The crypto map — what the sectors are and what they actually do

There is a question that comes up in every crypto conversation sooner or later, and that rarely gets a decent answer: what is actually out there? Not “which coin”, but: which areas exist, what do they do, and how do they fit together?

This post is the map. It is the anchor of a series — most of the areas here will get their own deep dive, and every one of them will point back to this page. That is why the section headings stay put, no matter how often I update the text beneath them.

Three rules to make the map hold:

One as-of date. All figures are from 23 July 2026, rounded, with a source. Bitcoin stood at roughly $64,800 that day, ether at roughly $1,890 — well below year-ago levels. Drawing a map to two decimal places without naming the date is drawing fiction.

One guiding question per area: who pays whom for what? Not: what does the project promise. Not: how high is the price. But: does money flow from someone who gets a benefit to someone who does the work — or does the money come from issuing new tokens?

No ranking. The order here is a structure, not a verdict. I recommend nothing, I do not say what will go up, and the size of an area says nothing about its quality.

Symbolic image, AI-generated.

Why “sector” is not a hard number

Before the map, one thing has to go on the table, or you will misread every figure on it. The usual categories are not measurements, they are labels — and they overlap massively.

The 25 largest categories on CoinGecko add up to $8.42 trillion on the as-of date — against a total market capitalization of $2.30 trillion published by that same provider. That is 3.66 times the entire market, because any coin can sit in any number of categories. The two largest are nearly identical: “Smart Contract Platform” at $1.858T, “Layer 1 (L1)” at $1.839T. Fourth place goes to a category called “World Liberty Financial Portfolio” at $543.6B; there is also “Made in USA” ($278.6B) and “Made in China”.

That is not the provider’s mistake — categories are a search aid there, not a balance sheet item. It is a mistake in how those numbers get quoted. Two examples you will meet again below: the largest positions in the “Artificial Intelligence” category ($21.97B) are Chainlink and NEAR — two projects with no meaningful AI business. The largest position in the DePIN category ($7.56B) is Bittensor, an incentive network for models that has little to do with physical infrastructure.

Even the headline size is not a fixed number: on the same day CoinGecko reported $2.30T and 56.6% bitcoin dominance, CoinMarketCap $2.21T and 58.8%. An $88B difference, because the two count different tokens.

So I do not organize this by label but by four layers, each with its own guiding question and its own sensible metric.

Layer 1

Money and claims

What do I have a claim on — and against whom?

Key metric: outstanding face value

• Stablecoins Tokens pegged to a currency, mostly the dollar; issuers earn interest on the reserve ~$307B outstanding · two issuers account for 82.9%
• Tokenized securities (RWA) Treasuries, fund shares, credit, gold, equities as a transferable claim $34.9B freely transferable · treasuries $15.9B across 62,902 addresses
• Bitcoin as reserve + ETFs Holding as the main use case, increasingly via funds and corporate balance sheets largest ETF $48.0B · 6.1% of all bitcoin held by listed companies

Layer 2

Compute and security

Who produces block space, and who pays for it?

Key metric: user fees versus emission

• Mining and consensus Compute secures Bitcoin; it is paid for almost entirely in newly issued coins $919M in 30 days — 0.66% of it from users
• Layer 1 platforms The base chains programs run on fees over 30 days: Tron $26.2M · Solana $14.8M · Ethereum $7.7M
• Layer 2 / rollups Batch transactions and write only the result to the base chain $34.3B secured · 79 of 109 at trust stage 0
• DePIN Wireless, maps, weather, storage from private hardware, paid in tokens ~$72M revenue per year, the rest is emission
• AI × crypto Compute, incentive networks for models, machine-to-machine payments Akash: 124 GPUs rented · x402: $24M in 30 days

Layer 3

Applications

What does a user do with it, and what do they pay?

Key metric: fees actually paid by users

• DeFi Swap, lend, deposit — overcollateralized, without an institution $76.5B locked · $24.6B in fees over 12 months
• Prediction markets Contracts on events; the price reads as a probability $12.3B in 30 days (largest venue) · legal status contested
• Games, NFTs, collectibles Shrunk from investment narrative to collectibles logistics over 90% of game projects shut down · average NFT price $54
• Memes and social No earnings, only redistribution — the launchpads earn reliably $25.6B = 1.1% of the market · pump.fun $1.15B in fees
• Privacy Hiding who pays whom how much Monero $6.6B · Zcash $8.6B · EU ban for obliged entities from July 2027

Layer 4

Access, custody, glue

What connects the parts — and who holds the key?

Key metric: third-party assets held or secured

• Exchanges and custodians Where the market actually happens $245B in reserves · one exchange holds 56% of it
• Oracles Bring prices and events on-chain — and attract attacks $40.0B depends on a single provider · $3.85M in fees per year
• Bridges Translate between chains; the single most dangerous component $3.29B in damage = 19.6% of all crypto theft
• Identity and names Readable names, proofs without an ID — “the next big thing” for ten years ENS name service: $214,788 in revenue over 30 days

Layer 1: money and claims

Guiding question: what do I have a claim on — and against whom? This is the layer where crypto moves the most money and looks the least spectacular.

