SBI's Exit Was a Symptom: The 60% Hashrate Concentration Was Already Settled

Math doesn't care about a pool's brand. On July 31, SBI Crypto's seven-day average hashrate read 0.452 EH/s. One month earlier, on June 30, it read 16.222 EH/s. That is a 97.2% collapse in thirty-one days. The pool's Stratum endpoint stopped serving miners; attributed blocks dropped to zero; the operation closed. Headlines will call this another blow to Bitcoin's decentralization. The data tells a different story. The top three pools — Foundry, AntPool, F2Pool — had already crossed 60% of attributed blocks before SBI completed its exit. On July 20, their combined share was 64.8%. On July 27, it was 60.8%. SBI's departure did not create the concentration. It was the lagging indicator of a structure that had formed while everyone was watching price charts. Mining pools are not part of Bitcoin's consensus layer. They are coordination layers. A pool aggregates hashrate from many independent miners, submits candidate blocks on their behalf, and splits rewards according to contributed work. The protocol sees a single block producer. The pool operator decides which transactions go into block templates, how fees are bid, and when a block is broadcast. That is a lot of unexamined authority. SBI Crypto was not a minor operator. It had been running Bitcoin mining services through SBI group, with a footprint in Japan and some enterprise mining. Its exit is a business decision. But because miners route their work through Stratum, the shutdown forced thousands of connected machines to find a new endpoint. The technical act is trivial: change a URL, update a port, restart the miner. The market act is not. Those miners will land somewhere, and if they land with the big three, the sector tightens. The core chain remained untouched. Bitcoin's PoW consensus, difficulty adjustment, UTXO model, and block subsidy schedule changed exactly nothing. A pool's closure is not a protocol event. It is an infrastructure event. Yet because infrastructure shapes who can participate, it deserves forensic attention. Start with the word "attributed." Hashrate cannot be measured directly. It is a probabilistic estimate derived from solved blocks. When a pool solves a block, the coinbase address inside that block marks the solution as belonging to that pool. Indexers count those blocks over a window and divide by total blocks. The result is a share estimate, not an exact measurement of connected hardware. This matters because the 60% figure everyone quotes is precisely that: an attributed share number, not a physical control measurement. A 60.01% reading is a snapshot. It can move 10 to 15 points in a week due to luck, sudden migration, or a single large industrial miner renegotiating a contract. It is not proof of persistent majority control. In my work auditing protocol infrastructure, I have learned to treat leaderboards as political signals, not engineering measurements. The first thing I check is whether a data source reports confidence intervals. The second is whether attribution is based on observed work or self-reported work. Most public mining dashboards provide neither. Their numbers are useful for trends, dangerous as thresholds. The real number worth inspecting is SBI's trajectory. The July 31 telemetry endpoint showed 0.452 EH/s in a 24-hour window. The seven-day average had fallen to 5.817 EH/s by July 30. That means the pool's own infrastructure had been losing hashrate long before the official announcement. In forensic terms, the pool was not killed by the market; it was already decaying. Its miners were leaving, and the closure merely formalized the decline. Why did they leave? The 2024 halving cut block subsidy from 6.25 BTC to 3.125 BTC. Pool revenue scales with block rewards plus transaction fees; when subsidy halves, a pool with constant hashrate earns half as much after payout. SBI, facing Japanese electricity costs and corporate overhead, no longer had the unit economics to justify operation. The closing of a pool in that context is not a technical failure. It is the normal output of a cost function with too many negative terms. But the aggregate numbers miss the flow. When miners migrate from a closing pool, the old pool's telemetry shows a decline, and the winning pools' shares increase with a lag. Because attribution is based on solved blocks, not on live work, the migration may be partially invisible for hours or days. My confidence in the reported "SBI dropped to 0.452 EH/s" is high. My confidence that all 0.452 EH/s disappeared from the market is low. The machines did not evaporate. They reconnected to other pools, likely without a single human touching them. This is the blind spot of any leaderboard: it records outcomes, not intentions. Now examine the structural claims about small pools. The same seven-day window that showed SBI falling also showed Luxor rising, Braiins falling, and NeoPool absent. That is not noise. It is the mid-tier churn that occurs when mining margins compress. Small pools survive by offering better data analytics, hashrate derivatives, or customized services. They die when their differentiation no longer pays for their infrastructure costs. SBI had no such differentiation. It was a corporate pool with a Japanese brand, not a technology edge. The market replaced it with cheaper or more useful service providers. The top three pooled 60% of attributed blocks before SBI closed. Does that mean the network is endangered? Not in the way the panic tweets suggest. A mining pool cannot unilaterally rewrite history without controlling actual hashrate. Miners can switch pools in seconds. The moment a pool signals a malicious block template or an abusive fee structure, its hashrate evaporates. This is the forgotten equilibrium. In game-theoretic terms, the pool's power is constrained by the threat of coalitional exit. Miners are not hostages. They are fee-sensitive rational agents. Trust is a vulnerability, not a virtue. The honest check against pool power is not shareholder goodwill; it is the fact that mining pool switching cost is near zero. Change Stratum parameters; wait six blocks; collect funds at a new address. That is why "60% concentration" is less dangerous than it seems — but not for the reason you think. The real danger is not a rogue majority. It is the quiet standardization of block template policy. If top pools all run the same software, use the same default transaction selection, and operate under the same jurisdiction or compliance framework, they become behaviorally identical without colluding. A 60% share of identical behavior is indistinguishable from central planning at the network layer. In my experience auditing mining-adjacent code, the most under-examined path is the block template builder. Most miners accept whatever their pool sends. They do not inspect the template. Pool operators have the technical power to exclude transactions from blocks, reorder tx, or include specific transactions for political reasons. That is a form of censorship — not at the protocol level, but at the aggregation level. Privacy is a protocol, not a policy. When a pool's block template becomes a filter, user privacy and network neutrality are modified at the service layer, below any official protocol change. SBI's exit should be a reminder that pool closures are normal business events, not existential threats. What should scare the industry is a market structure where only the largest pools can survive because of compliance overhead, financial services, and institutional relationships. That produces an oligopoly of behavior, not just hashrate. Foundry at 26.67%, AntPool at 17.13%, F2Pool at 16.21% — those percentages are less important than whether they all start adopting the same transaction review policies. Watch the block templates. Watch the pool operators' terms of service. Watch whether miners are receiving raw, unmodified work or a sanitized version. The next vulnerability is not a pool shutdown. It will be a month when no one notices that all top pools, under separate flags, decided to exclude the same category of transaction. That is how consensus becomes a rumor. Math doesn't care about the next price rally; it only cares whether the network's incentives are still aligned. Maybe SBI's closure is just one data point. The useful question is not "Why did a pool close?" It is "What will a miner do when the block template arrives with an empty mempool?" That question has no comfort in a leaderboard. It has to be answered by code, by incentives, and by the stubborn fact that the network is only as decentralized as the choices miners are actually willing to make.