31 Days to Zero
SBI's telemetry does not lie. On June 30, the Tokyo-based institutional mining pool carried 16.222 EH/s across its Stratum infrastructure. Thirty-one days later, that figure registered 0.452 EH/s. A 97.2% collapse in a single calendar month. By July 29, the pool had stopped producing attributed blocks entirely. The Stratum servers โ the coordination layer connecting roughly 0.07% of the network's hashpower to Bitcoin's consensus layer โ were dismantled in a phased, orderly shutdown.
Here is the number the headlines missed. The top three pools had already breached 60% aggregate share a full week before SBI disconnected its final worker. July 20: 64.8039%. July 27: 60.7843%. The concentration predated the exit. SBI's departure is the consequence, not the cause.
That distinction carries analytical weight. When a mining pool closes, the reflex narrative attaches causality to the event itself: SBI exited, therefore hashrate concentrated. This is backwards. The structural drift toward a triopoly was already embedded in the network's pool distribution. SBI's exit merely rendered it legible.
Stratum, Attribution, and the Coordination Layer
To understand why this matters, the mechanics have to be precise. Bitcoin miners do not connect directly to the consensus layer. They connect to Stratum โ a text-based protocol that coordinates work distribution across thousands of ASICs. A pool aggregates individual miner hashpower, assigns work packages, validates shares, and recombines results into network-valid blocks. The pool, not the miner, chooses the block template. The pool, not the miner, constructs the coinbase transaction. The pool, not the miner, decides which transactions enter the block.
This is a structural feature, not a bug. Pool-level concentration is not a technical deficiency in Bitcoin's Core protocol. The PoW difficulty adjustment, the UTXO model, the 21 million hard cap โ none of it changes. What concentrates is the point of coordination. Stratum is a trust-minimized lattice in theory and an operational chokepoint in practice.
Attribution data complicates the picture further. The 60% reading is not a measurement of hashrate. It is a count of attributed blocks โ blocks whose coinbase signatures identify a specific pool operator. Foundry USA's 26.67%, AntPool's 17.13%, and F2Pool's 16.21% are derived from a sampled window of approximately 1,000 blocks. This is a proxy, not a measurement. A pool producing one block per five minutes on average can experience statistical variance of several percentage points across any given window. The 60.01% reading captures an instant. It does not capture sustained control.
Proofs verify truth, but context verifies intent. The context here: SBI's Stratum service ran at full capacity until the final week. Its miners were not orphaned mid-stream. They were given transition windows. The orderly shutdown is itself a data point. SBI was not a distressed operator abandoning infrastructure. It was an institution making a calculated business decision.
The Predated Threshold: When 60% Actually Formed
My background is in code verification, not mining economics. But the forensic discipline transfers. When I manually audited ZKSwap's rollup aggregation logic in 2019, the lesson was that state mismatches hide in aggregation layers โ components that combine many inputs into a single output. Mining pools are aggregation layers. The same analytical discipline applies.
The first forensic question: when did the 60% actually form? The weekly bucket data is unambiguous. July 20 recorded 64.8039%. July 27 recorded 60.7843%. SBI's formal closure ran through late July. The concentration threshold was crossed before the closure event. This is not a post-exit effect. It is a pre-existing structural condition.
This reframes causation correctly. Bitcoin's mining industry has been consolidating since the 2024 halving compressed block subsidies from 6.25 BTC to 3.125 BTC per block. Pool revenue โ derived from the 1% to 4% fees charged on miner rewards โ was cut in half at the network level. Miners earning 3.125 BTC per block have less gross margin to absorb pool fees. Pools carrying institutional overhead, like SBI's Tokyo operation, face an operating-cost structure that scales poorly against declining revenue per unit of hashpower.
Logic holds until the gas price breaks it. The gas price here is the cost of maintaining a globally competitive Stratum fleet while charging competitive fees. SBI's economics broke first. Its 7-day average hashrate fell from 16.222 EH/s on June 30 to 5.817 EH/s by July 30 โ a 64% sustained decline over the month, not a sudden drop. Then came the cliff: 0.452 EH/s by July 31. The 24-hour average collapsed only after the pool stopped producing blocks on July 29.
The SBI case is consistent with a lagged response to the 2024 halving. I assign medium confidence to this interpretation, but the evidence aligns. Mining pools generate revenue exclusively from block subsidies plus transaction fees. The halving cut the subsidy. Transaction fee spikes from Ordinals and BRC-20 activity partially offset this โ but those fees are volatile, not contractual. A diversified financial institution like SBI Group would model that volatility and conclude that mining pool operation is a poor risk-adjusted use of capital. Japan's industrial electricity prices, among the highest in developed Asia, compound the problem.