Symbolic image, AI-generated.

Stablecoins

Tokens meant to hold a fixed value, almost always the US dollar. They are the unit of account for the entire market: trading on an exchange usually means trading against a stablecoin, not against actual dollars.

On the as-of date roughly $307 billion were outstanding (DefiLlama, time series, retrieved 23 Jul 2026) — up about 17.8% over twelve months (from $261.0B). And the first look at measurement limits is right here: for the same day, rwa.xyz says $298.75B and CoinGecko $304.11B. A twelve-billion-dollar spread for a number that supposedly exists precisely.

Two issuers provide nearly all of it: USDT (Tether) $184.06B (59.1%), USDC (Circle) $74.03B (23.8%) — together 82.9%. The third largest, Sky Dollar, sits at $6.64B; every other one is below 2.2% market share.

Who pays whom here? Nobody pays for usage — stablecoins are free to hold. The business model is interest on the reserves: the issuer receives dollars, buys US Treasuries, and keeps the yield. Circle reported $694M in total revenue for the first quarter of 2026, $653M of it (94%) from reserve income; after $407M in distribution costs, $55M in net income remained. Tether reported over $10B in profit for 2025 and around $193B in reserves (self-reported; Tether publishes attestations by BDO, not an audited annual report — an audit process began in early 2026, according to the company).

Which leads to the defining property of this area: the earnings depend on the policy rate, not on the crypto market. The US target range stood at 3.50–3.75% in mid-2026. If it falls, issuer profits fall almost one for one.

The sector is now large enough to feed back: a Bank for International Settlements study finds that a $3.5B inflow into stablecoins pushes three-month T-bill yields down by roughly four basis points over ten days. At this point, crypto and sovereign debt are not separate worlds.

How it fails: a stablecoin is a design claim, not a law of nature. Two documented patterns. The reserve gets stuck: in March 2023, $3.3B of USDC reserves (about 8%) sat at the collapsed Silicon Valley Bank; USDC fell to $0.95 and only recovered once Treasury, the Fed and the FDIC jointly declared an exception on 12 March and guaranteed all deposits at that bank. Issuance is badly secured: on 22 March 2026 an attacker minted roughly 80 million USR with no backing and extracted 11,409 ETH (about $23.7M); the price fell to $0.025 within 17 minutes. The cause: minting rights sat with a single address, no multi-signature, no volume limit.

A third pattern is new and quieter: yield-bearing stablecoins are shrinking. In the second quarter of 2026 they lost about 15% of their supply, sUSDe alone 52% — while plain Treasury-backed tokens grew. In the same quarter, total supply shrank on a quarterly basis for the first time since 2023.

In the EU, MiCAR applies: stablecoins need authorization, and the numbers are soberingly short — 21 authorized e-money token issuers, 41 white papers, zero authorized asset-referenced tokens (ESMA register, as of 21 Jul 2026). Tether never applied for USDT and has disappeared from regulated EU venues. The largest euro stablecoin, EURC, sits at roughly $430M; all eight MiCA-compliant euro stablecoins together came to about $674M in June 2026 — less than a quarter of one percent of the sector. The digital euro is a different thing entirely: central bank money, not a public blockchain.

In the US, the GENIUS Act has been signed since July 2025 but is still not in force a year later: the agencies missed the implementation deadline of 18 July 2026; the law takes effect no later than 18 January 2027. A February 2026 rule proposal by the OCC would extend the statutory ban on paying interest to affiliates via a rebuttable presumption — a reading that is contested in the consultation.

What stablecoins are actually used for — payments, remittances, inflation protection — is covered at length in “Crypto in practice”.

Check questions: who holds the reserve, and is there an audit or only an attestation? Is the issuer in a supervisory register — in which country? Who may mint new tokens, and how many signatures does that take? And if a stablecoin promises yield: where does it come from?

Tokenized securities and real-world assets (RWA)

A claim on something real — treasuries, fund shares, credit, gold, equities — represented as a token and transferable around the clock. This is the area where traditional finance has actually built.

And it holds the most instructive number on the whole map. rwa.xyz publishes two values: “distributed” $34.94B — assets that sit freely transferable on public chains — and “represented” $366.97B, everything that is tokenized in any sense. Of those $367B, $327B sit on a single network (Canton) that reports zero dollars in freely transferable assets there. Anyone reading “tokenization is at $367 billion” is reading the second number. The market you can touch is an order of magnitude smaller.

