Investing in Bitcoin

On the evening of April 19, 2024, US time (a few minutes into April 20 in UTC), at block 840,000, the reward paid to whoever mined the next block of bitcoin transactions fell from 6.25 coins to 3.125. Nothing was announced. No committee met. The rule had been in the software since January 2009, and the network executed it the way it executes everything, by counting to a number. The price that day, on Coin Metrics’ daily close, was $63,763.

Every previous halving had been followed, within twelve to eighteen months, by the kind of advance that makes people quit their jobs: 92 times the halving-day price after 2012, 30 times after 2016, 8.5 times after 2020. This time the peak came 535 days later, on October 6, 2025, at $124,824, exactly twice the halving-day price. Then the asset spent the next nine months doing what it has done in every cycle of its life, giving most of the advance back. By July 1, 2026 it was at $58,566 on our tape, below where it had traded on the day the reward halved.

That is the whole shape of the thing you are considering buying. The supply side is a solved equation, public to the block. The demand side is people, and people behave the way they always have. This guide puts the two halves in front of you with the numbers attached.

One fact should frame every other number in this guide: the buyer base of bitcoin was transformed between 2024 and 2025, and the return distribution was not.

On January 11, 2024, eleven spot bitcoin ETFs began trading in the United States with the coin at $46,381. BlackRock’s iShares Bitcoin Trust, IBIT, reached $80 billion of assets within 374 days, the fastest any ETF had reached that size as The Block and CoinDesk reported in July 2025, and by the price peak on October 6, 2025 the group held $169.5 billion, 6–7% of every coin in existence, on cumulative net inflows of about $61.5 billion on Farside Investors’ daily tallies. Vanguard opened its platform to the funds on December 2, 2025, Harvard’s endowment reported $443 million of IBIT at September 30, 2025, and one Virginia software company had accumulated more than 4% of the eventual supply. If you had written the institutional-adoption thesis in 2020, you could not have asked for more.

Then the price fell 53% anyway, and the flows followed it out: the ETFs were net sellers for four straight months after the peak, and cumulative net inflows had shrunk to $54.9 billion by August 27, 2026 (section 9 has the month-by-month record). The wrapper turned out to be a distribution channel, not a floor. On this hub the flagship Investing in Crypto guide covers the whole asset class, Investing in Ethereum the second coin, and Crypto ETFs and Treasury Companies the wrappers in depth; this guide stays with bitcoin, what it is, what it has returned, and how to hold it.

Bitcoin’s supply schedule is the one property of the asset that is not a matter of opinion. The software pays new coins to the miner who adds each block of transactions to the chain, roughly every ten minutes; the reward began at 50 coins a block in January 2009 and halves every 210,000 blocks, about four years, so the series converges on 21 million, with the last fraction of a coin due around 2140. No board can vote to change this. The rule is enforced by every computer running the software, and a version that printed more would be a different network that nobody’s coins were on.

By the fourth halving on April 19, 2024, 19,687,293 coins had been issued, 93.75% of the total; by May 23, 2026 it was 20,033,382, or 95.4%, on Coin Metrics’ supply series, and about 20.08 million, 95.6%, on public trackers at the end of August 2026. Coin Metrics’ ten-year forward projection on May 23, 2026 was 20,838,420 coins, so only about 805,000 new coins arrive in the whole decade to 2036. Annual supply growth was 3.8% in the year to the May 2020 halving, 1.65% in the year to May 2024 and 0.83% in the year to May 2026, and it falls to roughly 0.4% after the next halving, due at block 1,050,000 around April 2028, when the reward drops to 1.5625 coins. Gold’s above-ground stock grows about 1.5–2% a year from mine supply. That is the entire basis of the digital-gold argument, and it is a fact rather than a slogan.

What the cap does not mean

Two corrections the marketing leaves out. First, 21 million is the ceiling on units issued, not on units available to sell. Chainalysis has estimated that 2.3–3.7 million coins are lost for good, most from the early years when a coin was worth cents, and roughly 1.1 million further coins attributed to the creator’s own mining pattern in Sergio Demian Lerner’s 2013 analysis (later estimates run from about 700,000 to 1.1 million) have never moved; the effective float is nearer 16–17 million and shrinks every time an executor cannot find a seed phrase. Second, a fixed supply says nothing about demand, the only variable that has ever set the price: the schedule was exactly as fixed in November 2022, at $15,758, as in October 2025 at $124,824. Scarcity is necessary for the asset’s case. It has never been sufficient.

The pattern that made bitcoin’s cycle famous broke in 2025, and you should see exactly how. The halvings, dated by Coin Metrics’ block count and priced at that day’s close, were November 28, 2012 at block 210,000 ($12.33), July 9, 2016 at block 420,000 ($651.94), May 11, 2020 at block 630,000 ($8,592) and April 19, 2024 at block 840,000 ($63,763). The theory that grew up around them was simple: halve the flow of new coins, hold demand constant, and the price must rise until the market clears. For three cycles it looked like physics.

