What is The Graph (GRT)?

Blockchains are terrible databases. They are excellent at agreeing on an ordered list of transactions and hopeless at answering a question like "show me every trade this address made in the last month, sorted by size". Getting that answer means scanning the chain and reorganising the data, which every application would otherwise have to do for itself.

The Graph does it once, as a network. Developers define a subgraph — a specification of which events to watch and how to structure them — and indexers compete to serve queries against it, staking GRT and earning fees. Curators signal which subgraphs matter and delegators back indexers without running infrastructure. It is genuinely load-bearing: a large share of DeFi front-ends depend on it, usually invisibly.

What Horizon changed

The Graph spent years as one product — subgraphs, queried by GraphQL. Horizon, live since December 2025, made it a modular platform hosting several data services at once: Substreams for high-throughput streaming, Token API for balances, transfers and pricing, and analytics alongside the original subgraphs.

The revenue split in the quarter before tells you why this mattered. Substreams revenue grew fourfold while traditional subgraph query fees fell 8.7% quarter on quarter. The old product was shrinking and the new one was growing, and Horizon was the restructuring that let GRT be paid for both rather than only the declining half.

x402 and the agent thesis

The Graph integrated x402, a payment protocol allowing AI agents to pay for on-chain data queries in real time using GRT or stablecoins. The argument behind it is specific: an autonomous agent needs data, cannot open an account or sign a monthly contract, and needs to pay per request at machine speed.

Whether agents become a large buyer of on-chain data is unproven and the reasoning is sound. It is also one of the few AI pivots in crypto where the company already sells the product the AI would need, rather than inventing one to fit the narrative.

• Query fees paid by applications consuming data — the core revenue, now across several services rather than one.
• Indexer stake: operators bond GRT as a guarantee of honest service, and can be slashed.
• Curation: signalling GRT on subgraphs to direct indexing resources and earn a share of their fees.
• Delegation to indexers, for holders who want exposure without running infrastructure.

The Graph's value capture is better than most infrastructure tokens because the service is metered and paid for in GRT at the point of use. The structural weakness is the same one Chainlink has: a large share of consumption happens through hosted gateways and subsidised tiers, so usage and token demand are connected but not tightly.

GRT launched in December 2020 with 10 billion tokens and annual issuance to indexers, offset by burns — a portion of query fees and curation taxes are destroyed. Net supply depends on how much querying happens against how much is emitted.

For years emissions clearly exceeded burns, because query revenue was small and the network was subsidising indexers to show up. Horizon's broader product line is the attempt to grow the fee side of that equation, and the Q4 2025 split — Substreams up fourfold, subgraphs down 8.7% — shows the mix shifting even where the total is not yet decisive.

The hosted service sunset

For most of its life The Graph ran a free hosted service alongside the decentralised network, and almost everyone used the free one. Sunsetting it forced users onto the paid network and was the single most important economic decision in the protocol's history — it converted usage into revenue, at the cost of losing the users who were only there because it was free.

GRT is staked in two distinct roles and the difference matters. Indexers bond GRT as collateral for serving queries honestly and can be slashed for misbehaviour — that is a business requiring infrastructure. Delegators back an indexer without running anything, sharing in that indexer's rewards minus their cut.

Delegation has specifics that catch people out. There is a deposit tax on delegating, an unbonding period of several weeks before withdrawal, and returns depend entirely on the indexer's query volume and fee settings. An indexer with no queries pays you almost nothing regardless of how much you delegate.

Curation is a third activity again: signalling GRT on a subgraph to direct indexing toward it, earning a share of its query fees, on a bonding curve where early signal earns more and the position can lose value if the subgraph is abandoned.

The original product is shrinking

Subgraph query fees fell 8.7% quarter on quarter in Q4 2025. Substreams growing fourfold is the right response, and the network is in the middle of replacing its main revenue line rather than adding to it — a transition, with the execution risk that implies.

Competition from doing it yourself

Alchemy, QuickNode and Goldsky sell indexed data as a conventional service, faster to integrate and with support. Large applications increasingly run their own indexers, because at scale the economics favour it. The Graph's decentralisation matters to a subset of customers and is not what most are buying.

Free tiers weaken the link to the token

Much consumption runs through gateways and subsidised access rather than direct GRT payment. Usage grows; token demand grows less. The same structural gap that follows Chainlink around.

The agent thesis is unproven

x402 is a well-designed bet on AI agents becoming significant buyers of on-chain data. They may not, or they may buy it from a centralised provider with a simpler API. Judge it on revenue rather than on the integration existing.

Delegation is less passive than it looks

A deposit tax, a multi-week unbonding period and returns that depend on your indexer's actual query volume. Delegating to a large indexer with no traffic earns close to nothing, and the position is illiquid while you find out.