The largest category is tokenized US treasuries and money market funds: $15.92B across 85 products — up from roughly $1B in early 2024. The issuers are not startups: BlackRock’s BUIDL ($2.52B), Circle’s USYC ($2.96B), Ondo’s USDY ($2.16B), Franklin Templeton.

Except: 62,902 holder addresses across the entire category. BUIDL spreads $2.52B across 113 addresses, USYC $2.96B across 46. This is not a retail market, it is wholesale with a token surface. Fittingly, Forbes found in early July 2026 that of roughly $60B in tokenized assets, $32.9B sat in 910 assets with no transfer activity at all over a full week. Tokenized does not mean traded.

Tokenized equities are the area with the most small holders (692,410 addresses) and the smallest volume: $1.92B. Here the legal fine print is the entire point. Ondo’s tokens are, per the issuer, “not themselves shares” and carry no right to delivery of the underlying; xStocks are tracker certificates from a Jersey entity under a base prospectus approved by the Liechtenstein regulator, without voting or shareholder rights. In January 2026 the SEC set out four tokenization models in a joint statement by three divisions and noted that the synthetic ones convey “no ownership and no rights”. You are buying the price movement, not the security.

Germany has had its own law for this since 2021. The BaFin register of crypto securities lists 275 entries from around 175 issuers (as of 2 Feb 2026), registered between December 2021 and January 2026. Two registrars account for three quarters of them. There are real issues in there — KfW with a €100M bond, DekaBank, Deutsche Bank — and the list is honest about failure too: for two securities it notes deletion “due to the insolvency proceedings of the issuer”. The European DLT pilot regime has applied since March 2023; as of the ESMA cut-off at the end of May 2025, three infrastructures were authorized, one of them with a bond issue of — not a typo — €401. In late 2025 the European Commission proposed raising the ceiling from €6B to €100B.

Check questions: what exactly do you hold — ownership, a fund share, or a claim against a special purpose vehicle? Which jurisdiction is the issuer in? How many holder addresses does the product have, and when was it last moved at all? And who custodies the underlying?

Bitcoin as a reserve asset and its institutional wrappers

The strangest area on the map: an asset whose main use is being held — increasingly through products that did not exist three years ago.

The largest bitcoin ETF, IBIT, holds $47.99B at a 0.25% fee; the older GBTC sits at $8.77B and charges 1.50%. For comparison, because it sorts the proportions of this entire map: a single ETF holds more value than everything locked in DeFi applications on Ethereum ($41.56B).

Alongside that are companies putting crypto on the balance sheet: 1,285,045 bitcoin sit with listed firms — 6.12% of all bitcoin in existence. Strategy alone holds 843,775, 4.018% of all bitcoin that will ever exist, bought for $63.68B and worth $54.70B on the as-of date (down 14.1%). With ether the concentration is sharper still: a single company, BitMine Immersion, holds 4.787% of the entire ether supply.

That is a structural point, not a price call: a meaningful share of supply now sits on a few balance sheets that answer to their own constraints — quarterly reports, credit agreements, shareholders. Who really owns the coins is a more relevant question on this map than where the price stands. How holdings are distributed across all size classes is in “Shrimp, fish, whale”.

Layer 2: compute and security

Guiding question: who produces block space, what does it cost, and who pays for it? One metric here is more revealing than all the others: how much of the security budget do users pay — and how much does the network print itself?

Mining and consensus (Bitcoin)

The production side of Bitcoin, which almost never shows up in market reports. Data centres worldwide solve a puzzle that orders blocks; whoever wins gets new coins plus the fees of the transactions included.

Hash rate stood at 876 exahashes per second on the as-of date (mempool.space). Miners took in $919 million over 30 days (blockchain.com) — $11.18 billion annualized. And now the number that matters: $6.06 million of that came from transaction fees. 0.66%. The other 99.34% is newly issued coin.

That is not a criticism, it is the design: Bitcoin pays for its security with emission for now, and that emission halves on a schedule — next time in about 630 days. It is also the yardstick against which every other area should be measured. Because the same question — does the money come from users or from the printing press? — applies below to DePIN, to the AI networks, and to most tokens on this map. Bitcoin’s security budget alone is 16 times the combined fee income of the four largest smart-contract chains.

On energy consumption: the Cambridge index is the standard source, but it was not machine-readable on the as-of date. I am not filling in a number from memory — this will be added later.

Layer 1 platforms: the base chains

The chains programs run on. Three properties matter for a non-specialist: what they cost, how many independent machines run them, and whether they can stop.

By market capitalization: Ethereum $228.18B, BNB $75.32B, Solana $44.41B, Tron $30.99B (XRP sits in between at $69.15B; I leave it out here because it was built primarily as a payment network — but exactly these judgement calls are why sector numbers wobble).