Read the chart from the top. The first epoch multiplied the halving price 92 times in 371 days, the second 30 times in 525 days. The third managed 8.5 times, and even that needed the ETF launch to print its final high in March 2024, four years after the halving; the November 2021 peak most people remember was 7.9 times. The fourth peaked at 2.0 times, 535 days out. One year after each halving the price stood at +8,069%, +284%, +559% and, after the 2024 event, +34%. The decay is arithmetic, not sentiment: each halving removes half of a flow that is already a smaller share of the outstanding stock, and against $1.5 trillion of market value and about $100 billion of ETF holdings, the difference between 450 and 225 new coins a day is a rounding error in the order book.

What broke in 2025

The cycle model said the year after a halving is the blow-off and the year after that is the crash: 2013, 2017 and 2021 were the up years, each closing at least 59% higher. Calendar 2025 closed at $87,517 against $93,390 at the end of 2024, −6.3%, the first post-halving year in bitcoin’s history to fall, and from a peak the on-chain data never flagged as euphoric: the market-value-to-realized-value ratio, defined in the watch list, was 2.29 at the top against 4.4–4.7 in 2013 and 2017. The honest reading is that the halving is now a calendar event, not a catalyst. What remains of the cycle is the leverage cycle inside crypto and global liquidity, neither of which runs on a four-year clock.

What secures the ledger is electricity, and the shrinking reward that pays for it is the most serious open question in the asset’s design. The right to add a block, and collect the reward, goes to whoever finds a valid hash first, which in practice means whoever burns the most power guessing. That total guessing power, the hash rate, is measured in exahashes per second (EH/s, a billion billion guesses) and is the cleanest measure of real capital committed to the system, because the machines are useless for anything else and take years to pay back.

It has multiplied about thirty times in eight years, from 31 EH/s in May 2018 to 986 EH/s in May 2026 on Coin Metrics’ data, and it kept rising through most of the drawdown: the thirty-day average peaked at 1,111 EH/s on November 11, 2025 and had eased to 977 by May 23, 2026, with public seven-day trackers near 934 EH/s on September 6, 2026, roughly 16% off the peak. CoinShares’ first-quarter 2026 mining report recorded the first first-quarter decline in network hash rate since 2020.

The security budget

Miners are paid in two ways, the block subsidy and the fees users attach to get into a block, and the sum is the security budget: what an attacker would have to out-spend to rewrite history. At 3.125 coins a block and about 144 blocks a day the subsidy is roughly 450 coins, about $35 million at September 2026 prices. The design assumes fees will replace the subsidy as it halves away; on the data they are doing the opposite. Fees averaged 275 coins a day in 2017 (12.6% of miner revenue), 41.5 a day in 2024 (6.5%, lifted by the Runes token mania around the halving), then 4.7 coins a day in 2025 and 2.6 a day in the first five months of 2026, 1.0% and 0.6% of revenue. On May 20, 2026 the network’s users paid its miners 2.5 bitcoin for the day’s security.

Follow that arithmetic forward. If fees stay where they are, each halving cuts the security budget nearly in half in coin terms, and the network is relying on the dollar price doubling every four years to stand still; so far it has. The 2028 halving takes the subsidy to about 225 coins a day, $17.7 million at today’s price for a $1.5 trillion asset, still enormous against any attacker’s budget and still falling on a public schedule. Nobody has solved this. It is the one line in bitcoin’s design an honest holder should re-read every four years. Investing in Bitcoin Ordinals works the same arithmetic from the other side, asking what fee demand would have to be for fees alone to pay for the network, and answering it with the fee record the 2023–24 inscription wave actually produced.

Hashprice, daily revenue per petahash of computing power, is the industry’s profit and loss in one number, and Luxor’s Hashrate Index put the three lowest monthly averages in its history in 2026 ($31.27 in March, $30.37 in June, $31.21 in July). At those levels an operation paying more than about seven cents a kilowatt-hour loses money on most hardware generations, and older S19-class machines need power below about five and a half cents to break even, on 2026 industry estimates. The rest sell coins or turn their sites into landlords for artificial-intelligence computing, a pivot CoinShares tallied at more than $70 billion of contracts across 2025 and early 2026; the listed miners are therefore no longer a clean way to own bitcoin, and every megawatt that leaves mining for AI is one the security budget no longer buys.

Here is the return history in the form that does not let anyone choose the start date, and then what the choice of start date does to the number you will be sold. The figure the industry quotes is true: from $429.68 at the end of 2015 to $87,517 at the end of 2025, bitcoin compounded at about 70% a year for a decade, against about 15% a year for the S&P 500 with dividends over the same ten calendar years (S&P Dow Jones Indices data via SlickCharts).

The same series, drawn from Coin Metrics’ daily closes, December 31 to December 31, also shows five losing years out of the last twelve: −56.0% in 2014, −73.5% in 2018, −64.4% in 2022, −6.3% in 2025, and 2026 running at about −11% on our tape to September 8. The up years are +125.5% (2016), +1,337% (2017), +94.4% (2019), +304.9% (2020), +59.7% (2021), +155.5% (2023) and +121.2% (2024). Not one calendar year since 2011 has landed between 0% and +30%, the range where an equity index spends most of its life; over the same twelve years the S&P 500’s total return was negative twice (2018 and 2022) and its best year was about 31%.

