Bitcoin Mining Centralizes as AI Moves to Edge Models

Alex Thorn, head of Galaxy Research, said on Sunday that Bitcoin mining has moved from a hobbyist activity on home computers to a process dominated by specialized hardware and large farms. He contrasted that change with recent developments in artificial intelligence.

Thorn explained that mining now requires application-specific integrated circuits, known as ASICs, and industrial-scale operations to be competitive. Large operators able to buy equipment and secure low-cost power have captured much of the network’s hashing capacity, he said, reducing the number of individual machines participating in mining.

On AI, Thorn described an opposite trend. He said the field started with massive clusters run by cloud providers, but limits at the frontier — including data scarcity, short context windows and memory bottlenecks — are encouraging smaller, more efficient models. He pointed to open-source software and efficiency improvements as factors that could let more AI tasks run on phones, laptops and dedicated edge devices.

“If local models keep getting smaller, cheaper, and more efficient, AI may become increasingly personal and on-device,” Thorn added.

Market data cited by Thorn shows strong growth in edge AI. Grand View Research projects the global edge AI market to rise from about $25 billion in 2025 to roughly $119 billion by 2033, driven by more Internet of Things devices, demand for low-latency processing and concerns about data privacy.

An analysis from a crypto exchange noted rising energy costs have made mining unprofitable in parts of the United States, with the cost to mine a single bitcoin exceeding $100,000 in some regions. That has shifted hashing power to countries with surplus cheap electricity, including Paraguay and Ethiopia, where hydroelectric capacity has attracted mining activity.

Despite geographic shifts, mining concentration remains high in many places. Industrial farms and large pools still account for a large share of hash rate. On the AI side, major cloud providers retain advantages in access to large datasets, specialized chips and capital, which can affect how quickly on-device models handle more tasks.

ASICs are chips built specifically for cryptocurrency mining; they perform hashing far more efficiently than general-purpose processors. Edge AI refers to running machine-learning models on end devices instead of sending all data to central servers, which reduces latency and can keep sensitive data on the device.

Future changes in hardware costs, energy prices, regulation and improvements in model efficiency will affect where mining and AI computing concentrate or spread.