By fees users actually paid over 30 days (DefiLlama), the order looks different: Tron $26.22M, Solana $14.77M, BNB Chain $8.64M, Ethereum $7.69M — Ethereum down 29.4% against the prior period. Familiar names further down took in, over those same 30 days, $132,000 (Aptos), $88,800 (Sui), $68,630 (Avalanche) and $40,814 (Cardano). A network with a billion-dollar valuation can have revenue in the low five figures. That is not a contradiction — it just means the price is valuing something other than current usage.

Ethereum is very cheap right now: at a daily average around 0.5 gwei, per the Ethereum Foundation a transfer cost about $0.025 and a token swap $0.21 (as of 5 May 2026). Staked are 40.9 million ETH (33.6%) across 886,508 validators, yielding about 2.64% — which comes overwhelmingly from emission, not from user fees.

Decentralization can be roughly read off the number of block producers, and the range is enormous: Cardano over 2,000 stake pools, Solana 698, Avalanche 625, Sui 125, Aptos 94, BNB Chain 45 (21 per round) and Tron 27. Solana had 2,560 validators in March 2023 — a drop of over 70%, while the chain itself became markedly more reliable: no official outage has been recorded since February 2024. Conversely, Sui stood still for almost six hours on 28 May 2026 and was halted three times within 48 hours, due to a bug in the gas accounting of a new version. New chains are faster and more fragile; that is a trade-off, not a disgrace — you should just know about it.

Two things you can learn here and will need everywhere else:

The gap between market cap and FDV. Market capitalization counts circulating tokens; fully diluted valuation counts every token that will ever exist. For Hyperliquid that is $13.24B against $59.52B — a factor of 4.5. The difference is future supply nobody holds yet. For Ethereum, BNB and Tron that gap is essentially zero.

And the test case for the whole map: Hyperliquid. The same project shows up in the statistics as a layer 1 ($13.24B market cap), as a venue for perpetual futures, as an entry in the rollup database, and as an FDV example — with four different sizes. None of it is wrong. It just shows that the category sticks to the observer, not to the thing. So when you read that a sector is “$X billion large”, the first follow-up is always: measured how, and who is simultaneously counted in which other drawer?

Layer 2: scaling by outsourcing

Rollups batch many transactions and write only the result to Ethereum. For users that means near-identical security at a fraction of the cost. Median costs on the as-of date: $0.0014 on Base, $0.0048 on Arbitrum — against $0.1026 on Ethereum itself.

Value secured across all rollups stood at $34.25B (L2BEAT) — a year earlier $55.49B, at the peak $66.28B (December 2024). Base ($11.68B) and Arbitrum ($10.54B) together hold roughly 65%.

The economics of this layer are the real finding. In the 30 days to 22 July 2026, all tracked rollups took in $8.41 million in user fees — and paid $40,670 to Ethereum for it, $2,315 of that for blob storage. Half a percent. Since the upgrade that introduced blobs, data storage on Ethereum has become so cheap that it has all but vanished as a revenue source: in December 2024 Ethereum still earned $4.07M a month from it, now it is $2,138 across 885,544 blobs. The scaling problem was solved by driving the price of the scarce good to near zero — with all that implies for who pays for the base chain in future.

How decentralized is this? L2BEAT rates 109 projects: 79 sit at stage 0 (the operator can effectively intervene), six at stage 1, four at stage 2 — and those four genuinely trust-minimized rollups secure about $509,000 between them. For Base, Arbitrum, Starknet, Linea, Scroll and ZKsync Era the “exit window” is listed as “none”: on an upgrade, there is no guaranteed window to get out.

What that means in practice was demonstrated on 21 April 2026: Arbitrum’s security council changed a core contract with 9 of 12 signatures and without a token-holder vote, in order to redirect 30,765.67 ETH out of an ongoing attack. In substance that was a rescue — and simultaneously proof that the emergency brake exists and gets used. Both belong on the map.

Outages here are routine, not exceptional: Base was down for nearly two hours on 25 June 2026, Starknet for four and a half hours on 5 January 2026 (discarding 18 minutes of history in the process), and Linea recorded eighteen incidents in 2026 alone. Two networks were wound down: Polygon zkEVM stopped sequencing on 1 July 2026, and Zero Network announced its wind-down after roughly 18 months.

Check questions: what stage is the rollup at on L2BEAT? Is there an exit window? Who can change contracts, with how many signatures? And how much does the network pay to the chain whose security it claims?

DePIN: infrastructure from other people’s living rooms

People run hardware at home — radio antennas, dashcams, weather stations, storage — and get paid in tokens. The appeal: the product is not money, it is infrastructure.

The reality check is unusually clear here, because revenue is measurable. Messari puts total on-chain revenue across all DePIN networks for 2025 at roughly $72 million — against a sector market capitalization around ten billion. The rest of what operators receive comes from token emission.