Holding period is everything

The more useful way to read the record is to ask what happened to every investor who bought on any day since the start of 2013 and held for a fixed period. We ran that calculation across Coin Metrics’ daily closes to May 23, 2026: 4,527 one-year windows, 3,432 four-year windows.

Three things fall out of the chart. Nearly three in ten one-year holders since 2013 have lost money, and the worst of them lost 84%, so anyone who tells you a year is a long time in this asset is selling something. The four-year window has never lost, with a worst case of +32% for a buyer at the April 2021 top, and its median outcome, +921%, is the number that built the “just hold” faith. The five-year window is negative in 8 of 3,067 windows, worst case −14%, all for buyers within a few weeks of the December 2017 top who sold into the FTX winter of December 2022; the row is there to show you that “never” has a start date. Every one of these windows is drawn from a period in which the asset went from $13 to $124,824. A buyer in September 2026 is buying a $1.5 trillion asset with a 2.0× epoch behind it, and the arithmetic of the next four years is not the arithmetic of 2013.

Against equities, honestly

Over ten years to 2025 bitcoin beat the S&P 500 by a distance no risk adjustment closes, but at 68% annualised volatility across 2015–2025, roughly four times an equity index, its Sharpe ratio for the decade works out at roughly 1.0 against the S&P’s 0.7–0.9 (both our approximations from annual returns): a good asset per unit of risk, not a miraculous one. Over the twelve months to September 8, 2026 the comparison inverts: bitcoin −29.3% on our tape, the S&P 500 up about 12.5% on the monthly-average Shiller series from December 2025 to August 2026, gold +22.1%. Every trailing comparison in this market is a choice of window.

Bitcoin has lost half or more of its value five times. The number that matters most is not the depth but how long each took to regain the old high, because that is what decides whether you will still be holding when it matters.

The first ran from $29.03 on June 8, 2011 to $2.11 on November 18, 92.7% in 163 days after the first Mt. Gox hack, and regained its high on February 19, 2013. The second, from $1,134.93 on December 4, 2013 to $175.64 on January 14, 2015, took 406 days to bottom through the collapse of Mt. Gox and 1,177 days from peak to recovery, the longest wait in the record. The third, from $19,640.51 on December 16, 2017 to $3,185.07 on December 15, 2018, bottomed in 364 days and recovered on November 30, 2020. The fourth, from $67,541.76 on November 8, 2021 to $15,758.29 on November 9, 2022, bottomed in 366 days through Terra, Celsius and FTX and recovered on March 4, 2024, with the ETF launch as the final push.

The pattern is worth stating plainly. The bottom arrives about a year after the top; it is marked by a failure inside the industry rather than a macro event; and the round trip to the old high takes two to three years from the trough. Each of those bear markets was survivable for an investor who had sized the position to be ignored, and fatal for one who had borrowed against it.

The fifth drawdown

From $124,824 on October 6, 2025 the coin fell to $63,495 by February 5, 2026 on Coin Metrics’ closes, 49.1% in 122 days, and then, after a spring bounce, to $58,566 on July 1, 2026 on our tape: −53.1% against the Coin Metrics peak, the shallowest of the five and one of the fastest to its low. The optimistic reading is that a market ten times larger, with a standing ETF bid, has halved the amplitude. The cautious reading is that the 2021 cycle also had a 53.1% mid-cycle drawdown, from $63,446 on April 13 to $29,767 on July 20, 2021, that was not the bottom, and that the recovery to $78,576 since July is the size of the mid-2022 rally that preceded the FTX leg. Both readings fit the data; the record says the difference between them is another 40% of your money.

Volatility and correlation determine how bitcoin behaves inside a portfolio, and both have changed direction since 2020 in ways the sales material does not describe. Bitcoin’s annualised volatility, computed from Coin Metrics’ daily closes, was 149% in 2013, 93% in 2017, 78% in 2020, then 43% in 2023, 52% in 2024, 42% in 2025 and 50% in the first five months of 2026. The asset is calmer than it was, in the sense that a 60%-volatility year is now unusual, and still three to four times an equity index.

The tails are where that matters: there were 21 days in 2017 and 20 in 2018 when the price moved more than 10%, none in 2023 or 2025, and two in 2026, both in the first week of February, when bitcoin fell 14.1% on the 6th and rose 11.9% on the 7th. The worst single day in the record remains March 12, 2020, −37.5%.

Correlation went the wrong way

The case for bitcoin as a diversifier rests on low correlation with the things you already own, and that case was strongest before anyone made it. Using monthly-average prices (the Shiller S&P 500 series and Coin Metrics’ daily closes averaged by month, with our tape for June to August 2026), the correlation of monthly returns between bitcoin and the S&P 500 was 0.27 across 2014–2019 (71 monthly returns) and 0.47 from January 2020 to August 2026 (79); over the 2025–2026 stretch alone it is 0.55 (19). Bitcoin sold off with technology stocks in the 2022 rate shock, rallied with them through 2023–2024, fell 23% between Inauguration Day and April 7, 2025 while the tariff announcements took the S&P into a 19% drawdown, and fell 6.5% close to close on October 10, 2025 (about 14% from the day’s high to its low) when a tariff post hit a leveraged market. Monthly averages smooth the daily noise, so treat the levels as indicative; the direction is not in doubt.

