Timechain Wiki

ASICs and mining hardware

11 min read

Bitcoin mining is performed by Application-Specific Integrated Circuits (ASICs) — silicon chips engineered to compute SHA-256 hashes at orders-of-magnitude higher efficiency than general-purpose CPUs or GPUs. The hardware market is structurally concentrated among three principal manufacturers (Bitmain, MicroBT, Canaan), with smaller participants such as Block's Proto effort and Bitdeer's Sealminer. Current-generation flagships (Bitmain's Antminer S23 family, launched early 2026 — the S23 Hyd at ~9.5 J/Th; MicroBT's Whatsminer M7x series at ~14.5 J/Th) push efficiency to roughly 9.5-15 J/Th, down from the ~13-15 J/Th S21/M60 generation of 2024 and 30+ J/Th in 2020. Obsolescence is driven by efficiency rather than hardware failure — older ASICs become unprofitable as difficulty rises past break-even power costs, but the chips remain operational and migrate to cheaper-energy jurisdictions. Manufacturing depends critically on TSMC and Samsung foundries; supply-chain disruptions affect the entire industry simultaneously. The hosting-vs-self-mining distinction structures the operational landscape.

Why this note matters

ASIC hardware is the physical substrate of Bitcoin mining. The economic competition over hash production is mediated entirely through ASIC efficiency: who can produce hashes most cheaply per joule wins the long-run mining race. Understanding ASIC generations, manufacturer dynamics, and supply-chain dependencies is the precondition for evaluating mining-industry concentration, energy-mix changes, and the geopolitical implications of mining location.

The hardware layer also constrains the broader mining-industry dynamics. Hardware-manufacturer concentration (effectively three principal manufacturers globally) is a real centralization vector that the analytical-critique notes engage; this section provides the operational-empirical foundation those critiques rest on.

The ASIC efficiency trajectory

Bitcoin mining has progressed through five hardware eras:

  • CPU era (2009-2010). Original Bitcoin mining used general-purpose CPUs. Efficiency was effectively unmeasurable in J/Th terms because hashrate was so low.
  • GPU era (2010-2013). GPU mining offered substantial efficiency improvements over CPU; the era ended when FPGAs and then ASICs made GPUs uncompetitive.
  • FPGA era (2011-2013). Field-Programmable Gate Arrays provided an intermediate-efficiency option but were quickly displaced by ASICs.
  • Early ASIC era (2013-2017). Avalon (Canaan), KnCMiner, Cointerra, and early Bitmain (Antminer S1, S3, S5, S7, S9). The S9 (2016) ran at ~100 J/Th and remained operationally relevant until 2020 in cheap-power jurisdictions.
  • Modern ASIC era (2017-present). Generational progression from ~30 J/Th (S17 era, 2019) to ~13-15 J/Th (S21 Pro era, 2024) to ~9.5-11 J/Th (Antminer S23 era, 2026). Improvement curve is slowing as silicon-physics limits approach.

The efficiency trajectory. Per-joule hash production has improved by roughly 5× over the past five years and roughly 100× over the past decade. The trajectory is slowing — silicon-physics-driven limits (transistor size, leakage current, thermal dissipation) constrain how much further efficiency can be pushed without major-architectural changes.

The hash-per-joule metric. The standard operational efficiency measure is Joules per Terahash (J/Th) — lower is better. Current flagship ASICs achieve ~9.5-11 J/Th (Antminer S23 generation, 2026); the prior S21/M60 flagship generation ran ~13-15 J/Th; deployment-grade ASICs across the fleet are typically 15-20 J/Th; legacy hardware (S19 family) operates at 25-35 J/Th and is profitable only in very-cheap-power jurisdictions.

The manufacturer landscape

As of 2026, three principal manufacturers dominate the ASIC market:

Bitmain. The largest. Based in China; produces the Antminer S-series (current flagship: the S23 family, launched early 2026 — the hydro-cooled S23 Hyd at ~9.5 J/Th is the most efficient production miner available). Estimated 60-70% of the ASIC market by hashrate. Bitmain has navigated multiple geopolitical transitions (Chinese mining ban 2021; export controls; chip-foundry access) and has remained the dominant participant. Bitmain also operates substantial mining operations directly through subsidiaries.

MicroBT. Bitmain’s principal competitor. Also Chinese-headquartered; produces the Whatsminer M-series (current flagship: the M7x line, e.g. the M79 at ~14.5 J/Th, succeeding the M60 series). Estimated 25-30% of the ASIC market. MicroBT has historically offered competitive efficiency at similar price points to Bitmain.