Helium shows the economics in one sentence: the network cut compensation from $0.50 to about $0.10 per gigabyte while carrier data traffic rose from 24,000 (June 2025) to roughly 97,000 gigabytes per day (April 2026). More real usage, less payout per unit — that is the moment a subsidized network starts becoming a business, and simultaneously the moment the return on the hardware disappears. The token sits 99.6% below its 2021 high. A governance decision in 2025 abolished proof of coverage as a reward basis.

Similar picture in storage: Filecoin has 651 pebibytes stored against roughly 2 exbibytes of raw capacity — about a third utilized; the younger Walrus reached 467 terabytes after a year, 250 of them from a single customer. At Hivemapper/Bee Maps, $32M in venture funding and a customer list naming HERE, TomTom and Lyft (company statements) stand against a token 99.8% below its high.

Check questions: how much revenue comes from paying customers — in dollars, not tokens? How much emission flows out at the same time? Does the hardware still pay for itself if rewards halve? And: is the device a capital investment or a bet on a price?

AI and crypto

The area with the widest gap between story and evidence. It has three branches that need to be kept apart.

Decentralized compute. Networks that broker GPU capacity. The numbers are small: Akash had 263 GPUs in the network on the as-of date, 124 of them rented; lease revenue in the first quarter of 2026 was $253,250 (down 45%), and cumulative revenue since launch $5.67M. io.net advertises 327,000 GPUs but verifies roughly 6,720 daily — about two percent — at an annualized revenue around $12.5M. Aethir self-reports $127.8M in revenue for 2025; its own recurring-revenue figures contradict each other between $147M and $166M.

The price advantage, the core argument of the category, has melted away: an H100 hour cost $1.89 to $1.99 at commercial providers on the as-of date; Akash advertises around $1.32. For scale: the large cloud providers are investing between $660 and $725 billion in data centres in 2026.

Incentive networks for models. Bittensor pays for contributions with emission — after the December 2025 halving, roughly 3,600 TAO per day, or roughly $254 million a year. The largest subnet receives about $36.5M of that annually and sets it against $1.3M to $2.4M in customer revenue. In other words: the service is more than 90% funded by issuing new tokens.

Machine-to-machine payments. The most interesting branch, because there is a real problem here: software agents do not have credit cards. The open x402 standard reached roughly 75 million transactions and $24 million in volume over the 30 days to mid-July 2026, across 94,000 buyers. Caution is still warranted: in March, the analytics firm Artemis classified about half of the then-current volume as artificial — the mix has since shifted toward genuine payments.

And as a reminder of the opening: CoinGecko’s “AI” category covers $21.97B, led by two projects whose business has nothing to do with AI. Provenance for media, often sold as a crypto use case, runs in practice on the C2PA industry standard — explicitly without a blockchain, using signed metadata.

Check questions: how much paying revenue is there against token emission? Is the advertised capacity verified or merely registered? Would the service be cheaper than a normal provider without the subsidy? And does the application need a blockchain at all?

Layer 3: applications

Guiding question: what does a user do with it, and what do they pay? The metric for this layer is fees paid by users — not value locked, not volume.

DeFi: trading, credit, interest without a bank

Programs instead of institutions: swap, lend, deposit. Value locked stood at $76.5 billion on the as-of date — down from $171.0B at the peak on 7 October 2025. More than half of it ($41.56B) sits on Ethereum.

Important against the quick verdict: the decline is mostly a price effect, not a capital outflow. Lido held more ether on the as-of date than at the dollar peak in August 2025 — same quantity, half the price. Reading TVL charts as a confidence gauge means measuring the price.

What is actually happening:

Trading. Decentralized exchanges turned over $6.53B in 24 hours, $4.17 trillion over twelve months. Their share of spot trading against centralized exchanges was 13.65% in June 2026 (peak: 21.75% in June 2025). In perpetual futures, by contrast, the decentralized share is growing structurally: from 3.1% of open interest (early 2024) to 12.0% (end of January 2026); roughly 88 to 90% still runs through centralized exchanges.

Credit. Aave V3 has $13.81B locked against $10.78B in outstanding loans, Morpho Blue $7.49B. All of it collateralized: you post more than you borrow. Uncollateralized lending does not work on-chain so far — at Goldfinch, $56M in issued loans are stuck, of which roughly 30% has come back after six years.

Interest. And here comes the single most useful number in this post: staked ether yielded 2.16% on the as-of date, and the large decentralized savings product from Sky 3.52% on $4.70B in deposits. That is the market rate of this sector. If someone offers you a double-digit yield on stablecoins, the difference comes from a risk somebody is carrying — or from token emission. Incidentally: those $4.70B are spread across 5,893 depositors.

Who earns? Over twelve months, $24.61B in fees flowed through all tracked protocols; $2.10B of that went to token holders. In many large protocols that share is simply zero: Morpho Blue took in $26.59M over 30 days and distributed nothing, Uniswap V4 $35.29M and likewise nothing. A governance token is usually not a claim on earnings.