On our own tape, which records CoinGecko’s daily price, bitcoin is down 29.3% over the twelve months to September 8, 2026, from $111,133 to $78,576, while our gold series rose 22.1%; the daily-return correlation between the two across 199 common trading days is 0.09, and that is not the good news it sounds like, because zero correlation with gold means bitcoin did not hedge what gold hedged. Index both to 100 on October 6, 2025, the day bitcoin peaked, using our tape for both: on July 1, 2026 gold stood at 103 and bitcoin at 47; on September 8, gold at 112.5 and bitcoin at 63.6. Whatever bitcoin is, it did not trade like gold in the year gold mattered most. It is a liquid, 24-hour, high-beta risk asset with a fixed supply. Own it where you keep equities and venture, and size it against those.

Our tape is Invest Alternative’s own daily collection of CoinGecko’s bitcoin price, 374 observations from August 29, 2025 to September 8, 2026: our record of one vendor’s daily print, not a market-wide index. It turns the drawdown above into a sequence of dated prices. On it, the peak was $124,740 on October 7, 2025 (Coin Metrics’ October 6 close was $124,824); the low was $58,566 on July 1, 2026, 267 days later, a 53.0% fall; and the September 8 print of $78,576 is 34.2% above the low and 37.0% below the peak.

The shape is a staircase down with two landings. October 2025 gave back the peak within four days of setting it, the cascade of October 10–11 (about $19 billion of leveraged positions liquidated on Coinglass’s count) taking the tape from $121,719 on the 10th to $108,409 at month end. November was the worst month, −16%, as the ETFs turned to net selling; February printed the tape’s worst and best days back to back; April was the first landing, June’s low the second, and the August rally is where the tape stands. Of its 374 daily prints, 123 are below $70,000 and only five below $60,000.

What averaging in would have done

The tape lets us test the standard advice. Twelve equal purchases of $2,083.33 on the 8th of each month from September 8, 2025 to August 8, 2026, at prices from $121,463 (October) down to $63,274 (June), would have bought at an average of $79,110 and been worth $24,831 on September 8, 2026: −0.7% on $25,000, against −29.3% for a lump sum on the first date and −37.0% for one bought at the tape peak. Averaging did not make money; it turned a year that would have ended a lump-sum buyer’s interest in the asset into one that was survivable, and on our tape that is the whole of what it did.

A spot bitcoin ETF is a grantor trust that holds coins with a custodian (Coinbase Custody for most issuers) and issues shares against them; a share of IBIT is a fractional claim on the trust’s bitcoin and tracks the coin to within a basis point or two during US hours, because authorised participants arbitrage any gap. The SEC’s July 29, 2025 order allowing in-kind creations and redemptions made the mechanism cheaper and more tax-efficient; options on IBIT, listed on Nasdaq since November 19, 2024, made it the most liquid bitcoin instrument in the world. You buy it commission-free at any major broker, and the cost is the sponsor fee, deducted inside the fund by selling a sliver of coin.

The fee differences are small in dollars: on $25,000, IBIT costs $62.50 a year, Grayscale’s mini trust $37.50 and Morgan Stanley’s MSBT, the cheapest since its April 8, 2026 launch, $35, and IBIT’s liquidity (the tightest spread and the deepest options book) is worth that to anyone who will ever trade it. What you should not own is the legacy GBTC at 1.50%, which exists because its holders would owe tax to leave it.

The flows, 2024–2026

The ETFs took in roughly $35 billion net in their first calendar year on Farside Investors’ tallies and about $61.5 billion cumulatively by the October 6, 2025 peak, when group assets reached $169.5 billion. Then: about $3.5 billion out in November 2025, $1.1 billion in December, $1.6 billion in January 2026, $0.2 billion in February, a $1.3 billion inflow in March, and $4.1 billion out in June into the July low, the worst month since launch on CoinDesk’s count of Farside’s tables (some tallies put June at $4.5 billion); IBIT’s net assets fell from $67.4 billion at the end of 2025 to $53.4 billion at March 31, 2026 on its 10-Q. August 2026 brought the turn, about $3.5 billion in, cutting the year-to-date outflow to $1.8 billion and putting group assets at $99.6 billion and IBIT at $61.4 billion on September 1; cumulative net inflow since launch was $54.9 billion on August 27, 2026, so the ETF investor as a class had given back about $6.6 billion of what it put in by the top.

The record daily inflow ($1.4 billion, November 7, 2024) came the week after the election; the record outflow ($1.1 billion, February 25, 2025) came as the price broke. One caution about reading flows as adoption: a meaningful share of 2024–2025 inflows was the long leg of the basis trade, a hedge fund buying the ETF and selling CME futures at a premium to lock in a dollar yield, which unwinds when the premium closes. Read the flows against CME open interest, never alone.

