20M Bitcoin Mined, 1M Left: Public Miners Are Pivoting to AI Before the Finish

Bitcoin just crossed 20 million mined coins. Only 1 million remain—and those will unlock gradually over roughly 115 years. The supply clock stretches out; miner business models do not. The defining story now is miners retooling into high‑performance computing and AI to survive the next halving cycle.

The economic squeeze is straightforward. Hashprice has stayed stubbornly low while the 2028 halving approaches, cutting the block subsidy again. Many public operators sit near breakeven today, according to Needham’s John Todaro, while net operating income margins for HPC can run north of 80%. That math drives behavior: he expects a wave of public miners to wind down core Bitcoin operations in 2027–2028 and to sell down nearly all BTC on balance sheets before the end of 2026 to fund AI‑focused capex.

That potential selling sounds scary until you size it. Miners now hold roughly 0.5% of circulating BTC. Strategy alone holds about seven times what all miners combined do. If miners liquidate treasuries and redirect production, the impact on price likely skews to near‑term flow from newly minted coins rather than a structural overhang from giant miner wallets. Long‑time HODLers still anchor supply.

Survival, then, looks less like “mine more” and more like “own the stack.” Bitdeer’s posture is the template: convert select facilities into AI data centers, design next‑gen high‑efficiency ASICs in‑house, and lock in long‑dated energy. Control over chips, sites, and power grants optionality—route electrons to the highest‑margin workload at any given time. What used to be a nice dual engine (Bitcoin monetization plus AI infrastructure) increasingly reads like a requirement.

HIVE Digital illustrates the benefit of early optionality. Years before many peers, it began building HPC alongside its mining footprint, funded in part by Ethereum mining margins. When Ethereum’s 2022 merge turned proof‑of‑work obsolete, that diversification cushioned the transition. HIVE’s playbook—sourcing hydro and otherwise stranded power in Sweden and Iceland, then converting it into durable compute—captures a broader industry truth: cheap, reliable energy plus competent structuring beats scale for scale’s sake.

Zoom out to the milestone. It took 16 years to mine 20 million BTC. The final million drips out over a century. As the subsidy trends toward zero, the fee market matters more, and miner treasuries matter less. The whitepaper’s gold analogy still holds—new issuance is akin to miners adding metal to circulation—an idea echoed over the years by figures like BlackRock’s Larry Fink, Strategy founder Michael Saylor, and even Fed Chair Jerome Powell. Unlike gold, Bitcoin’s terminal supply and schedule are known. That predictability softens reflexive selling fears as halvings progress.

What to watch from here: - Power contracts and sites: fixed‑price, long‑tenor energy wins. Overpaying for electrons kills both mining and AI economics. - Retrofits: immersion cooling, fiber, and rack density upgrades determine whether a mine can credibly host AI/HPC or gets stranded as a single‑use asset. - Capital stack: project finance and off‑take agreements are replacing equity‑funded hash rate growth. Operators that secure non‑dilutive capital will outlast tourists. - Silicon cadence: efficiency gains in ASICs remain meaningful, but AI demand can crowd foundry capacity. Vertically integrated firms with design leverage get a margin edge.

On price, I’d fade blanket “miner capitulation” narratives. The glide path of issuance dampens shocks, while the miner share of supply is small. If pressure emerges, it likely comes from newly produced BTC hitting the market during transitions, not from legacy hoards being dumped. Meanwhile, the fee market’s evolution and broader institutional demand should carry heavier weight in medium‑term price discovery.

The next million coins won’t be mined by the same cohort that got us here. Operators that control energy, own critical parts of the stack, and can fluidly allocate compute between Bitcoin and HPC will still be around in 2028. Everyone else is renting time.