In June a headline made the rounds: Bitcoin's on-chain activity at a seven-year low, the network as quiet as it last was in the 2019 bear market. Anyone reading that draws the obvious conclusion. Less activity means fewer users, fewer users mean less value — and that is where Metcalfe's Law comes in, because it claims exactly that link.
Is that right? Does the law hold for Bitcoin, and does less activity really mean less value? To answer that you need to know what the law says, where it came from, where it was ever measured — and what a Bitcoin address still counts today.
A law that started as a sales slide
Metcalfe's Law says the value of a network grows with the square of its participants. Two fax machines can hold one conversation, ten can hold 45, a hundred can hold 4,950. Each new participant adds not one unit of value but as many as there are participants already.
The idea is older than its name. Robert Metcalfe, co-inventor of Ethernet, showed it around 1980 as a slide at his company 3Com to sell network cards: past a critical size, the benefit exceeds the cost, so buy more. It became a "law" only in 1993, when the writer George Gilder called it one in Forbes ASAP. The law was a sales slide before it was a law.
For a long time nobody measured it. In 2006 Briscoe, Odlyzko and Tilly argued in IEEE Spectrum that not every connection is worth the same — most participants talk to a few others — so value grows more like n·log n than n². Only in 2013 did Metcalfe himself present data: ten years of Facebook users against Facebook revenue, and the curve fit. In 2015 Zhang, Liu and Xu confirmed the same with Tencent and Facebook data. Thirty years after the slide, there was data for the first time.
The tally after 45 years: for social platforms with revenue per user, the quadratic form holds. For telephone and the internet it was never shown that way, only asserted. Anyone who says the law "held for telephone, the internet and social media" is quoting two-thirds folklore.
Bitcoin fits over 15 years — almost
The law was applied to Bitcoin in 2018. Timothy Peterson set market capitalisation against network size and found that over 70 percent of the variation in value was explained. The two large upward deviations, 2013 and 2017, looked in the model like bubbles that came back down.
The most recent test is from August 2026. Stephen Perrenod ran the relationship between price and addresses with a balance across the entire history in five tests. Result: price grows with the address count to the power of 1.84 to 1.92. Metcalfe requires 2.0. Over a decade and a half, Bitcoin has grown almost quadratically. That is one half of the truth.
The other half: the last year
Here are the twelve months to today, from the same series the platform draws from the Bitcoin Research Kit:
| 16 Sep 2025 → 16 Sep 2026 | Value |
|---|---|
| Addresses with a balance | 54.38m → 56.83m (+4.5 %) |
| What Metcalfe expects from that (n²) | +9.2 % |
| Market capitalisation, actual | $2,326bn → $1,522bn (−34.6 %) |
The network grew, the model would have said up, the market lost a third. Metcalfe tells you where Bitcoin ended up after 15 years. Not where it will be next year.
That does not contradict Perrenod's exponent. A relationship across the whole history is a trend line; the deviations from the line are years, sometimes a whole cycle. Anyone who reads the trend line as a one-year forecast is trading on noise.
Since 2018 an address no longer counts a user
That leaves the claim that the model has been "breaking" since 2018. It traces back to an indicator by Charles Edwards (Capriole): a fair value built from daily active addresses, squared. Since February 2018 that line has sat permanently below the price. Edwards himself names as the most likely explanation not overvaluation but the gauge: more and more trading never touches the chain.
Perrenod's measurement puts a number on it. The share of supply held in institutional custody has risen from roughly 2 to roughly 20 percent. An ETF share creates no address. A purchase on an exchange that stays there creates no address. Ten thousand customers of a custodian share a handful of wallets. The people are there; the addresses are not.
A fair value that has sat below the price for eight years is not an overvalued coin. It is a tape measure with a piece missing.
What the seven-year low was made of
The 4 June 2026 report referred to the 60-day average of active addresses: just above 600,000, a level last seen in 2019. Bitcoin traded at $63,950 that day. It was the month of the lowest daily close in the past twelve months, $58,625 on 30 June. The report did not come out of an empty network; it came out of a price low.