Tax and the IRA

Because the trust is a grantor trust, you are taxed as if you held the coins: long-term capital gains rates after a year, no corporate layer, and a small stream of taxable events each year as the trust sells fractions of a coin to pay its fee, reported on your broker’s ordinary 1099-B. Two things differ from holding coins directly. The wash-sale rule applies to the ETF, because it is a security, so a loss harvested in a drawdown means thirty days out of the market or a switch to another issuer’s fund.

And the ETF can live inside a traditional or Roth IRA at any major brokerage, where none of the above applies until withdrawal, or ever in the Roth case; the 2026 contribution limit is $7,500, or $8,600 at fifty and over, under the IRS’s November 2025 announcement, so a $25,000 IRA position is a rollover or several years of contributions. Fidelity’s crypto IRA, launched in April 2025, holds coins directly at a 1% spread per trade and no account fee; against a 0.25% ETF it makes sense only for someone who wants the coins themselves in a tax-deferred wrapper.

Strategy and the treasury companies

The other listed route into bitcoin is a company that holds it. Strategy, formerly MicroStrategy, held 845,050 coins bought for $63.73 billion, about $75,400 each, in its 8-K for the week to August 31, 2026, financed by equity, about $8 billion of convertible notes and perpetual preferred shares paying 8–10% or more. The trade worked while the stock traded above the value of its coins, the multiple called mNAV; that premium was above 2.5× in November 2024 and below 1.0× on an enterprise basis by late June 2026, at which point the company sold coins for the first time since 2022: 32 in the last week of May 2026, then 3,588 for about $216 million between June 29 and July 5, to pay preferred dividends and refill its dollar reserve. MSTR closed at $137.57 on September 8, 2026, about 75% below its intraday peak of $543 on November 21, 2024 and 71% below its record close of $473.83 the day before, over a period in which the coin fell about 17% to 20% depending on the day taken. A treasury stock is a bet on other people paying more than a dollar for a dollar of bitcoin. For bitcoin exposure, own bitcoin.

The exchange route is the one most first-time buyers take, and an exchange balance is not what most of them think it is. Coinbase is the regulated US on-ramp, Nasdaq-listed and an S&P 500 member since May 19, 2025; Kraken is the second US venue, its IPO pushed on September 2, 2026 to the second quarter of 2027 at the earliest; Fidelity and Robinhood also sell coins.

The fee schedules, checked on September 9, 2026, are where the industry relies on you not reading. Coinbase’s simple buy button charges roughly 1.5% on a $25,000 purchase plus a spread of about 0.5%. Its Advanced interface, a real order book, charges 1.20% taker and 0.60% maker at the entry tier (under $1,000 of trailing thirty-day volume, where a first purchase lands), falling to 0.40% and 0.25% between $10,000 and $50,000 of monthly volume; a taker order takes the price on offer, a maker order posts a limit and waits. Kraken Pro rebuilt its tiers on July 9, 2026 into a single cross-platform schedule and now charges 0.80% taker and 0.40% maker at the entry tier, roughly double its previous entry rate — a reminder that an exchange schedule is a fact about a date, not a constant. Fidelity Crypto quotes a 1% spread. Card and “instant” purchases run 3–4%.

The arithmetic on $25,000: $500 through Coinbase’s simple button, $300 for a market order on Coinbase Advanced, $150 for a limit order there, $200 for a market order on Kraken Pro, $100 for a limit order on Kraken Pro, $250 through Fidelity’s spread, against $62.50 plus a few dollars of spread for the ETF’s first year. The exchange route is competitive only if you post limit orders on the cheaper venue and then hold for years, which is the route the worked example prices.

What you own on an exchange

Coins bought on an exchange sit in its omnibus wallet. Your ownership is a line in the company’s database, and your recourse if it fails is the bankruptcy court; Coinbase segregates customer assets and is overseen by New York and the SEC, but its own filings say customers could be treated as unsecured creditors in an insolvency. The record: Mt. Gox lost about 850,000 coins in February 2014 and is still distributing what was recovered, with the deadline extended to October 31, 2026; FTX filed on November 11, 2022 and has since paid customers 118–143% of their claims in dollars fixed at the November 2022 price of about $16,900, so that a customer with one bitcoin there received about $20,000 for it. Even Coinbase disclosed on May 15, 2025 that support contractors had been bribed to hand over customer records, home addresses included. Nobody breaks the cryptography; they break the company.

The rule that follows is simple. An exchange is for buying, not for storing. The moment your balance there is more than you would leave in a checking account, it should be in the ETF or in your own custody.

A bitcoin is spendable by whoever holds the private key that controls its address, and self-custody means holding that key on a device you control rather than in a company’s database. The standard kit is a hardware wallet, a small signing device that never exposes the key to an internet-connected computer, plus the twelve or twenty-four seed words from which the key can be regenerated, engraved on steel and stored away from the device. At 2026 list prices the devices are cheap: Ledger’s Nano S Plus about $79, Trezor’s Safe 3 $59–79 (cut from $79 in 2026), Trezor’s Safe 5 $129–169, Coinkite’s Coldcard Mk4 about $178, BitBox02 about $150; a steel seed plate is $50–100 and the withdrawal from an exchange costs a network fee of a few dollars. After that, holding costs nothing for as long as you hold, which is the honest financial case for the route over an ETF that charges 0.25% every year.

