Home → Phone Farms
What an Android Phone Farm Actually Looks Like
A glossy YouTube tour of "my crypto farm" shows neat rows of phones blinking in unison. A real poker-bot farm looks like a half-broken bookshelf with three handsets temperature-throttling and a cable rats-nest you'd be embarrassed to show your mother. Here's what it actually is.
The hardware floor
A 60-seat build is the sweet spot for a single operator. Below that and your fixed costs (mini-PC controllers, racks, proxy contracts) dominate. Above that and you're hiring a second pair of hands to swap handsets at 3am. Sixty seats is also roughly the number of tables one strong agent can actually drive without queue-jitter ruining decision latency.
Physically, sixty Pixel 4a or Samsung A-series handsets sit on three steel shelving units, screens facing out. They USB-hub up to four mini-PC controllers — I use M75q-1 Tinys, but anything with enough USB lanes works. Each controller runs scrcpy or a custom adb-shell wrapper that mirrors device state to the agent. Power is split across two 1500W UPS units; a brownout will brick three or four handsets, and that's a $600 morning.
The cost structure, line by line
Numbers below are mine, June 2026, based on actual purchase receipts and proxy invoices. Markets differ, but ratios hold.
| Line item | One-off | Monthly |
|---|---|---|
| 60 × handsets (refurb Pixel/Samsung) | $5,400 | — |
| 4 × mini-PC controllers | $1,000 | — |
| Racks, hubs, cabling, UPS | $1,400 | — |
| Residential proxies (60 sticky IPs) | — | $1,200 |
| Electricity | — | $420 |
| Agent licence / engineer retainer | — | $600-$2,000 |
| Handset replacement (20-25% / month) | — | $220 |
That gives you a $9-12K upfront commitment and a $2.4-4K monthly burn before a single hand is played. Anybody quoting you "$300 lifetime licence and play forever" is either lying or selling you a binary that won't survive a week.
The failure modes nobody mentions
Three things will kill a farm faster than detection.
Temperature. Sixty handsets in a closed room idle at 36°C ambient. The screens dim, the SoC throttles, the agent's input timing drifts, and the rooms-side anti-collusion notices that your decisions take longer in the afternoon than the morning. You either AC the room or you accept the throttle and lose money to it.
Software updates. The room's Android app updates twice a month. Each update can shift a button two pixels, and if your agent reads the screen by template matching, you wake up to sixty bots calling check-fold every hand. Whoever's building your agent needs to ship a UI-diff layer, not just a poker brain.
Proxy quality. "Residential" proxy providers cycle their pools. Half of what you're paying for resolves to mobile carriers, the other half to AS-numbers that look very datacenter from the inside. Quarterly audits with a $20 IP-intel API are not optional.
Where the edge actually lives
The agent layer eats roughly half your theoretical edge. That sounds harsh; it's not. It just means a "GTO bot" that wins 8 BB/100 in a vacuum will win 3-4 BB/100 in production, after decision latency, missed timing tells, opponent profiling lag, and the implicit penalty rooms apply to suspected accounts. Below 3 BB/100, rake and proxy cost eat the rest. That's why operators with controlled liquidity — private clubs, NDA agreements, custom rake structures — are the only ones running this profitably at scale.
So why even bother with Android?
Because the rooms that matter are Android-first. PPPoker, Pokerrrr2, ClubGG, every Asian club app — those are mobile-native. There's no desktop client to script, no website to puppet. If you want to play where the soft games are, you play on a handset, full stop. iOS is harder still, so Android it is. The substrate is fine; the public APKs are not.
If you're sitting on an idea for a real build — your own private club, a managed liquidity contract, a custom agent for an app the public projects don't support — talk to me. We build, we don't ship downloads.