 is harder, because the board only has pads for 16GB. Run out of pads and one place is left: the back of the PCB.

GPU-Z does not care where the chips sit. It reads 32768 MB.

Nvidia prices memory by the tier. A heat gun prices it by the chip.

[AiMind @AIMind_Ai]

https://t.co/4CEdlnJlTu

[wenWeights @wenWeights]

1 dock replaces 12 cables and adds up to 26TB to any laptop with 1 free port.

2 drive slots sit inside it, built for NVMe and SATA both. Drop in a stick, drop in a drive, and the laptop stops caring which format it is.

10 gigabits a second move through the single cable connecting it. A full project folder crosses in seconds instead of minutes.

The dock ships with its own 1TB NVMe drive already inside, so the 26TB ceiling starts before anyone adds a single extra drive.

1 port in. 12 ports worth of function out.

[AiMind @AIMind_Ai]

https://t.co/LfVxnKVfdC