Where the kit fails

The operational risks are the ones that do not appear on a fee schedule: a seed on paper lost in a move, a phishing site that asks you to “verify” your words, a counterfeit device from a marketplace, a bad firmware update. The last stopped being theoretical on July 31, 2026, when Coinkite disclosed that a 2021 firmware change had left seeds generated on affected Coldcard versions with about 72 bits of randomness instead of 128 (about 40 bits on the older Mk3), and that attackers had swept on the order of a thousand bitcoin from affected wallets before the advisory; a firmware update does not fix a seed already generated, only moving the coins does.

A single device and a single seed is a single point of failure, and the industry’s answer for any serious sum is multisignature: two or three of several keys required to spend, held on different devices in different places, so that no one device, location or person can lose the funds.

Collaborative-custody services build the multisig for you and hold one of the keys, so that neither you nor they can move coins alone: Casa charges $250 a year for its Standard three-key plan and $2,100 for Premium on its 2026 price list; Unchained offers a basic vault free and, per its own and third-party 2026 comparisons, a concierge Signature tier at $6,000 in the first year and $4,500 after. Those are the real carrying costs of self-custody done properly; on $25,000 the Casa Standard fee is 1% a year, four times the ETF.

Inheritance

Chainalysis’s estimate that 2.3–3.7 million coins are lost for good includes every estate whose executor could not find the words. A hardware wallet in a safe with no instructions is a gift to the remaining holders, and “my spouse knows the password” is not a plan; the multisig services exist partly to solve this, with a named heir, a lawyer or the service holding a key that works only in combination with what the estate can find. At death the coins receive a stepped-up basis, but only if they are found and valued, and under the wallet-by-wallet basis rules in force since January 1, 2025 the file that tells your heirs where the coins are should also tell the IRS what you paid for them.

The risk that grew fastest with the price is the one no wrapper except the ETF removes, and the one the marketing never mentions. The old cartoon has a criminal, faced with unbreakable encryption, hitting the owner with a five-dollar wrench until he types the password. Jameson Lopp, a bitcoin security engineer, has kept a public list of known physical attacks on crypto holders since 2014, and we parsed its fully dated entries on September 9, 2026: 78 recorded attacks in 2025 against 36 in 2024, and 57 in the first eight months of 2026 to a September 6 entry. The maintainer states the list is not comprehensive, and a further fifty or so entries with only a partial date were not counted.

France is 39 of the 57 cases in 2026. The wave’s defining case came on January 21, 2025 with the kidnapping in Vierzon of David Balland, a co-founder of the wallet maker Ledger, whose finger was severed before police rescued him, and in 2026 it has become routine: a family of five held hostage in Ploudalmézeau on April 20 and €700,000 extorted; an attempted kidnapping of the wife of The Sandbox’s co-founder in Seine-et-Marne on May 20, stopped by neighbours; a man held in his house at Jouy-en-Josas on August 24 against a €6 million demand.

Nor is it confined to France. Armed men demanded a hardware wallet from a couple in Queens, New York on June 29, 2026; a holder was kept for thirty hours in Bali on July 2 and gave up $5 million; and on September 1, 2026, in Atizapán de Zaragoza outside Mexico City, a musician, his pregnant wife, their three-year-old daughter and their domestic worker were killed in their home by two men hunting a hardware wallet that local media said held about $1.5 million in bitcoin. Chainalysis’s parallel count, published in July 2026, had 46 violent incidents worldwide by late June against 40 in the same period of 2025, more than $30 million taken, and home invasions rising from 14% of cases to 37%.

The victims share two things: someone knew they held coins, and someone knew where they lived. The 2020 leak of Ledger’s customer database, 272,000 names and postal addresses, remains the foundational dataset for the crime, and every exchange breach since adds to it. The defences are discretion (no social media, no devices shipped to your home), structure (multisig with a third-party key, so there is nothing in the house to hand over) and, for large holders, the security thinking a jeweller uses. It is also the strongest single argument for the ETF: nobody has been kidnapped for a brokerage account. The trade-off is sovereignty and 0.25% a year.

The rules changed materially in 2025 and 2026, and the change lands on your 2026 return; this is the treatment as it stands on September 9, 2026. The IRS decided in Notice 2014-21 that bitcoin is property, not currency, and everything follows from that: each sale, each swap into another coin, each purchase made with bitcoin is a disposition, a gain or loss measured against what you paid. Coins held more than a year are taxed at long-term capital gains rates, which for 2026 under Revenue Procedure 2025-32 are 0% on taxable income up to $49,450 single or $98,900 joint, 15% up to $545,500 single or $613,700 joint, and 20% above, plus the 3.8% net investment income tax above $200,000 single or $250,000 joint. Held a year or less, the gain is ordinary income at up to 37%. Bitcoin is not a collectible under section 408(m), so the 28% rate that applies to the art, wine, cars and coins elsewhere in this series does not apply, and the spot ETFs, as grantor trusts, are taxed identically to the coins.

