Poker Bots on Dedicated Servers: Infrastructure, Software and Real Risks

Poker bots have travelled a long way from crude scripts to neural networks trained on billions of hands. The people who run them increasingly deploy them on dedicated servers rather than a desktop at home. The reasons are practical: a home PC needs reboots, shares resources with everything else and goes offline whenever the power or internet does. A server in a data centre solves all three problems at once.

This article is an explainer, not a manual. It covers what the hosting side of bot operations looks like, which commercial bots dominate the market, where the technology comes from, and, most importantly, why the economics rarely work out once detection and confiscations are priced in.

Why bots moved from home PCs to servers

A home computer can run one or two bots. Beyond that, the operation hits a ceiling fast:

• Uptime. Home internet and electricity are not designed for 24/7 workloads. Every reboot means lost hands.
• Resources. A browser, a stream or even a Windows update competes with the bot for CPU and RAM.
• Scaling. Adding a fifth bot means a second machine, a second desk and a second monitor.

Infrastructure: virtualisation with Proxmox

What Proxmox is and why it became the standard

Proxmox Virtual Environment (VE) is an open-source platform for creating and managing virtual machines. It lets one physical server run several isolated operating systems side by side. For bot operators that means every bot lives in its own VM with its own OS and its own MAC address.

The advantages of the approach are the same ones any sysadmin would list:

• Isolation. A crash or compromise inside one VM does not touch the others.
• Scaling. A new bot is a new VM. No new hardware.
• Snapshots. Before the poker client updates, the operator snapshots the VM and can roll back in seconds if the update breaks the bot.
• Web interface. Every VM is managed from a browser tab.

Typical hardware tier

Operators usually rent a mid-range dedicated box: a 6- to 8-core CPU, 32–64 GB of RAM and an NVMe drive. Each Windows VM running an Android emulator and a poker client comfortably needs 2 cores and 4 GB, so a 64 GB machine hosts roughly 10–12 bots before latency creeps into the decision loop. Latency matters: a bot that takes six seconds to act on every street is the first thing a security analyst notices.

The main commercial bots, from classic to AI

Shanky Holdem Bot

The oldest commercial bot on the market, sold since 2007. Version 12 supports cash games, tournaments and sit-and-gos with up to six tables at once.

• Pros: eighteen years of polish, ready-made profiles, simple to launch.
• Cons: limited adaptivity; it does not use live opponent statistics.
• Price: subscription plus paid profiles.

Warbot

An advanced bot built on the open-source OpenHoldem engine. It reads the table visually (screen scraping) and takes decisions from external profile files.

• Pros: highly customisable, supports dozens of rooms.
• Cons: needs technical skill; strong profiles are sold separately.
• Price: paid, despite its open-source roots.

Hydra

Marketed as an "AI poker bot". It integrates with PokerTracker, reads opponent statistics from the database and adapts its strategy.

• Pros: adaptive play, HUD integration.
• Cons: well known to rooms (BetOnline has banned it by name).
• Price: premium, with profile packs (Kraken, Medusa) sold on top.

Poker Bot AI+

A service rather than a standalone program. It runs neural networks trained on billions of hands and targets mobile club apps such as PokerBros through Android emulation.

• Pros: state-of-the-art AI, mobile emulation.
• Cons: high cost, custom setups, legal grey zone.

Comparison table

Where the technology comes from

In 2017 Libratus, built at Carnegie Mellon, beat four top professionals in heads-up no-limit hold'em over 120,000 hands. In 2019 Pluribus, a Facebook and CMU project, became the first AI to win in six-max NLHE. The techniques behind those systems, counterfactual regret minimisation, reinforcement learning and neural networks, have steadily migrated into commercial products. The gap between an academic paper and a Fiverr listing is now about three years.

How rooms detect server-hosted bots

Operators like to believe that a data-centre box with a fresh VM is invisible. Security teams disagree, and the numbers are on their side. PokerBros reports a proactive detection rate of 96–98% for its Game Integrity BROS team, which combines machine-learning models trained on millions of legitimate hands with a real-time alerts engine. In 2026 the app added a Photo Rotation Verification check at the table that a fully autonomous bot cannot pass at all. We took that system apart in our review of the PokerBros anti-bot feature.

What gets flagged, in rough order of how quickly it happens:

1. Timing signatures. Identical decision times street after street, or a session that never pauses.
2. Statistical anomalies. Perfect GTO frequencies at NL10, where nobody plays that well.
3. Session patterns. Sixteen-hour sessions, seven days a week, with no losing days.
4. Network patterns. Many accounts from one hosting provider's IP range.
5. Player reports. Regulars notice a bot long before the model does, and reports trigger a manual review.

What a ban costs

Even with perfect infrastructure, the ban eventually arrives, and the consequences are financial, not theoretical:

• Account closure with confiscation of the balance.
• Personal ban tied to documents and payment details, which follows the operator to future accounts.
• Confiscated winnings. 888poker seized roughly $290,000 from bot accounts in 2022. PartyPoker confiscated more than $700,000 in 2019 and redistributed it to the affected players.

Add the server rent, the bot licence, the profile packs and the operator's own time, and the expected value of a bot farm turns negative for everyone except the people selling the software. The Fiverr poker bot market makes that asymmetry very visible: the sellers are the only party with a guaranteed profit.