Hashrate Accounting and the Variance Problem
The second forensic question: what does the attribution data actually tell us? Hashrate Index and mempool.space both derive pool shares from attributed blocks. The method is sound as a directional indicator and imprecise as a measurement of control. Attribution counts blocks. It does not count hashes. A pool with exceptional luck can temporarily inflate its attributed share; a pool with poor luck can understate it.
Consider the math. Over a 1,000-block window at a ten-minute average block time, the network produces roughly 144 blocks per day. A pool with a true 20% share expects about 29 blocks per day. The standard deviation of a Poisson process at that rate is roughly 5.4 blocks. That translates to a daily variance of nearly two percentage points of attributed share โ without any real change in deployed hashrate. The weekly buckets smooth this, but they do not eliminate it. The 64.8% reading on July 20 and the 60.8% reading on July 27 could differ by nearly four percentage points of pure statistical noise.
There is also a migration blind spot. When SBI's Stratum service shut down, its miners faced a binary choice: reconfigure their Stratum connection parameters to a new pool, or cease operation. The technical switching cost is negligible โ modifying a connection string in mining firmware. The economic switching cost is a few hours of missed earnings during transition. But aggregated data cannot identify where those miners went. SBI's telemetry reflects only the hashrate the SBI name serves, not the fate of its miners. The computed hashrate is a bookkeeping relic.
In the dark, zero knowledge is just a guess. I put medium confidence in the hypothesis that a portion of SBI's residual hashrate spontaneously migrated to other pools in the final days of the shutdown window. Mining services markets have no friction beyond firmware configuration. Rational miners do not idle profitable ASICs.
The Economics of Exit
The tokenomic layer is cleaner. Bitcoin's supply model is unchanged โ a hard cap of 21 million, block subsidies halving every 210,000 blocks. The current subsidy is 3.125 BTC. Pool fees range from 1% to 4%, settled in real time. SBI's exit removes approximately 0.07% of network hashrate from the accounting ledger.
That figure โ 0.07% โ is the most underweighted number in this entire dataset. A 16.222 EH/s pool in June was roughly 2.5% of a network running above 650 EH/s. Its complete disappearance shifts aggregate difficulty by a rounding error. The difficulty adjustment algorithm compensates silently within two epochs, roughly 2,016 blocks. Surviving miners see a marginal increase in expected block income; the adjustment does the bookkeeping. The tokenomic shock is negligible.
The signal, however, is not negligible. SBI's exit was a portfolio-level judgment about the return on mining capital. This is the kind of signal an institutional analyst should read carefully. A Japanese financial conglomerate with deep pockets and patient capital concluded that operating a Bitcoin mining pool does not clear its internal hurdle rate. That is a market verdict on mining infrastructure margins, rendered by a sophisticated actor with no ideological attachment to the sector.
The revenue model of pools is, at its base, sound: fees paid for a real service โ work coordination and block construction. There is no Ponzi dynamics here. But the margin structure has tightened to the point where mid-tier pools with high operating costs are no longer viable. This mirrors the pattern I flagged in my Convex Finance analysis during the 2021 bull market, where I argued that CRV emission misalignment threatened long-term sustainability. The market ignored the report for months; the liquidity crunch validated it in late 2021. The analog in mining is simpler and faster: when fee revenue compresses below operating cost, the operator exits. The market clears.
What SBI leaves behind is not a security hole. It is a balance-sheet signal. The pool's departure also raises the question of where the capital goes. In 2025, I reviewed an emerging AI-agent protocol and identified an oracle-data flaw that created an "AI-Oracle Attack Vector." The broader lesson applies here: capital leaving mining infrastructure is increasingly redeploying toward AI compute. SBI Group's broader portfolio โ exchange operations, securities, banking โ suggests the mining pool was not strategically sacred. It was a line item. When the line item underperformed, it was cut.
The Survivors' Market
The competitive ranking tells the same story from a different angle. Foundry USA commands 26.67% of attributed blocks, with institutional-caliber compliance infrastructure in the United States. AntPool holds 17.13%, leveraging its Asian channel relationships and multi-product financial services. F2Pool at 16.21% rounds out the triopoly. SBI falls from relevance to 0.72% to zero in a single quarterly window.