No wash-sale rule, for now

Section 1091 disallows a loss on a security repurchased within thirty days. Bitcoin held directly is not a security for this purpose, so as of September 2026 you can sell at a loss, buy back within the minute and book the loss against other gains. Congress has tried to close this in every budget since 2021, in Senator Lummis’s digital-asset tax bill (July 2025, revised April 2026), in the bipartisan PARITY Act and in a June 2026 House bill; none had passed by September 9, 2026, and Treasury’s 2024 budget estimate that closing it raises about $23.5 billion over a decade is the reason it eventually will. Two cautions. The rule does apply to the ETFs, which are securities. And a harvested loss is a deferral: your basis resets lower and the gain returns when you finally sell.

Form 1099-DA and wallet-by-wallet basis

The era of self-reporting from a spreadsheet ended with Form 1099-DA under Treasury’s final broker regulations (T.D. 10000). Exchanges reported gross proceeds for the 2025 tax year on forms issued in early 2026, and from the 2026 tax year must report cost basis, but only for “covered” coins acquired on or after January 1, 2026 and held at the same broker until sold; anything bought earlier or transferred in is reported without basis, and the IRS will assume zero unless you can substantiate it. In parallel, Revenue Procedure 2024-28 ended the “universal wallet” method on January 1, 2025: basis is tracked wallet by wallet, so a coin bought at $30,000 and moved to a hardware wallet carries that basis only if you kept the transfer record. If you hold coins in more than one place, a crypto tax package (Koinly, CoinTracker, CoinLedger) is the cost of doing business. The reporting rule for decentralised exchanges was repealed under the Congressional Review Act on April 10, 2025, so trades there are reported by nobody but you.

Gifts, donations, death and the states

Gifts of appreciated coins carry your basis to the recipient; the 2026 annual exclusion is $19,000 per recipient. A donation of coins held more than a year avoids the gain and deducts fair value, but because bitcoin is not a “publicly traded security” in the Code, any donation above $5,000 needs a qualified appraisal. At death the coins receive a stepped-up basis, the federal estate exemption is $15 million in 2026, and none of it helps an estate that cannot find the keys. No state taxes bitcoin differently from other property: none in Texas, Florida, Wyoming and five other states; Washington’s capital gains tax reaches crypto gains above a $278,000 annual deduction (the 2025 figure, indexed), at 7% and, above $1 million of gain, 9.9%; Puerto Rico’s Act 60 offers decree holders 0% on gains accrued after residency, under rules the territory amended in March 2026 (Act 38-2026), which add a 4% rate for later applicants. The worked example assumes 15% federal and 5% state.

The same $25,000 and the same price path run through the three honest ways to own bitcoin, with every fee and the tax shown, so that the difference between the routes is visible in dollars rather than in philosophy. The assumptions, stated once. All three investors buy on September 8, 2026 at our tape’s $78,576 and hold five years. The price falls to $40,000 in the second year, a 49% drawdown, roughly the shallowest bottom in the record, and finishes the fifth year at $130,000: +65.4% in total, about 10.6% a year, modest by this asset’s history and aggressive by anyone else’s. Everyone pays 15% federal long-term capital gains and 5% state, 20% combined, on sale. Figures are rounded to the dollar.

Route A: IBIT inside a Roth IRA

The investor has the balance in an existing IRA (a $25,000 single-year contribution is not possible at the $7,500 limit). Buys $25,000 of IBIT at a $5 spread. The 0.25% sponsor fee is taken inside the fund every year, so after five years the investor holds 0.9975 to the fifth power, or 98.76%, of the coins the initial shares represented. Gross value at $130,000: $24,995 × 1.6545 × 0.9876 = $40,839; the fee drag over five years is $514 in end-value terms. Sells at a $5 spread: $40,834. Inside a Roth, no tax. Net $40,834, a 63.3% return. Total costs: $524.

Route B: IBIT in a taxable brokerage account

Identical purchase, identical $40,834 at the end. Taxable gain $40,834 − $25,000 = $15,834; tax at 20% is $3,167. Net $37,667, a 50.7% return. During the drawdown the wash-sale rule blocks a same-day loss harvest in the same fund; the investor could switch to FBTC for thirty-one days, which we have not modelled.

Route C: Kraken Pro to a hardware wallet

Posts a limit order on Kraken Pro at its 0.40% maker rate — the entry rate since the July 9, 2026 restructure, and the figure this example was rebuilt on — paying $100, withdraws to a Trezor Safe 3 for a $5 network fee, and spends $150 on the device and a steel seed plate. Coins bought: $24,895 (the $25,000 less both fees) ÷ $78,576 = 0.316827 bitcoin; basis $24,995, the cash plus the trading fee. Holds five years for nothing. Optionally, at the $40,000 trough, sells and repurchases within the hour (two trades, about $101 in fees) to book a $12,373 loss against other gains, worth $2,475 at 20%, resetting basis to $12,774; there is no wash-sale rule for property, so this is legal as the law stands.