How it fails. In 2026 up to the as-of date: $1.045B in losses across 145 incidents. The median is $219,000, the mean $4.7M — a few large cases dominate the statistics. The key finding from TRM Labs for the first half of 2026: 76% of the total damage came from compromised keys and infrastructure, not from faulty contracts. The weakest point is rarely the mathematics, usually the key management. On top of that is the quiet dying: ZeroLend announced its wind-down in February 2026 at $6.6M, after a peak of $359M; SummerFi in July 2026 after seven years.

Prediction markets

The fastest-growing use case on the map, and the most interesting open legal question. You trade contracts that pay a fixed amount if an event occurs — elections, rate decisions, sports results. The price reads as a probability.

The scale: Kalshi turned over $12.34B in 30 days ($52.21B over twelve months), Polymarket $4.31B. In the second quarter of 2026 the sector grew 48.7% to $113.8B in notional volume. That makes this single area larger by orders of magnitude than the entire fee income of all rollups.

And it is being fought over in real time: Michigan barred Kalshi from offering sports contracts in late June 2026, and in mid-July 2026 the US derivatives regulator CFTC in turn barred the company from complying with that order — the jurisdictional dispute between federal and state level is pending. In France, the gambling authority ordered internet providers to block Polymarket in July 2026, and the Czech Republic put the service on a blacklist. In July 2026 the European securities regulator ESMA stated that event contracts with a binary structure may qualify as financial instruments — which would bring prospectus, authorization and investor-protection rules with them.

There are technical and economic warning signs too: a study by Stanford and SMU showed that five-minute contracts on the bitcoin price create manipulation incentives — with fifteen-minute contracts the effect disappeared. And in June 2026 Polymarket’s web front end was attacked through a compromised third-party dependency; eleven wallets lost $2.94M, and the operator promised full reimbursement.

A nice detail for the measurement nerds: DefiLlama now counts prediction markets in the DEX category. Kalshi and Polymarket together make up about eight percent of the volume there — so anyone quoting “DEX volume” has for some time also been quoting bets on football matches.

Games, NFTs and collectibles

The area with the deepest fall — and a resurrection almost nobody expected.

Games: the trading firm Caladan examined over 3,200 web3 gaming projects and classified over 90% as “effectively dead”; more than 300 titles were shut down. Studio funding collapsed by 93%, and gaming’s share of crypto venture capital fell from 62.5% (2022) into the single digits. The tokens of the best-known names sit 98 to 99.8% below their highs. What remains of “play to earn” is covered in “Crypto in practice”.

NFTs: here the nuance matters more than the verdict. The market is not dead, it has become mundane. More NFTs sold in October 2025 than in any other month that year — at an average price of $54, one sixth of the January figure. The highest-volume collection in the week before the as-of date was not CryptoPunks (about $991,000) but Courtyard with $8.72M across 182,492 transactions: tokenized, physically vaulted trading cards. The area lives as logistics for collectors, not as an art market.

Two structural findings worth knowing before buying any of it: a measurement study of 12,353 Ethereum contracts holding over 6.2 million tokens found that in 25.24% of contracts the referenced files were unreachable. The token is on the blockchain, the image is on a web server — in April 2025 nearly 20,000 images from a well-known Nike collection vanished because a hosting contract was downgraded (a class action over this is pending in the US; the allegations have not been decided in court). And on airdrops, the sector’s favourite acquisition tool: per DappRadar, 88% of distributed tokens were below their launch price within three months.

Meme coins and social

The most honest area on the map, because it never claimed to produce anything. The meme category comes to $25.57B — 1.1% of the total market and about 54% below the year-ago level; Dogecoin alone accounts for 42% of it.

Where money is earned here is precisely measurable: the launchpad pump.fun has brokered $92.74B in trading volume since launch and taken $1.147B in fees from it, at 0.95% per trade. Over the last 30 days that was $24.91M — down roughly 31% against the first quarter of 2026. The infrastructure earns reliably; the tokens on it earn nothing. Every payout to a seller is the deposit of a later buyer. That is not a judgement, it is the mechanism — covered in full, with risks and legal context, in “Meme coins — the honest map”.

Among social applications, the gap between attention and size is striking: the Farcaster protocol took in $60,037 in fees over 30 days.

Privacy

Tools that hide who pays whom how much — on a technology whose default state is total publicity. Monero ($6.58B) hides sender, recipient and amount by default; Zcash ($8.64B) makes it optional — 26.03% of supply actually sits in the shielded pool.

How hard the cryptography is here was shown in June 2026: a flaw was disclosed in Zcash that had been live since May 2022 and could in theory have allowed undetected inflation; it was found by a commissioned reviewer and fixed within days — the price fell 38% in 24 hours. Four years undetected, in one of the most-audited protocols in the field.

