A directed acyclic graph (DAG) is a structure of nodes joined by one-way links that never loop back. In distributed ledgers, a DAG lets many transactions or blocks reference several predecessors at once instead of forming a single chain, which can raise throughput. The idea has matured: today DAGs appear less as "blockchain replacements" and more inside the consensus engines of high-performance networks.
Chain versus graph
A classic blockchain is a DAG with one strict rule: each block points to exactly one parent, so history is a single line. Competing blocks are discarded, which wastes work and caps throughput. A DAG ledger relaxes the rule. A new unit can reference multiple earlier units, so units produced in parallel can all be kept and later placed in a consistent order. The hard problem moves from "which chain wins" to "how does everyone agree on the order of a graph".
Major DAG designs
| Network | Structure | How ordering is agreed |
|---|---|---|
| Hedera | Hashgraph: events record gossip history between nodes | "Gossip about gossip" plus virtual voting, an asynchronous Byzantine fault tolerant algorithm run by a council-governed node set |
| IOTA | Originally the Tangle, where each transaction approved earlier ones | Early versions relied on a central Coordinator; the network has since been rebuilt, and its 2025 "Rebased" upgrade moved to a validator-based, Move-language object ledger |
| Nano | Block-lattice: each account has its own chain | Representatives vote only when two blocks conflict (Open Representative Voting) |
| Kaspa | BlockDAG of proof-of-work blocks | GHOSTDAG ordering lets parallel blocks coexist, enabling fast block rates |
| Sui, Aptos and similar | Linear blocks for users, DAG internally | DAG-based BFT protocols (the Narwhal/Bullshark family, Sui's Mysticeti) separate data dissemination from ordering |
The last row is the important trend. Research on DAG-based Byzantine consensus showed that letting validators broadcast batches in parallel, building a DAG of certificates, and then ordering it gives high throughput with strong safety guarantees. Users of these chains never see a "DAG"; they see fast finality.
Claimed benefits, honestly assessed
- Throughput: real, because parallel units are not discarded. But throughput also depends on networking, execution and storage, so a DAG alone does not deliver huge numbers.
- Low or zero fees: some DAG networks have tiny or no fees, which suits micropayments. Fee-free systems need other spam defenses, such as small proof-of-work per transaction or rate limits based on stake.
- Fast finality: achievable with DAG-based BFT; so is it with many linear BFT chains.
- Decentralization: varies widely. Several early DAG networks depended on coordinators or small permissioned node sets for years. Check who runs consensus today.
How DAGs handle double spending
Because units arrive in parallel, two conflicting transactions spending the same funds can both appear in the graph. Every DAG ledger therefore needs a rule for picking one. Hashgraph orders all events by virtual voting, so the first in consensus order wins. Nano's representatives vote only on the conflicting blocks. Kaspa's GHOSTDAG ordering decides which of two conflicting blocks counts. When marketing says a DAG "prevents double spending", it means its ordering rule does; the graph alone does not.
Where DAG ledgers fit
DAG networks were long pitched for IoT and machine-to-machine payments, because devices could send tiny, frequent value transfers. That remains plausible for data integrity and micropayment use cases, but most IoT products still do not need any ledger; see the honest look at blockchain IoT development. Enterprises tend to pick Hedera for its governance model and predictable fees, covered in the guides to Hedera development and the broader hashgraph algorithm.
Building on a DAG-based network
For developers, the data structure usually matters less than the platform. Questions to ask:
- Smart contracts: Is there a contract layer, and in what language? Hedera supports EVM contracts through its smart contract service; Sui and Aptos use Move; Nano has no general smart contracts.
- Finality and ordering: How long until a transaction is final, and can ordering be influenced (MEV)?
- Tooling: SDKs, explorers, indexers, wallets and auditors familiar with the platform.
- Governance: Who controls protocol upgrades and validator admission?
- Ecosystem: Liquidity, stablecoins and users if your product needs them.
When to choose something else
If your product needs deep DeFi liquidity and the broadest tooling, EVM chains and their rollups still lead; the overview of scaling with layer 2s explains that path. If you only need a tamper-evident log inside one company, a database with signed, hash-chained records is simpler than any distributed ledger, as discussed under private blockchain development.
Frequently asked questions
Is a DAG a blockchain?
Technically a blockchain is a special case of a DAG where each block has one parent. In everyday use, "DAG ledger" means a design where units can reference multiple parents, so parallel activity is kept rather than discarded.
Are DAG ledgers more scalable than blockchains?
They remove one bottleneck, the single chain of blocks, but scalability also depends on networking, execution and state growth. Modern linear chains using DAG-based consensus internally get similar benefits.
Is Hedera a blockchain?
Hedera uses the hashgraph algorithm, a DAG of gossip events, rather than a chain of blocks for consensus. For developers it behaves like a fast public ledger with tokens, consensus messages and EVM smart contracts.
Why did early DAG networks use coordinators?
Pure DAG security models were hard to make safe at low participation, so some projects used a central coordinator as a temporary safeguard. That compromised decentralization and was a common criticism until those networks redesigned their consensus.