At the end the coins are worth 0.316827 × $130,000 = $41,188. Sends them back to the exchange ($5), sells at 0.40%: $165. Proceeds $41,018. Without the harvest: gain $41,018 − $24,995 = $16,023, tax $3,205, net $37,813, less the $150 of hardware, $37,663, a 50.7% return. With the harvest: gain $41,018 − $12,774 = $28,244, tax $5,649, less the $2,475 saved in year two, net tax $3,174. Net cash after the hardware is $37,694, and after the $101 of harvest fees paid along the way, $37,592: about $70 worse than not harvesting, because the harvest only moved tax between years at the same rate and cost two trades to do it. Add a Casa Standard plan at $250 a year and the net falls to about $36,413.

Side by side: $40,834 in the Roth, $37,667 for the ETF in a taxable account, $37,663 for self-custody, $36,413 for self-custody with a paid multisig. The tax wrapper is worth $3,167 on $25,000. The choice between the ETF and self-custody, over five years, is worth four dollars, and the ETF has them: Kraken’s July 9, 2026 fee restructure took away the edge self-custody used to have on this arithmetic, which is the plainest illustration in this guide that a fee schedule is a fact about a date. The case for holding your own keys is a ten-year case rather than a five-year one, because the ETF’s 0.25% recurs and self-custody’s $420 does not: run the same example to 2036 and the fund’s drag roughly doubles while the wallet’s does not move. Against that stands the home-invasion record in section 12, which no fee table prices.

The sequence, in the order that avoids the expensive mistakes.

1. Decide the size first, in dollars, as a loss you can absorb. Take the amount you would accept losing outright and divide by 0.6; that is the most you should hold. The allocation literature (BlackRock’s December 2024 paper, Fidelity Digital Assets’ work) converges on 1–5%, and every study that finds more starts at a bottom.
2. Choose the wrapper by the account you have. A retirement account with room: a 0.14–0.25% spot ETF, and stop here. A taxable account only: the ETF still, unless the sum and your competence justify self-custody. Never a treasury stock as a proxy, never a lending product for the yield.
3. If you self-custody, buy the kit before the coins. A hardware wallet from the manufacturer, shipped to an address that is not your home, a steel plate for the seed, and a written plan naming where the seed is and who may use it. Test the whole path with $100 before you send $25,000.
4. Buy on a schedule, with limit orders. Twelve equal purchases on a fixed day, posted as maker orders on Coinbase Advanced or Kraken Pro, at 0.40–0.60% on the schedules in force in September 2026 rather than the 1.5–2% of a buy button. On our tape over the last year that schedule cost 0.7% where a lump sum cost 29%. Do not read the news on the day you buy.
5. Set the basis file up on day one. Every purchase, transfer and fee, in a crypto tax package or a spreadsheet; the 1099-DA regime assumes zero basis for anything you cannot document.
6. Write the rebalancing rule and the exit rule now. A band (rebalance when the position drifts more than 25–50% from target) and a level at which you would conclude the thesis has failed, which should be a fact about the network (fees, hash rate, a protocol failure) rather than a price.
7. Tell one person, and no one else. Your executor or spouse needs the plan; the internet does not need to know you own any.

The readings that would change the view, and the levels at which they would change it.

• Realized price and MVRV. Coin Metrics’ realized price, the average cost basis of every coin, was about $54,165 on May 23, 2026, and the market-value-to-realized-value ratio 1.41; every prior bear market bottomed at or below realized price (MVRV 0.56, 0.69 and 0.75 in 2015, 2018 and 2022), and the 2026 low did not get there (1.15 on February 5). A close below realized price would be, on the record, the capitulation this cycle has not had; an MVRV above 3.0 would be the first euphoric reading since 2021.
• ETF cumulative flows. $54.9 billion on August 27, 2026 on Farside’s tallies. A return above the $61.5 billion peak-day figure would mean the wrapper’s holders have replaced everything they sold; a fall below $50 billion would mean the August turn was a rally trade.
• Hash rate and fees. Coin Metrics’ thirty-day average was 977 EH/s in May 2026, public seven-day trackers about 934 in early September, against the 1,111 EH/s peak (CoinShares, on its own measure, had the network near 850 EH/s in early February). A thirty-day average below 850 EH/s would mean miners are leaving faster than the AI pivot explains; daily fees sustained above 20 bitcoin (against 2.5) would be the first evidence in two years that the security budget has a second leg.
• Strategy’s coin count. 845,050 at August 31, 2026. Any month in which the company sells more than it buys, or an enterprise mNAV below 0.8×, would signal forced supply of a size the ETFs could not absorb.
• Washington. The Senate cloture vote on the CLARITY Act scheduled for September 15, 2026; a wash-sale provision in any tax bill, which ends the loss-harvest route in the worked example the day it passes; the Labor Department’s final rule on alternatives in 401(k) plans, proposed March 30, 2026 and not final as of September 9, 2026; and the first 1099-DA forms with basis, due in early 2027 for the 2026 tax year.
• The calendar. The fifth halving at block 1,050,000, due around April 2028. If the July 2026 low holds, the base rate of the four prior recoveries (two to three years from trough to new high) points to 2028–2029; if the price makes a lower low first, the fourth epoch will have been the first in which the halving marked nothing at all.