How does Bitcoin actually work? Not the whitepaper summary — the version you only learn by watching the thing nearly break. Bitcoin's circulating supply sits at 19.8 million out of a hard cap of 21 million, consensus is Proof-of-Work, launch year 2009, and the last all-time high printed at $109,000 on January 20, 2025. Those are the clean numbers. The dirty version — the one that explains how the protocol actually behaves under stress — lives inside five specific moments where the chain was tested in production. That is the version worth your time.
January 2009: The Genesis Block And The Headline Baked Into It
Block 0. The genesis block. Mined on January 3, 2009, with a coinbase transaction that nobody can spend because Satoshi hardcoded it that way in the reference client. The 50 BTC reward from block 0 is unspendable — not by consensus rule, but by implementation detail. That matters, and I will come back to why.
What people miss about the genesis block is the string embedded in its coinbase scriptSig: "The Times 03/Jan/2009 Chancellor on brink of second bailout for banks." This is not decoration. It is a timestamp proof. By quoting a newspaper headline from that specific morning, Satoshi proved the chain could not have been pre-mined before January 3, 2009. The headline is the receipt.
Here is how Bitcoin "works" at the base layer, revealed by this one block. The chain is a sequence of blocks. Each block references the hash of the previous one. The genesis block has no previous — its `prev_block_hash` field is zeroed out, 32 bytes of nothing. Every block mined since then descends from this root. Change block 0 and every subsequent block's hash chain breaks downstream. That is the immutability property, and it is not magic. It is just SHA-256 applied recursively.
The launched_year on this coin is 2009. The consensus is Proof-of-Work. The max supply is 21 million. The circulating supply today is 19.8 million. Every one of those numbers traces back to rules the genesis block set in motion — not by declaration, but by making the first entry in a ledger that nobody has been able to roll back since. Sixteen years and counting.
August 2010: The Value Overflow Incident And The Hard Fork Nobody Voted On
On August 15, 2010, block 74,638 was mined containing a transaction that created 184,467,440,737.09551616 BTC out of thin air. That number is not a typo. It is roughly 184 billion Bitcoin, generated in a single transaction, on a chain whose entire max supply is capped at 21 million. The exploit abused an integer overflow in the output-value validation code. Two outputs, each worth about 92 billion BTC, summed in a way that wrapped around the 64-bit integer check and passed as valid.
The chain accepted the block. For about five hours, the longest valid chain by Proof-of-Work included a transaction that violated the economic cap that defines what Bitcoin is. If you were running a node at hour three and you queried the UTXO set, you would have seen 184 billion coins. Hear me out — this is the moment that disproves the lazy framing that "Bitcoin is just math." Math did not save the chain. People did. Specifically, Satoshi and Gavin Andresen pushed a patched client (0.3.10) within hours, the community coordinated on IRC, and miners agreed to abandon the chain containing block 74,638 and continue from block 74,637 with the fix applied.
That is a hard fork. Not voted on, not announced through governance process, just coordinated under duress because the alternative was Bitcoin ceasing to be Bitcoin. The 184 billion coins stopped existing because the social layer agreed they would stop existing. If you want to understand how Bitcoin actually works, you have to accept this: the protocol is enforced by nodes, nodes are run by humans, and humans retain the ability to coordinate out of an invalid state. The cap is 21 million because people keep running the software that says so.
March 2013: The Chain Split That Proved Consensus Is Software, Not Math
March 11, 2013. Block 225,430. Bitcoin version 0.8 had just been released, built on a new database backend (LevelDB) that handled large blocks differently from the older Berkeley DB used by 0.7 nodes. A miner running 0.8 produced a block that was valid under 0.8's rules but failed to validate under 0.7's — specifically because 0.7 hit a BDB lock limit when processing it.
The result: half the network accepted block 225,430 and built on it. The other half rejected it and continued the chain from 225,429. For about six hours, Bitcoin had two canonical histories running in parallel. Any transaction confirmed on one side was not confirmed on the other. Any merchant or exchange that assumed finality after one confirmation during those six hours was wrong on roughly half of their counterparties. Mt. Gox paused deposits. BTC-e paused deposits. OKcoin had not launched yet.
Resolution came through social coordination again. Core developers posted on bitcoin-dev and in #bitcoin-dev IRC asking miners running 0.8 to switch back to 0.7 and abandon the 0.8-only chain, even though the 0.8-only chain had more Proof-of-Work accumulated. The miners complied. The chain reorganized. The 0.8-only side lost roughly 24 blocks of work.
