Get live statistics and analysis of Sudo su's profile on X / Twitter

I love hardware too much. There’s no force on earth that can stop me from hoarding GPUs. Not markets, not power bills, not even common sense. Silicon is my addiction, and every new card feels like another lungful of oxygen. I’ll keep stacking them until the lights flicker in my city.

Hot take: Most "AI infrastructure" advice is from people who've never run their own hardware. They'll tell you: "Just use the API, it's cheaper" What they won't tell you: - How to debug when PCIe bandwidth becomes your bottleneck - Why your "software bug" is actually a loose cable - When thermal throttling matters more than FLOPS Real infrastructure knowledge comes from hardware that fights back.

What's stopping you from renting a space and building your own local compute lab? own the hardware, own the inference, own the learning. The barrier isn't technical anymore. It's deciding ownership > convenience.

My biggest competitor isn't another company. It's my own inconsistency. And that guy's winning.

Running GPU clusters for AI infrastructure. Some nodes sit idle between projects. So I rent them out on Vast ai during downtime. $17.72 so far, climbing daily. 3090s earn $0.2-0.3/hr depending on demand. Covers power costs, generates passive income, hardware stays productive instead of idle. The economics are simple: owned hardware that isn't running costs money. Hardware that's running makes money. Best part? You can use them for your own workloads AND rent them out when idle. Not either/or. Both. Your GPUs work for you 24/7. Training your models when you need them. Earning passive income when you don't. By the time I scale to full utilization, the side hustle will have covered a chunk of the infrastructure cost.

"Just use the API" assumes three things that aren't true: 1. Your usage stays constant (it never does) 2. Their pricing stays constant (it never does) 3. You'll switch when it gets expensive (you won't) Year 1: APIs cheaper Year 2: Break even Year 3+: Ownership wins The math isn't complicated. The switching costs are. [Can expand into thread about: specific cost comparisons, switching cost breakdown, when APIs make sense vs don't]

GPU poverty forces creativity. When you can't throw compute at the problem, you have to throw intelligence at it. Constraints breed solutions, not excuses.

Nobody tells you that building your vision means living in the gap between who you were and who you need to become. That gap is uncomfortable. But it's where everything happens.

Would you rather be underestimated, or overhyped?

Git commit messages are documentation for your 3 AM debugging self.

Fine-tuning without understanding your data is just expensive random number generation.

The age of acceleration doesn't wait for you to feel ready. At 26, I realized: the gap between 'I should build this' and 'I'm building this' is just a decision. Not talent, resources or permission. A decision. What I learned: many optimizing for the perfect moment. Choose the path where you ship messy over plan perfectly. Acceleration rewards action, not intention.

This is what happens when you take infrastructure constraints seriously and have the capital to think beyond incremental solutions. Space has effectively unlimited solar and cooling. The engineering challenge is getting the hardware there and getting the data back. Elon has both pieces: SpaceX for transport, Starlink for data transmission. Timeline is Elon standard optimistic. Strategy is sound though. When you can't scale horizontally on Earth, you scale vertically into orbit.

Trend nobody's talking about: Every major API customer eventually becomes a competitor. Cursor paying Anthropic → Cursor builds Composer OpenAI's biggest users → evaluating local LLMs My $20K/year API bills → bought $9K in GPUs API businesses work until your customer does the math on ownership. Then your best customer becomes your biggest threat.

All plans are provisional plans. The difference: some people update their plans when reality changes. Others defend their old plans until they become failures. Stay fluid or stay wrong.

My biggest competitor isn't another company. It's my own inconsistency. And that guy's winning.

I love hardware too much. There’s no force on earth that can stop me from hoarding GPUs. Not markets, not power bills, not even common sense. Silicon is my addiction, and every new card feels like another lungful of oxygen. I’ll keep stacking them until the lights flicker in my city.

Running GPU clusters for AI infrastructure. Some nodes sit idle between projects. So I rent them out on Vast ai during downtime. $17.72 so far, climbing daily. 3090s earn $0.2-0.3/hr depending on demand. Covers power costs, generates passive income, hardware stays productive instead of idle. The economics are simple: owned hardware that isn't running costs money. Hardware that's running makes money. Best part? You can use them for your own workloads AND rent them out when idle. Not either/or. Both. Your GPUs work for you 24/7. Training your models when you need them. Earning passive income when you don't. By the time I scale to full utilization, the side hustle will have covered a chunk of the infrastructure cost.

Would you rather be underestimated, or overhyped?

29 days of building in public: 61 → 232🔥. I’ll keep following back until 1,000🫱🏻‍🫲🏼.

The age of acceleration doesn't wait for you to feel ready. At 26, I realized: the gap between 'I should build this' and 'I'm building this' is just a decision. Not talent, resources or permission. A decision. What I learned: many optimizing for the perfect moment. Choose the path where you ship messy over plan perfectly. Acceleration rewards action, not intention.

Nobody tells you that building your vision means living in the gap between who you were and who you need to become. That gap is uncomfortable. But it's where everything happens.

Momentum isn't motivation. It's proof you can execute when nobody's watching. That's the real shift.

Hot take: Most "AI infrastructure" advice is from people who've never run their own hardware. They'll tell you: "Just use the API, it's cheaper" What they won't tell you: - How to debug when PCIe bandwidth becomes your bottleneck - Why your "software bug" is actually a loose cable - When thermal throttling matters more than FLOPS Real infrastructure knowledge comes from hardware that fights back.

What's stopping you from renting a space and building your own local compute lab? own the hardware, own the inference, own the learning. The barrier isn't technical anymore. It's deciding ownership > convenience.

Trend nobody's talking about: Every major API customer eventually becomes a competitor. Cursor paying Anthropic → Cursor builds Composer OpenAI's biggest users → evaluating local LLMs My $20K/year API bills → bought $9K in GPUs API businesses work until your customer does the math on ownership. Then your best customer becomes your biggest threat.

The difference between those who build and those who imagine is the ability to sit with one task. Repetition. Crushing it through consistency. The rest is noise.

Watched 5 companies migrate APIs this year: OpenAI → Anthropic (safety) → back to OpenAI (performance) → Google (cost) → back to Anthropic (quality) → evaluating local (control) Each migration: broken integrations, retrained teams, changed costs, 2week delays. This isn't strategy. It's vendor hopping. Local deployment looks expensive until you price out the 5th migration. Then it looks like stability.

Users don't care about your tech stack. They care about their problems.

Context switching is the enemy of deep work.