Today, three and a half months later:
| Metric | Reading |
|---|---|
| Bitcoin price | $76,012 (+18.9 % since the report) |
| Transactions per day, avg. January 2026 | 393,989 |
| Transactions per day, avg. last 31 days | 702,516 (1.78×) |
| Lowest day in the last 12 months | 290,317 (11 January) |
| Highest day in the last 12 months | 915,712 (19 July) |
The low in addresses coincided with the low in price, and both are behind us. Anyone repeating the headline in September is holding up a June picture as a September picture.
One caveat belongs here: since 2023, transaction counts include inscriptions, entries with no payment purpose. So the 1.78× shows only that the chain has not emptied out. It does not show user growth. What it disproves is enough: a dying network looks different.
The objection: so Bitcoin is undervalued by Metcalfe?
"If addresses undercount users, the true value sits above the line. Then Bitcoin is cheap."
That sounds coherent and cannot be tested. The constant in the model was calibrated in years when an address was still roughly one user. Revise n upward after the fact because the price does not fit, and you have not rescued the model — you have built the result into it. That is the price of hindsight: a line that already knows the outcome is no longer a line.
How model lines fare when measured against their dated calls is shown by the platform's cycle-model scorecard as of 15 September: Stock-to-Flow sees a fair value of $843,552 against a price of $77,720, a deviation of −90.8 %, zero of two dated calls hit. The Power Law model sits at −47.1 %, the Rainbow chart at −71.3 %, the Bitcoin Wave model at −10.8 % with likewise zero of two hits. Metcalfe is not on it, because nobody has derived a dated price call from it that could be checked. A model line is a claim. It gets measured only once it carries a date.
What to do with this
Three rules for the next address headline:
- Ask what is being counted. "Active addresses" comes in half a dozen definitions: daily unique, averaged per block, with or without change outputs. The platform labels its series explicitly as a per-block average, not daily users — and says so inside the dataset itself. A number without a definition is not a number.
- Put a second series next to it. Addresses with a balance, transactions per day, value per transaction. If one falls and two rise, the one is measuring something other than usage.
- Take the trend line as a trend line. Metcalfe explains why Bitcoin stands where it stands after 15 years. On whether you buy or sell this month it says nothing that the last twelve months have not refuted.
What is not claimed here: that Bitcoin is "fair" or "cheap" by Metcalfe, or that addresses work as a buy signal. The measurement still missing: whether lows in the address average have historically preceded better or worse twelve-month returns. That can be computed. Until it is, it is a guess.
Method
Addresses with a balance, transactions per day and market capitalisation come from the Bitcoin Research Kit (BRK, an open-source chain index), retrieved through the platform's connector on 16 September 2026. The "addresses with a balance" series is the current stock of funded addresses and corresponds to network size in the Metcalfe literature. The Metcalfe expectation is (1.045)² − 1. The seven-year-low report refers to Bitcoin Magazine data as reported by crypto.news on 4 June 2026. Perrenod's exponent is from his 21 August 2026 publication; the Capriole indicator is on TradingView (published October 2019).
FAQ
Is the platform's address series the same as Glassnode's "daily active addresses"? No. BRK publishes the average of active addresses per block over 24 hours, not daily uniques. The two numbers differ by orders of magnitude and are not comparable. The seven-year-low report therefore cannot be recomputed with the platform, only put in context.
Why market capitalisation rather than price? Metcalfe values the network, not the share. Over twelve months it makes little difference because supply grows slowly; over 15 years market capitalisation is the right quantity.
Don't transaction counts include spam? Yes. Since February 2023 they include inscriptions, in bursts since the BRC-20 waves of 2023. That is why the transaction count appears in the text only as evidence that the chain did not empty out, not as a usage gauge.
Not investment advice, not a recommendation, not a forecast — historical patterns are not a promise.
Study the Past — Improve your Future 🥋