Here is the lesson that custody desks took away, and that I think retail traders still underprice. "Consensus" is not a property of the Nakamoto algorithm. Consensus is a property of what software the economic majority of nodes is running. When those implementations diverge on edge cases — database limits, memory thresholds, parser corner cases — the chain can split along software-version lines rather than Proof-of-Work lines. The lesson got written into every serious custodian's deposit-crediting policy afterward: never credit on one confirmation during a client upgrade window.
August 2017: The SegWit Activation And The Fee Market Nobody Designed
August 24, 2017. Block 481,824. Segregated Witness activated via a soft fork, after roughly three years of acrimony that broke the development team in half and spawned Bitcoin Cash as a competing chain the week before.
The engineering substance of SegWit is this: signatures (the "witness" data) got moved outside the base transaction structure. Legacy nodes still see transactions as valid; upgraded nodes verify the witness. Because the block weight calculation discounts witness data to 25% of its byte count, SegWit effectively increased block capacity without changing the 1MB base block size — a block weight ceiling of 4 million weight units replaced the raw byte cap. That is the trick. That is why the block size "debate" was never really about one megabyte versus two.
But the production lesson from August 2017 was not the block weight math. It was the fee market. Before SegWit, during the activation stretch of mid-2017, the Bitcoin mempool regularly carried over 200,000 unconfirmed transactions and median fees crossed $30 per transaction. For a protocol designed around "peer-to-peer electronic cash," paying $30 to send $400 broke the use case. Users learned that Bitcoin's fee market is a priority auction where the next block clears roughly 2,000 transactions and the top bidders win. If you are not in the top 2,000 bidders in the next 10 minutes, you wait. If you wait long enough and the backlog grows, your fee becomes stale and you wait more.
Nobody designed this fee market. It emerged from the block space constraint hitting demand. SegWit relieved it temporarily — mempool drained within weeks of activation. But the structural lesson stayed: Bitcoin's throughput ceiling is architectural, and the "fees go up during congestion" dynamic is not a bug to be fixed. It is the mechanism by which the network allocates its scarce resource. Custody desks that batch withdrawals learned to time them around mempool depth. Retail users learned the hard way.
January 2025: The All-Time High At $109,000 And What The Custody Desks Did Quietly
The ath_date for Bitcoin is January 20, 2025. The ath_usd is $109,000. The market_cap_usd_b at today's $64,349 price sits at $1,290.97 billion against a circulating supply of 19.8 million BTC. Those are the clean numbers from the grounding. The dirty version of what happened around that peak is less reported.
What qualified custodians did during the climb to $109,000 — the ones registered under NY DFS Trust Company charters like Coinbase Custody and Fidelity Digital Assets, and the OCC Federal Trust Charter held by Anchorage Digital — was execute a quiet choreography of cold storage migrations. When institutional allocations grow faster than custody infrastructure, you have two choices: expand hot-wallet exposure to handle withdrawal flow, or stage multisig signing ceremonies to move more coins into deep cold while keeping a hot float for operational liquidity. The second option is what prudent custody does. It is also what nobody tweets about.
Here is what Bitcoin "working" looks like at this scale. A qualified custodian holding, say, $20 billion in BTC on behalf of ETF issuers and family offices does not keep that in a single wallet. It is sharded across cold storage vaults, each requiring multi-party signing — typically 2-of-3 or 3-of-5 multisig schemes with keys held in geographically separated HSMs. The signing ceremony to move coins from cold to warm takes hours, requires multiple officers, and gets logged with timestamp receipts for the auditor.
The hardware wallet companies — Ledger (Paris), Trezor (SatoshiLabs, Czech), GridPlus Lattice1 — serve the retail and small-institution end of this same architecture. The firmware audit histories matter here because a signed firmware update that leaks seed material is the single highest-consequence attack surface on a cold-storage setup. Trezor's firmware is open source and reproducibly built. Ledger's firmware is closed, which is a design choice I will not litigate here but will name. GridPlus runs a co-signer abstraction that lets you require a second signature from a GridPlus-hosted key for policy enforcement.
When BTC printed $109,000 on January 20, 2025, the measure of whether Bitcoin "worked" was not the price. It was whether the custody layer handled the inflow without a single publicly reported seed compromise, bridge exploit, or exchange insolvency. By that measure, the ATH print was uneventful. Which is the highest compliment you can pay a financial system.
What It All Means
Five events, sixteen years. The pattern is consistent. Bitcoin does not "work" in the way the whitepaper describes — a trustless, mathematically perfect ledger that runs itself. Bitcoin works the way any sufficiently complex production system works: it has bugs, it has social-layer coordination mechanisms, it has operational practices built up around it by the people who hold other people's money on it. The protocol is a floor, not a ceiling.
