The free Hyperliquid API has hard limits. Most info endpoints give you about one call a second. One IP address gets 1,000 WebSocket subscriptions. Hyperliquid lists 234 perp markets and 326 spot pairs, so that budget goes fast.
So you need paid access. Ask most providers what a Hyperliquid node costs and you get a contact form.
Dwellir sells Hyperliquid L1 access two ways: a metered shared endpoint, or a Dedicated Node Cluster. Every price is on this page. This post explains what the cluster adds and which workloads need it.
This post covers Hyperliquid L1, also called HyperCore. That means the order book, fills, order statuses, and the info endpoint. The cluster also serves HyperEVM JSON-RPC. New to the two layers? Start with What Is Hyperliquid.
Shared endpoint or dedicated cluster
| Shared endpoint | Dedicated Node Cluster | |
|---|---|---|
| Price | API plan ($49, $299, or $999/month) plus add-ons | Starts at $4,000/month |
| gRPC Streaming API | $299/month add-on, metered | Included, unmetered |
| Order Book Server | $199/month add-on, metered | Included, unmetered |
| Billing | 1 response or streamed message = 1 API credit | Flat. No per-response billing |
| Nodes | Shared pool | Starts with 2 nodes |
periodic_abci_states over gRPC | Not available | Available |
| SLA | Standard plan terms | Extended SLAs available |
The shared endpoint is the right place to start. The API plans include 25M, 150M, and 500M responses a month. Overages cost $5, $3, and $2 per million. Most Hyperliquid workloads never need more.
What the cluster adds
Throughput that holds at peak hours
Nodes perform better when they are not shared. A cluster carries only your traffic, so you get more consistent throughput and response times during the busiest hours of the trading day.
Streams split across backends
A cluster starts with 2 backend nodes. We can help you spread your gRPC and Order Book subscriptions across them, so one heavy stream cannot slow the rest.
Peering you can rely on
A Hyperliquid node depends on the peer that feeds it. The community root peers are shared and oversubscribed. When a node loses its peer, it falls behind the tip and restarts sync.
Cluster nodes peer through the same gossip peers we sell as Hyperliquid Peering. Teams that run their own node pay $500 a month for one slot on those peers. On a cluster, that peering is part of the service.
An SLA and a team that knows your workload
Extended SLA terms are available on the cluster, along with priority support. We also configure the nodes around your access pattern: node flags, the feeds written to disk, regions, and load balancing. A shared pool has one configuration that has to work for every customer.
The periodic_abci_states stream
Hyperliquid nodes write periodic ABCI state dumps, which are binary snapshots of full node state. They are the input for rebuilding order books from a known point. Direct gRPC access to periodic_abci_states is available only on a Dedicated Node Cluster. The shared endpoint offers generated order book snapshots, not the raw state stream.
No per-response billing
On a shared endpoint, every streamed message costs one API credit. Hyperliquid volume is not steady. A volatile week means more book changes, more fills, and a bigger bill. You learn what that week cost after it happened.
The cluster starts at $4,000 a month, billed flat. The gRPC Streaming API and the Order Book Server are included, with no usage charges. At their standalone prices those two add-ons cost $498 a month before any usage.
Why a cluster and not a single node
The obvious middle step is to rent one Hyperliquid node of your own. We do not sell single nodes, for three reasons.
Redundancy. A single node is a single point of failure. If its server has a fault or the node falls out of sync, your feed stops until it recovers. A cluster starts with 2 backends. If anything goes wrong on one server, the other backend takes the load.
Zero-downtime upgrades. Node software needs updates and servers need maintenance. A node that restarts must resync with the network before it serves data again. On a cluster, we upgrade one backend at a time while the other keeps serving your traffic.
Performance. Hyperliquid's official machine specification for a non-validating node is 16 vCPUs, 128 GB of RAM, and 500 GB of SSD. One node serves a fixed number of requests per second, and you cannot add capacity to it. A cluster spreads your workload across its nodes, with headroom on each.
When you need a cluster
Three workloads point to a cluster.
High Order Book volume. One L2 order book subscription produces about 13 to 14 messages a second, or 1.1 to 1.2 million a day. About 28 subscriptions add up to 1 billion messages a month. On a shared endpoint with the Scale plan, that costs $2,198 a month: the $199 add-on, the $999 plan, and $1,000 in overage. Each extra 100 million messages adds $200, and at about 1.9 billion messages a month the metered bill passes the $4,000 starting price of a cluster. The Order Book usage calculator recommends a cluster above 1 billion messages a month.
Latency-sensitive trading. Nothing else runs on a cluster's nodes, so all of their capacity goes to your requests. That gives you the best performance and the lowest latency we offer. If a slow response means a chart redraws late, a shared endpoint is fine. If it means a worse fill, or a missed one, a cluster pays for itself in outcomes that never show up on the infrastructure invoice.
State from periodic_abci_states. If you rebuild order books or other state from ABCI state dumps, a cluster is the only way to stream them over gRPC.
Good fit for a cluster:
- Market makers quoting continuously
- High-frequency strategies and latency-sensitive execution
- Trading firms where performance consistency is the product
- Exchanges, custodians, and anyone with an uptime commitment to their own users
- Teams feeding many downstream consumers from one Hyperliquid feed
- Workloads that need
periodic_abci_statesto rebuild state
What about running your own node?
For some teams, running their own node is the right choice.
You need 16 vCPUs, 128 GB of RAM, 500 GB of SSD, and Ubuntu 24.04. Ports 4001 and 4002 must be reachable from the internet. Tokyo is the recommended location, because most of the network sits there.
The disk surprises people. With default settings a Hyperliquid node writes roughly 100 GB of logs a day. With every data type written to disk, as on our nodes, it writes about 2.6 TB a day. That fills the recommended 500 GB SSD in under 5 hours, so storage and pruning become a daily job.
One self-hosted node has no failover and a fixed capacity. A second node, failover, and load balancing are yours to build. You also need stable peers. If you run your own node, Hyperliquid Peering gives it a reserved slot on our peers for $500 a month.
Questions to ask before you buy
Ask any provider, including us, the same five things. Get the answers in writing.
- Is this one node or a pool of nodes? If it is one node, what happens when my workload outgrows it?
- How many nodes, and what happens to my traffic while one of them is being updated?
- Is billing flat, or is there a rate limit or usage charge behind the headline price?
- Which data services are included, and which are separate line items with usage charges on top?
- What SLA applies, and which peers do the nodes sync from?
The bottom line
Start on a shared endpoint. An API plan plus the gRPC or Order Book add-on covers many Hyperliquid workloads.
Move to a Dedicated Node Cluster when your business depends on these feeds. Market makers, execution engines, and services with their own uptime commitments get the lowest latency and highest throughput we offer. A second backend takes the load if one server fails, and upgrades happen without downtime. The cluster also includes unmetered gRPC and Order Book access, the periodic_abci_states stream, our peering, and an SLA. Clusters start at $4,000 a month and 2 nodes.
The Dwellir Hyperliquid pricing page lists every option. The Hyperliquid docs cover the endpoints and data feeds. To start on a shared endpoint, create an account. To size a cluster around your regions and SLA terms, contact our team, email us, or DM us on Telegram.
