Four ways to get a Sui RPC node, compared: the Foundation's public endpoints, your own full node, a managed RPC node provider, or a dedicated node. Setup, limits, history and cost.
"Sui RPC node" is one phrase for four different things: the Foundation's public endpoints, a full node you run yourself, a managed shared gateway, and a dedicated node someone else operates for you. They differ on almost every axis that matters in production, and the right one changes as an application grows. This post lays the four side by side and gives a rule for choosing.
TL;DR:
| Foundation public endpoint | Your own full node | Managed shared gateway | Dedicated node | |
|---|---|---|---|---|
| Time to first call | Minutes | Hours to days (snapshot download, restore, catch-up) | Minutes | 6 to 24 hours |
| Protocols | gRPC and GraphQL on mainnet and testnet, gRPC only on devnet (JSON-RPC retired) | Whatever you build on top | gRPC, GraphQL, JSON-RPC | gRPC, plus what you configure |
| Rate limit | Strict, undocumented | Hardware-bound | Per plan, 10 to 1,000+ req/s | None shared; hardware-bound |
| History | Live state, plus a public archival endpoint | Whatever you retain and pay to store | Full, since genesis | Archival mode on request |
| Who patches it | Sui Foundation | You | Provider | Provider |
| Cost shape | Free, capped | Hardware plus people | Per plan, from free | Flat monthly, per region |
| Production use | Not recommended by Sui | Yes, with a team | Yes | Yes |
The gateway and dedicated-node columns describe Inodra; other providers on Mysten's official list vary, particularly on history depth and JSON-RPC continuity.
Sui Foundation runs public-good gRPC, GraphQL and archival endpoints on mainnet and testnet, and gRPC only on devnet. The gRPC hosts are fullnode.mainnet.sui.io:443 and the testnet and devnet equivalents; the RPC providers page in Sui's docs is the reference for the rest. They are free, they are rate-limited, and Sui's documentation says plainly that they are for development and testing rather than production. JSON-RPC was retired from them in July 2026.
Use them for a first GetCheckpoint, a tutorial, or a script you run once. Do not put them behind a product: the limits are not published, they change, and there is no one to call when they do.
The full treatment is in Sui full node requirements in 2026. The short version: 8 cores, 128 GB of RAM and 4 TB of NVMe is the suggested minimum, disk grows past 7 TB without pruning, a new node should be restored from a snapshot rather than synced from genesis, and in our experience the operations budget (upgrades, monitoring, incidents, redundancy) exceeds the hardware budget within a year.
It makes sense when you already run chain infrastructure, when you need a configuration no provider offers, or when read volume is enormous and steady enough that fixed hardware beats per-request pricing and you have the people to keep it healthy. For everyone else it is the most expensive way to get an RPC endpoint.
A shared gateway is a provider's fleet of nodes behind one endpoint per protocol, metered by API key. On Inodra that means gRPC, GraphQL and JSON-RPC on the same key, every request routed to a live node or to archival storage automatically, and full history since genesis on all three. Rate limits and monthly credits scale by plan, starting with a free tier that needs no credit card.
This is the right default for most applications. There is no node to patch, history is someone else's disk, and the endpoint is already redundant across machines. The trade-offs are the ones you would expect from anything shared: a per-plan rate limit, and neighbours on the same hardware. For the large majority of Sui apps neither is ever the bottleneck.
The endpoints page has the URLs, the per-plan limits and a copyable first call for each protocol.
A dedicated node is a full node reserved for one customer: bare-metal hardware with NVMe, a private endpoint with no shared rate limit, and a configuration you control (archival mode, RPC tuning, custom patches) while the provider keeps it patched and monitored. On Inodra it is priced per region (EU, US and Singapore, with more on request), delivered within 6 to 24 hours, and billed separately from the plan so the two stack.
That last point is the pattern we see most: keep the shared gateway for the bulk of reads and for history, and route the latency-sensitive or highest-volume path to the dedicated node. You get isolation where it matters without giving up the parts of the gateway that are hard to replicate on one machine, such as archival depth across every protocol.
What is a Sui RPC node? A Sui full node that exposes an API (gRPC, GraphQL or JSON-RPC) so applications can read chain state and submit transactions without running a node themselves. On Sui the go-forward interface is gRPC; JSON-RPC is served only by providers that run it on their own infrastructure.
Is the Sui Foundation public RPC endpoint free? Yes, with strict rate limits. Sui's documentation describes the public-good gRPC, GraphQL and archival endpoints as suitable for development and testing, not for production workloads, and they no longer serve JSON-RPC.
Do I need a dedicated Sui RPC node? Only when a shared gateway's rate limit or latency profile is the bottleneck, or when you need a custom node configuration such as archival mode or a patch. Most applications never reach that point; those that do usually see it in their own request graphs first.
What it takes to run a Sui full node in 2026: Mysten's hardware spec, how disk grows with and without pruning, snapshot restore versus syncing from genesis, and where the real cost sits.
Sui JSON-RPC is deprecated and gone from public fullnodes. Inodra still serves the API on mainnet and testnet, with history back to genesis.
All 14 providers on Mysten Labs' official Sui gRPC list compared on gRPC, GraphQL, archival depth, streaming and webhooks, plus JSON-RPC continuity for legacy clients.