AI training work is mostly text-based. You don't need an RTX 4090, a multi-monitor curved gaming setup, or a custom mechanical keyboard. You do need a few specific things — most of them cheap, all of them practical. Here's what actually moves the needle.
The shortlist (everything you need)
- A laptop or desktop from the last ~5 years.
- A reliable internet connection (~50 Mbps minimum).
- One external monitor (1080p is fine).
- An external keyboard if you're on a laptop.
- A decent chair (this matters more than the computer).
- Two pieces of free software.
Total cost if you don't already have any of it: ~$400–$700. Total cost if you already have a laptop: ~$200–$400. The first month of contract income usually covers the entire setup.
Computer: what actually matters
Most AI training tasks run in a browser. The platform's interface (Outlier, Mercor, Surge) handles all the heavy work — model inference, task routing, scoring. Your computer is just rendering web pages.
What you need:
- 16GB of RAM if you have multiple browser tabs open (you will).
- A CPU from the last 5 years — any Apple Silicon, any Ryzen 5 5xxx or newer, any Intel 11th-gen or newer.
- SSD storage — 256GB is plenty unless you're doing video or image annotation.
What you don't need:
- A discrete GPU. Pointless for AI training contracting. Your tasks aren't running on your hardware.
- 32GB+ of RAM. Unless you also run heavy IDEs alongside, 16GB is the sweet spot.
- The latest M-series Mac. An M1 from 2020 is still completely fine in 2026.
If you're starting from zero, the cheapest viable option in 2026: a refurbished M1 MacBook Air ($550–$700) or a refurbished ThinkPad with 16GB ($400–$550). Both will outlast 3+ years of contract work without complaint.
Monitor: one is fine, two is overkill
One external monitor at 1080p or 1440p is the cleanest setup for this work. The platforms' interfaces aren't designed for ultra-wide; they look bad stretched across two monitors.
The actual benefit of an external monitor isn't pixel count — it's posture. Looking down at a laptop for 4-hour stretches is what wrecks your neck. Even a cheap $130 1080p monitor at proper height eliminates that problem.
If you already have one monitor, don't buy a second. The second monitor mostly fills with Twitter.
Keyboard: external if you're on a laptop
Laptop keyboards are bad for typing 6+ hours a day. Any external keyboard is an upgrade. You don't need mechanical (though it's nice). Membrane keyboards in the $30–$60 range are 95% as good as keyboards costing four times that.
What matters: full-size or tenkeyless layout, decent travel, no proprietary connector. Logitech MX Keys ($85, wireless) or Keychron K2 ($70, wired/wireless) are solid defaults.
Chair: this is the one to spend on
The most common occupational hazard in this work isn't carpal tunnel or eye strain. It's lower-back pain from sitting in a bad chair for 4+ hours at a stretch. A reasonable chair pays for itself in productivity within a month.
You don't need a Herman Miller. You do need:
- Adjustable seat height (so your feet can be flat on the floor).
- Adjustable lumbar support.
- Adjustable armrests (so your shoulders aren't shrugged while typing).
- A seat depth that lets you sit with your back against the chair.
Mid-range options that meet all four: SecretLab Titan Evo ($300–$500), IKEA Markus ($230), Autonomous ErgoChair Pro ($350). All three are dramatically better than the $80 chair you're tempted to buy. Don't skimp here.
Internet: 50 Mbps is enough
AI training tasks are bandwidth-light. You'll see a few hundred KB per task at most. The thing to optimize for isn't speed — it's stability.
- Wired Ethernet beats WiFi for stability. If you can run a cable to your router, do.
- Have a mobile hotspot or backup connection. Lost progress mid-task on a flaky connection is the most expensive form of internet failure.
- VPN: not required, sometimes problematic. A few platforms (Mercor especially) flag VPN connections during onboarding. Turn it off when working.
Software: just two things
1. A clipboard manager.
You'll be copy-pasting code snippets, prompts, references constantly. A clipboard manager that keeps your last 50 copies saves hours per week.
- Mac: Maccy (free, open-source).
- Windows: Built-in Win+V is surprisingly good now.
- Linux: CopyQ.
2. A pomodoro / break timer.
The single biggest predictor of high quality scores after the first month is not getting fatigued. A 5-minute break every 30–45 minutes keeps your error rate down. Use any tool that pulses in your menu bar.
- Mac: Be Focused (free), Tomato Bar (free).
- Windows: Pomotroid (free).
- Cross-platform: Just a kitchen timer also works.
What you don't need
- A "professional" webcam. Most onboarding doesn't need video. Mercor's AI screener works fine with a built-in laptop camera.
- An RGB mechanical keyboard. Adds nothing.
- A standing desk. Nice, but not required. Sitting properly in a good chair is fine.
- Coding-themed accessories. Posters, wrist rests with logos, light strips. None of this affects your output.
- A specific operating system. Mac, Windows, Linux all work. Pick whatever you're already fluent in; switching mid-onboarding costs you a week of fluency.
The total bill, realistic budgets
- Cheapest viable setup: ~$400 — refurbished laptop, your existing $20 keyboard, basic chair you already own, $130 monitor.
- Reasonable mid-range: ~$900 — used M1 Air, $80 external keyboard, $230 IKEA Markus, $200 monitor.
- "Real workspace": ~$1,400 — new laptop, mid-tier ergonomic chair, dual setup, decent desk.
The cheapest setup is fine. People obsess over equipment as procrastination from doing the actual work. The actual work pays $30–$130/hr. Optimize for getting started, not for the perfect setup.