Canaan. Smaller market share (5-10%). Publicly traded on NASDAQ since 2019. Produces the Avalon series (current: the Avalon A16XP at ~12.8 J/Th). Strategic positioning as the more-transparent publicly-accountable manufacturer.

Emerging and historical participants:

  • Intel. Entered the ASIC market with the Blockscale series in 2022; exited in 2023 after the chip-design approach proved uncompetitive. Intel’s exit was viewed as a setback for hardware-manufacturer diversification.
  • Block (Square). Has signaled ongoing ASIC-design effort through the Proto program; production hardware is not yet at scale.
  • Bitdeer / Sealminer. Singapore-based; positioned as a publicly-traded alternative; market share is growing but still small.
  • Auradine (US-based startup) and various smaller participants.

The geographic-manufacturer concentration. Both Bitmain and MicroBT are Chinese-headquartered; design and significant operational footprint is in China; foundry production is at TSMC (Taiwan) and Samsung (South Korea). This produces a structural concentration vulnerability: a Taiwan-related geopolitical disruption would affect the entire industry simultaneously.

ASIC design and the SHA-256 specialization

ASICs are specialized to compute one task: SHA-256 hashing. The chip’s architecture is dedicated hashing circuitry — no general-purpose computation, no memory subsystems beyond what’s needed for the SHA-256 algorithm, no I/O beyond the minimal control interface.

The specialization is extreme. A modern Bitcoin ASIC dedicates ~99%+ of its silicon area to SHA-256 hashing circuitry. By contrast, a general-purpose CPU dedicates only a small fraction of its silicon to any single computational task. The specialization produces the per-joule efficiency advantage.

The cost of specialization. ASICs cannot mine anything other than Bitcoin (and merge-mined Bitcoin-derivative chains; the algorithmic family of “double-SHA-256-on-block-headers” includes all Bitcoin forks). Switching to mining a different cryptocurrency requires different hardware. This is the structural reason mining hardware is specifically Bitcoin-aligned: ASIC investment is non-fungible across cryptocurrency ecosystems.

The chip-architecture progression. Modern ASICs use 5nm to 3nm fabrication processes (TSMC N3, N5 nodes). Sub-3nm fabrication is approaching but yield-and-cost considerations slow the transition. Power-and-cooling architecture is increasingly important — modern ASICs run hot enough that air-cooling has substantive limitations and immersion-cooling has emerged as the preferred large-deployment cooling mode.

Supply chain and foundry dependency

Bitcoin ASICs are manufactured at semiconductor foundries — primarily TSMC (Taiwan Semiconductor Manufacturing Company) and Samsung Foundry. Bitmain and MicroBT both order chip production from these foundries; the design is the ASIC manufacturer’s, but the silicon comes from the foundries.

The foundry-capacity allocation. Bitmain and MicroBT compete with AI-chip designers, smartphone designers, and other ASIC customers for foundry capacity. During the 2020-2022 chip shortage, ASIC manufacturers reported substantial delivery delays; this dynamic recurs whenever foundry capacity tightens.

The geopolitical-vulnerability concern. TSMC’s location in Taiwan creates a structural geopolitical exposure for the entire Bitcoin mining industry. A major disruption to TSMC operations (military, infrastructure, or political) would produce a multi-year shortage of new mining hardware. Samsung’s South Korean facilities provide some diversification but are smaller in capacity. US-based foundry expansion (TSMC Arizona; Intel) is in progress but not yet at scale for ASIC production.

The supply-chain segmentation. The ASIC supply chain operates in phases: chip design → foundry production → ASIC manufacturer integration → distribution → end-user mining. Each phase has its own bottlenecks and concentration dynamics.

ASIC obsolescence and the secondary market

ASIC obsolescence is driven by efficiency, not hardware failure:

  • Older-generation ASICs (e.g., S19 family at 25-35 J/Th) become unprofitable when network difficulty rises above their break-even hash price.
  • The hardware itself remains operational; the chips don’t fail at end-of-economic-life.
  • Obsolete-in-one-jurisdiction ASICs are routinely redeployed to lower-cost-energy jurisdictions where break-even economics still favor older hardware.

The secondary market. Used ASICs are actively traded on a large secondary market. Pricing tracks hash-price expectations and energy-cost projections for the target deployment jurisdiction. Major resellers (CompassMining, Sazmining, others) provide brokered ASIC sales.