Legally this is the touchiest area, and for EU readers the single most practically important line on the map sits here: the EU anti-money-laundering regulation bans anonymous crypto accounts and the handling of “anonymity-enhancing coins” by obliged entities from 10 July 2027 — exchanges, custodians, banks. Coins that offer anonymity only optionally are explicitly covered. The ban addresses companies, not possession; makers of hardware, software and self-custody wallets are exempt as long as they have no access. In practice: anyone holding such coins should know which licensed route they intend to use to convert back into euros later.

In the US the movement went the other way: sanctions against Tornado Cash were lifted in March 2025 following a court ruling. Criminal cases against individual developers continue or have concluded — for Tornado Cash developer Roman Storm a guilty verdict was returned on one count in August 2025 while the jury deadlocked on two others; the proceedings are not finished, and the presumption of innocence applies where no final decision has been made. The founders of the Samourai service were sentenced to five and four years respectively in November 2025.

How big is the problem being regulated? Chainalysis puts illicit inflows for 2025 at at least $154B — under one percent of attributed volume, 84% of it in stablecoins, not privacy coins. These figures come from commercial surveillance vendors with a business interest in large numbers; I quote them because they are the best available, not because they are neutral.

Layer 4: access, custody and glue

Guiding question: what connects the parts — and who holds the key? The invisible layer, where the largest sums and the largest losses sit.

Exchanges and custodians

Where the market actually happens. 78 centralized exchanges hold $245.05B in traceable reserves, Binance alone $138.11B — more than three times the entire DeFi market, about which incomparably more is written. A single exchange turned over roughly $6.71B in 24 hours on the as-of date; all decentralized exchanges combined did $6.53B that same day.

This is also where the only audited set of numbers on the map sits: in SEC filings, Coinbase reports $7.18B in revenue for fiscal 2025, after $6.56B in 2024 — with a 30.5% decline in the first quarter of 2026 against the prior-year quarter. A single regulated company thus earns about 29% of what every protocol on this map collects in fees over twelve months.

On where all of this is domiciled: the largest venues are incorporated in the Cayman Islands, the Seychelles, Panama and the British Virgin Islands. For anyone thinking about MiCAR and investor protection, that is the most relevant sentence in this section — the rules apply where the provider is licensed, not where the app is installed.

Oracles: data from outside

A blockchain knows nothing about the world. Oracles supply prices, rates and events — which makes them the point where attacks concentrate.

The market is extremely concentrated: Chainlink secures $40.03B across 541 protocols; 64.4% of oracle-dependent capital hangs on a single provider with no fallback. Since 2020, the incident database records 69 oracle-related events totalling $734.8M.

And then the economics that make this section representative of the whole map: Chainlink took in $3.85M in fees over twelve months — all of which goes to node operators. The token has a market capitalization of $6.34B. The Graph, the indexing service behind countless applications, earned $610,732 in query fees over a year across more than a trillion queries answered. These services are indispensable and essentially unpaid.

Bridges and swap routes

Chains do not talk to each other; bridges translate.

Symbolic image, AI-generated.

They are the single most dangerous component in the whole system: 57 bridge hacks caused $3.29B in damage — 19.6% of all recorded crypto losses, on a fraction of the capital. The list reads like a who’s who: Ronin $624M (March 2022, attributed by the FBI to the Lazarus group), Poly Network $611M, BNB $570M, Wormhole $326M.

The most recent large case shows the pattern: on 18 April 2026, a bridge configuration released 116,500 rsETH (about $293M) without backing, because message verification was performed by a single instance — operated by the bridge provider itself, which publicly acknowledged the error in May and changed the default. At Aave this produced $123.7M in bad debt; a community rescue package had recovered roughly 54% of the missing amount by the end of April. In 2026 up to the as-of date there were 27 bridge incidents totalling $399.6M — more individual cases than in any prior year, at smaller amounts.

The economics are remarkable here too: the large messaging protocols take a zero percent cut — LayerZero, Wormhole, Across and Axelar each report no protocol revenue.

Identity and names

The area that has been “the next big thing” for ten years. Worldcoin sits 96.8% below its high; the ENS name service took in $214,788 over 30 days. I list it because it is on every map — and because the numbers show how far story and usage can drift apart.

Across all layers: law, tax, measurement

Two topics belong in every row of this map, not in a chapter of their own.

Law. In the EU, MiCAR has governed providers since late 2024; the last transition period expired on 1 July 2026. From July 2027 the anti-money-laundering regulation adds the ban on anonymous accounts. And for readers in Germany, the rule that has applied since 1 January 2026 and that almost nobody knows: under the EU directive DAC8, crypto service providers report identification data, transaction counts and gross amounts per crypto asset to the tax authorities — automatically, for the current calendar year. German tax treatment (paying with crypto is a disposal under § 23 EStG) is covered in “Crypto in practice”. Not tax advice, just a note that the reporting is already running.

