LESSON 1

Traditional Proof of Work (PoW) in blockchains makes miners burn energy on meaningless puzzles, purely to secure the network . Qubic flips this script with Useful Proof of Work (uPoW) – a consensus mechanism where the computational work actually trains AI models instead of hashing useless data . In Qubic, your mining rigs perform machine learning tasks that contribute to a global neural network (nicknamed Aigarth) rather than wasteful cryptographic calculations . This means every joule of energy spent mining Qubic also pushes an AI project forward, aligning network security with real-world useful work.

Qubic’s uPoW mainly leverages CPU power for AI training tasks . Miners (often called “AI trainers”) run intensive operations like matrix multiplications and tensor updates to train artificial neural networks . The network validates that these tasks are done correctly via special validator nodes called Computors . Computors ensure miners are honestly contributing to AI objectives and not cheating.

Importantly, decentralization is preserved: instead of a single data center training an AI, Qubic distributes the workload across a global swarm of independent miners . From 31 December 2025, all Computor operators will require AVX-512, so optimisation should focus on modern CPUs.

How rewards work: Qubic runs on 7-day cycles called epochs. Throughout an epoch, every miner accumulates a performance score based on useful work completed (solving AI tasks = Solutions. At the end of the epoch, top miners become benign candidates first. Then they may be promoted to Computor and join quorum if they provide enough valid solutions and outperform competing miners during verification.

Computors receive the rewards, and rewards are distributed to miners based on valid solutions from the previous epoch.

If you don’t become Computor and don’t reach the required number of solutions, you may receive no reward. Pool miners still receive rewards because the pool handles distribution, and because the Quorum needs Computors.

In short, miners do the heavy lifting and Computors validate, with payouts every Wednesday.

Qubic’s uPoW directs mining power into training neural networks for the Aigarth project, rather than wasting it. This approach not only secures the network but also makes progress toward AI goals.

RAM mining (RandomX AI Mining) was originally designed as a clever “idle work” system — allowing miners to remain productive when no AI tasks were available. However, the network now always has active AI mining tasks, so RandomX serves purely as a Proof of Concept (PoC) fallback. There are no idle epochs today; miners always have live AI workloads to process.

If RandomX fallback is used for testing, the Monero (XMR) mined is not automatically converted or burned. Instead, it’s manually sold for QUs by core contributors at the end of the epoch. Those purchased QUs are typically burned, though in some cases, portions have been redistributed as community or Computor rewards. This means the original automatic-burn mechanism is no longer active, but the XMR-to-QU conversion process still benefits Qubic’s economy.

In practice, miners continuously perform AI work without switching to RandomX under normal conditions — keeping hardware fully utilized toward training Aigarth.

Below, you can watch a comprehensive video on Qubic's uPoW. You can download a PDF summary of the video here.