The hardware retirement question. The Bitcoin industry’s hardware-retirement-and-recycling dynamics are less well-documented than newer-generation deployment. Obsolete ASICs that cannot be profitably redeployed accumulate; the e-waste implications are real but the empirical scale is contested.

Hosting vs self-mining

Most modern Bitcoin mining operations split into two distinct models:

Self-mining. The ASIC owner also owns and operates the mining facility (power infrastructure, cooling, networking, security). Self-mining is the dominant model for large public miners (Marathon, Riot, CleanSpark) who own substantial facilities and operate hardware they purchased directly.

Hosting. The ASIC owner purchases hardware and pays a third-party operator to host it at the operator’s facility. Hosting models include flat-rate-per-kWh (the host charges electricity costs plus a per-kWh fee), revenue-share (the host receives a percentage of mining revenue), or various hybrid structures. Hosting is dominant for smaller-scale ASIC purchasers and for facilities operated by non-mining-focused energy companies.

The hosting market segmentation. Hosting providers vary in trust profile, contract structure, and operational reliability. Some hosts have collapsed (Compute North 2022; Core Scientific bankruptcy 2022-2023) producing material asset losses for hosted customers. Due-diligence on hosting providers is an active operational discipline.

Tradeoffs and design choices

Hardware-manufacturer concentration vs decentralization-ideal. Three principal manufacturers globally is a structural concentration. The alternative (many smaller manufacturers) would distribute supply-chain risk but at the cost of efficiency competition (ASIC design has substantial fixed-cost economies of scale). The competitive dynamic produces ~3 viable participants at any given time. See Mining centralization concerns for the substantive analytical engagement.

Foundry concentration vs supply-chain resilience. TSMC + Samsung covers essentially all advanced-node ASIC fabrication. The geopolitical-vulnerability concern is real; alternative-foundry development (Intel; US-based TSMC capacity) is slower than ideal. The industry has limited optionality on this dimension.

Hosting vs self-mining tradeoffs. Self-mining captures more margin but requires substantial capital, operational expertise, and facility-management capability. Hosting reduces capital requirements at the cost of margin share. The right choice depends on the operator’s scale and operational capability.

Air-cooling vs immersion-cooling. Air-cooling is operationally simpler and has lower capex; immersion-cooling is more efficient and supports higher-density deployments but adds capex and operational complexity. The contemporary trend favors immersion for large-scale new facilities.

ASIC obsolescence and e-waste. The hardware-retirement dynamics are an operational reality that the industry has not yet developed mature recycling-and-disposal protocols for. The empirical scale is contested but the structural concern is real. See Environmental and energy-consumption critiques for the substantive analytical engagement.

Substantive analytical critique of hardware-manufacturer concentration lives in Mining centralization concerns; environmental and e-waste concerns are engaged in Environmental and energy-consumption critiques.

Open questions for further development

  • What is the realistic ASIC-efficiency endpoint? Silicon-physics limits constrain how far J/Th can be pushed; with the 2026 flagship (S23 Hyd) already at ~9.5 J/Th, the empirical trajectory will likely plateau somewhere in the ~3-7 J/Th range over the coming decade.
  • Can hardware-manufacturer diversification meaningfully increase? Intel’s exit suggests the barrier-to-entry is high; Block’s Proto effort is in early stages.
  • How will TSMC-geopolitical risk evolve? This is the principal supply-chain exposure for the industry; the trajectory is unclear.
  • What is the appropriate e-waste-and-recycling framework for retired ASICs? Industry-level protocols are immature.
  • How does the AI-chip competition affect ASIC foundry-capacity access? AI compute demand is a major competitor for foundry capacity; the long-run capacity-allocation dynamics affect ASIC delivery timelines and pricing.

Canonical sources for this note

  • Hashrate Index (Luxor Technology) — quantitative mining-industry data including hardware-by-model hash share
  • Bitmain, MicroBT, Canaan, Bitdeer product pages and technical specifications
  • Compass Mining, BraiinsOS+, Hashrate Magazine — industry-level coverage and operational reference
  • Public-miner quarterly filings (Marathon, Riot, CleanSpark) include hardware-deployment composition
  • The Bitcoin Standard - Saifedean Ammous — Chapter 9 engagement with mining-energy and hardware dynamics
  • Broken Money - Lyn Alden — empirical-macro engagement with mining infrastructure

Graph View

Backlinks