Below the top tier, the rankings are in flux. Luxor is rising on the strength of its data services and hashrate derivatives. Braiins is declining despite its open-source pool software. NeoPool is absent from the recent attribution window entirely. This churn in the long tail is the real competitive story. The middle of the market is being squeezed between the triopoly's scale advantages and the long tail's specialized offerings. There is no room in the middle.
My L2 scalability work taught me that consolidation narratives are usually more complex than they appear. When I benchmarked Optimistic against ZK-Rollup finality times in 2022, the institutional takeaway was that the race to become the standard was really a race to onboard the next hundred projects. The analog in mining pools is identical. Scalability is a trade-off, not a promise. The top three pools did not necessarily become more efficient. They became the default destination for orphaned hashrate. Cheap, fast aggregation โ not superior technology โ built the triopoly.
Every migration event reinforces the incumbents. When a mid-tier pool closes, its miners re-deploy to the largest pools first because those pools offer the deepest payout liquidity and the most reliable block-timestamp regularity. This is a network effect, and it is self-reinforcing. The chain is fast; the settlement is slow. The consolidation compounds over years of exits rather than weeks.
The Real Blind Spot: Block Templates and Capital Reallocation
Now the uncomfortable part. The entire public debate around pool concentration treats the 60% figure as the hazard. I think that is the wrong frame.
Consider what a pool actually controls. It does not control the consensus rules. It controls block templates โ the specific sets of transactions that propagate into a candidate block. A pool with 26.67% of attributed blocks controls 26.67% of the network's block-template policy. That policy includes whether the template is compatible with Ordinals inscriptions, whether BRC-20 transactions are prioritized, whether the pool runs Bitcoin Core's standard template or custom filtering.
I assign low confidence to the emerging hypothesis that block-template policy becomes a pool-selection criterion for miners. But the direction of travel is clear. As transaction fee revenue grows relative to subsidies โ a slow but inevitable shift โ block-template composition becomes a larger share of a pool's value proposition. The Ordinals wave demonstrated that inscription-heavy blocks can generate meaningful fee spikes. A pool that filters or prioritizes those transactions differently will produce differently distributed revenue for its miners.
The second blind spot is statistical. The 60.01% reading is real but temporally thin. It captures an instant in a sampled window. The sustained control level may sit a few points lower. This does not mean the triopoly is safe. It means the numeracy of the crisis discourse is sloppy. We are debating measurements as though they were unmediated facts โ a category error that a forensic reader should reject on principle.
Third, and most counter-intuitive: SBI's exit has not materially degraded Bitcoin's security model. The network's actual security derives from the combination of difficulty adjustment, the game-theoretic alignment of miners with block validity, and the cost of acquiring 51% of active hashpower. A pool's exit does not reduce hashpower; it re-routes it. The ASICs still compute. The difficulty still adjusts. Reorganization resistance is unchanged. The concentration risk is real โ but it existed before SBI closed its doors, and it will exist after the industry consolidates further.
Complexity hides risk; simplicity reveals it. The simple truth is that the triopoly crossed 60% before any exit event. The simple truth is that attribution methodology cannot see where the hashrate went. The simple truth is that pool economics, not protocol design, determined SBI's fate.
Risk Checklist
Based on the institutional due diligence framework I developed in 2024 โ the same checklist I applied when evaluating a modular blockchain protocol for a European fund, a call that saved that fund from a 60% drawdown after a sequencer outage โ here is the actionable assessment:
- Centralized sequencer/validator: flagged. Pool coordination concentrates block-template authority at the Stratum layer.
- Excessive administrative authority: flagged. Pool operators unilaterally determine transaction inclusion and payout settlement.
- Unaudited code: not applicable. No smart contracts in the affected layer.
- Technical complexity risk: not applicable. Stratum is a mature protocol.
- Peer-review deficiency: not flagged. Hashrate Index and mempool.space are professionally maintained data platforms.
The checklist confirms the risk vector: operational and structural, not cryptographic. Bitcoin's consensus layer remains sound. Its coordination layer is drifting.
The Structure Came First
The lesson from SBI's quiet exit is that mining consolidation is not an event. It is a process with a lag. The 2024 halving compressed pool margins. The post-halving rate environment raised the opportunity cost of capital. The 60% threshold was crossed months before any formal announcement. The question is not whether SBI's exit was survivable โ it was. The question is how many more SBI-sized operators are already below their internal hurdle rate, waiting for the right accounting quarter to exit.
When the next pool closes, do not read it as a security event. Read it as an economic signal. The exit reveals the structure that was already there โ the proof is in the buckets. July 20: 64.8%. July 27: 60.8%. SBI's last block: July 29. The structure came first. Trust the data, but audit the interpretation.