Measurement. Five traps, all of which appeared in this post — and which explain every sector figure you will ever be quoted:

1. Categories overlap. $8.42T in category totals against a $2.30T market.
2. Same thing, two definitions. Base is quoted at $11.68B (value secured) and at $4.59B (locked in applications), Arbitrum at $10.54B and $1.24B. Both numbers are correct — they measure different things. Without asking, you will assume one is wrong.
3. Emission is not revenue. Bitcoin: 0.66% of miner income comes from users. DePIN: $72M in revenue a year. Rollups: they pass half a percent of their income to the chain that secures them.
4. FDV is not market cap. A factor of 4.5 at Hyperliquid, 2.5 at Sui.
5. Volume is partly fabricated. An academic study of 29 exchanges found substantial wash trading on unregulated venues. And even reputable aggregators differ by 17% on the same metric on the same day.

A last, uncomfortable point about the data: the map depends on providers who make a living measuring this market. Chainalysis, TRM, Messari, DefiLlama, L2BEAT, rwa.xyz — they supply the numbers this post quotes, and they sell software to agencies and investors. Some data simply disappears, too: part of the DefiLlama interface now answers only against payment, and three formerly central industry data sources are today a parked domain, a dead DNS entry and a redirect.

What the map says as a whole

Two statements are carried by every layer — and neither appears in any single section.

First: the market does not earn its money where the story is being told. Of $24.61B in protocol fees over twelve months, roughly 34% goes to two stablecoin issuers collecting interest on government bonds. Next to that: bitcoin mining at $11.18B a year, a single regulated exchange at $7.18B. And against that, the numbers from the areas that get written about most — Chainlink $3.85M a year, The Graph $610,732, the Farcaster protocol $60,037 over 30 days. The earnings come from interest, from emission, and from trading fees at centralized venues.

Second: concentration is the rule, not the exception. Two issuers provide 82.9% of all stablecoins. Two rollups hold 65% of value secured. One oracle secures 64% of dependent capital. One exchange holds 56% of reserves. One company holds 4.0% of all bitcoin, another 4.8% of all ether. Nine areas, nine independent pieces of evidence, one pattern. When you hear “decentralized”, ask: at exactly how many points?

And what cannot be shown, though it is constantly claimed:

“Adoption is rising.” This map contains no reliable user count. What it does contain are holder counts: 113 addresses for $2.52B, 5,893 depositors for $4.70B, 46 wallets for $2.96B. Wallets are not people, and nobody publishes an audited user number.

“Crypto is dying.” Equally unsupported. Prediction markets grew 48.7% last quarter, the decentralized share of futures trading has risen for two years, tokenized treasuries are up fifteenfold since early 2024, and NFT unit sales hit yearly highs in autumn 2025. The market has become smaller in dollars and more grown-up in some corners.

What remains is a landscape in which a few unglamorous areas earn real money, many loud areas live off their own printing press, and the line between the two becomes visible in exactly one metric: does the money come from someone who gets something out of it? That question applies before every purchase, every device, every promise — and it is the same attitude the Crypto Collection stands for: stay sober when everyone else gets loud.

Where this goes next

Most areas on this map will get their own post — what is inside, how to check it, what can go wrong. Already published: Crypto in practice (what actually gets used), Meme coins, Technical analysis, Leverage, Bear markets in numbers, running your own bitcoin node, and the glossary, which explains every technical term on this page.

This map stays put and gets updated — I do not change the section headings, so links to them keep working. The version history at the end shows what changed.

Sources and limits

All figures carry the as-of date 23 July 2026 unless stated otherwise. Primary sources where available: SEC filings (Coinbase), Tether and Circle quarterly reports, the BaFin crypto securities register, ESMA register and statements, EU legal acts (MiCAR, AMLR, DAC8, DLT pilot regime), the OCC rule proposal, the Ethereum Foundation, BIS working paper 1270, NBER working paper 30783. Aggregators: DefiLlama, L2BEAT, rwa.xyz, CoinGecko, Messari, mempool.space, blockchain.com, Chainalysis, TRM Labs, Immunefi, DappRadar, growthepie. Company self-reported figures (Tether, Aethir, Bee Maps, io.net, Akash, Helium, Chainlink) are marked as such in the text — they are documented, but not independently verified.

Three honest limits: first, the energy consumption of the Bitcoin network is not quantified here, because the standard source was not readable on the as-of date. Second, market caps, TVL and volume figures are snapshots from providers with differing definitions; where they diverge, the divergence is in the text. Third, the selection of areas is an editorial decision — a different structure would be defensible, and I have named the borderline cases (Hyperliquid, prediction markets, stablecoins sitting between money and application) rather than hiding them.

And once more, because it is the core: that an area is large does not mean it is good. That it earns money does not mean its token does. This map shows where things stand — not where you should go.

Questions, or found a mistake? Write to me.