What the timeline teaches is that every major Bitcoin failure mode has been patched not by the protocol fixing itself, but by nodes and operators coordinating to run compatible software. Max supply is 21 million because people run clients that enforce it. Confirmation depth is a safety heuristic because software versions diverge. Fees are a priority auction because block space is finite and demand is not. Cold storage practice is multisig because single-key compromise is a single point of failure that institutional underwriting cannot accept.
If you take one thing from this piece, take this. "How Bitcoin works" is not a question answered by the whitepaper or by any single architectural diagram. It is answered by the operational discipline of the people running nodes, signing cold storage transactions, and deciding — in moments of crisis — which chain is the real one. The protocol gives you the rails. The humans decide where the train stops.
None of this tells you whether the next stress event will be resolved the same way. The custody landscape of 2026 looks nothing like 2010's IRC-coordinated patch deployment, and the next time the chain splits — because of a client bug, a quantum milestone, or something I cannot name yet — the resolution will run through different institutions with different incentives. That question is where the real work starts, and it is not where this piece ends.
FAQ
If Bitcoin's supply is capped at 21 million, why does the circulating supply read 19.8 million in 2026?
The 21 million cap is the asymptotic maximum emitted through block rewards, which halve roughly every four years. As of 2026 the circulating supply sits at 19.8 million BTC, representing roughly 94% of eventual max supply. The remaining 1.2 million gets emitted over the next century via progressively smaller block rewards. Note that circulating supply does not account for provably lost coins — the genesis block's 50 BTC, early-era wallets with lost keys, and burned addresses are all included in the circulating figure but functionally unreachable.
What actually happens during a Bitcoin "hard fork" like the one in August 2010?
A hard fork is a change to consensus rules that makes previously invalid blocks valid, or vice versa — forcing every node to upgrade or get stranded on a dead chain. In August 2010, Satoshi released client version 0.3.10 patching the integer-overflow bug, and miners agreed to abandon the chain containing the 184 billion BTC block. The social coordination layer — not any protocol mechanism — chose which chain became "Bitcoin." Any node that failed to upgrade kept following the invalid chain into irrelevance.
Is self-custody actually safer than a qualified custodian like Coinbase Custody or Fidelity Digital Assets?
It depends on your threat model and your operational discipline. A qualified custodian carries NY DFS or OCC regulatory oversight, insurance, and multisig cold-storage infrastructure most individuals cannot replicate — Coinbase Custody holds a NY DFS Trust Company charter, Anchorage Digital holds an OCC Federal Trust Charter. Self-custody with a Ledger or Trezor removes counterparty risk but introduces operational risk you own entirely: seed phrase loss, firmware compromise, signing errors. The failure modes are different, not strictly ordered.
What is the difference between Ledger, Trezor, and GridPlus Lattice1 at the firmware level?
Trezor publishes open-source firmware with reproducible builds — you can verify the binary on your device matches the published source. Ledger ships closed-source firmware; you trust the vendor's attestation chain. GridPlus Lattice1 adds a co-signer abstraction: the device enforces policies by requiring a second signature from a GridPlus-hosted key before high-value transactions execute. For custody at scale, firmware audit history and update-signing architecture matter more than the hardware specs printed on the box.
Why do Bitcoin transaction fees spike so hard during congestion events?
Each block clears a fixed amount of transaction weight — up to 4 million weight units post-SegWit, which translates to roughly 2,000 to 4,000 transactions per block depending on their composition. When demand exceeds that throughput, miners prioritize by fee rate (satoshis per virtual byte). Your transaction either outbids the next 2,000 competing transactions or waits. During peak mid-2017 congestion before SegWit activation, median fees crossed $30 and the mempool held over 200,000 pending transactions. The spike is not a bug — it is how the network allocates scarce block space.
How does Proof-of-Work actually secure the chain against someone rewriting history?
Rewriting a confirmed block requires re-mining it and every block after it faster than the honest network produces new blocks. At current hashrate, re-mining even a one-week-deep block requires sustained hashpower exceeding the entire honest network's capacity for that same duration — a cost measured in hundreds of millions of dollars in electricity and ASIC depreciation. The security scales with how deep the block is buried. One confirmation is weak. Six is the convention. Custody desks typically require more for high-value credits, especially during known software upgrade windows.
What did the January 2025 all-time high at $109,000 actually change about how custody works?
Operationally, less than you would think. The $109,000 print on January 20, 2025 was absorbed by custody infrastructure that had been maturing since the 2020-2021 cycle: multisig cold storage with geographically separated HSMs, qualified custodian trust charters, auditor-logged signing ceremonies. The visible metric was price. The invisible metric was that no major custodian reported a seed compromise, no bridge exploit cleared, and no exchange paused withdrawals around the peak. That silence is the proof the architecture worked — the opposite of 2010, 2013, and 2017, when stress produced visible failures.