How to build an Instagram farm

Step 1. Decide the network before buying anything

Every account behind one connection shares an IP, which links them to each other permanently. Three options: a SIM per phone, which is cleanest and most expensive; mobile proxies covering groups of devices, which is the common middle path; or a shared connection, which is fine only at small scale. Choose now, because retrofitting is worse than planning.

Step 2. Buy hardware in one consistent batch

Refurbished iPhones in a single model. One control chip per phone plus spares. Powered USB hubs only. A rack with real airflow. Do not buy a hundred phones before ten are proven end to end.

Step 3. Set up the driving machine

Windows. The kernel service that talks to the chips does not run on macOS, and any plan that starts with Apple hardware on the controlling side has to be redone. A modest machine handles a surprising number of devices because the heavy work happens on the phones.

Step 4. Build the software layer

This is the project, and the order inside it matters too:

1. Drive one phone reliably. About a week.
2. One account, full cycle: create, verify, set profile, warm, post, read result.
3. Ten phones concurrently. This is where USB bandwidth, the controlling machine and your assumptions all get tested.
4. A queue with leasing, timeouts, retries and circuit breakers.
5. Ban detection and automatic replacement.
6. Encrypted credentials with access logging.
7. Observability, before you scale, not after.

Step 5. Instrument before you grow

Record every ban against the phone, the network group, the account age at ban, the warm up it received and the content posted. Within a few weeks the pattern appears, and it is almost always concentrated rather than spread. Growing without this multiplies a leak you cannot see.

Step 6. Ramp in stages

Ten phones, then thirty, then fifty, then a hundred. Measure ban rate at each step before the next. Two phones working proves the automation logic and nothing about two hundred, which adds concurrency, heat, queueing and detection at scale.

Step 7. Solve content supply in parallel

A hundred phone fleet at three accounts and three posts is nine hundred posts a day. Most agencies hit this wall after the fleet is built, which is the worst possible time to discover it.

The mistakes that cost most

• Buying a Mac to drive the fleet.
• Unpowered USB hubs.
• No airflow plan.
• Deciding the network after the rack is built.
• Scaling before instrumenting.
• Running the agent in a terminal window so it does not survive a reboot.