Running a Vault gateway: a practical operator's guide

A SolanaVault gateway is a node that relays JSON-RPC requests from light clients to the storage network, charges the client per query, and keeps 85% of the revenue. Storage and consensus nodes split the remaining 15%. If you are reading this thinking about running one, this guide covers what it takes, what it pays, and what we recommend you do in the first month.

We will be specific about numbers. Your mileage will vary, but the order of magnitude is what we observe on the production network in 2026.

What a gateway actually does

A gateway is the public face of the network. Light clients connect to it over HTTP or WebSocket and call standard Solana JSON-RPC methods. The gateway:

1. Authenticates the light client and decrements its prepaid balance.
2. Looks up the slot in question on the Kademlia DHT to find replicas.
3. Issues parallel requests to k replicas, runs the Byzantine quorum check, and returns the verified response.
4. Caches hot blocks locally for fast repeat hits.
5. Submits periodic attestations to the consensus layer.

The gateway is a service operator. It does not have to store the entire ledger. It does not have to participate in deep consensus. Its job is to be a fast, reliable, well-connected relay.

Hardware recommendation

For a single-region gateway serving up to about 200 queries per second, our reference hardware is:

• 16 vCPU (AMD EPYC or equivalent)
• 64 GB RAM
• 1 TB NVMe SSD for the local hot cache
• 10 Gbps network with low intra-region latency

You can run smaller. We have operators on 8 vCPU / 32 GB that handle 80 qps comfortably. The bottleneck on a tuned gateway is almost always the network, not the CPU.

The interesting hardware choice is the NVMe. The hot cache holds compressed blocks, so the effective storage capacity is 15-25x the raw size. A 1 TB NVMe holds the equivalent of 15-25 TB of raw block data, which is enough hot capacity for most workloads.

Network and peering

DHT discovery is helpful but not sufficient for production latency. We recommend:

• Anycast or a regional DNS for client-facing traffic.
• A second NIC or VLAN for the NNG peer mesh — keeping intra-network gossip off the public path matters.
• Outbound peering with at least one storage cluster in your region.
• A second region as warm standby if your SLA target is below 99.9%.

The cost of running this is dominated by egress. On a major cloud provider you should budget about USD 0.05-0.09 per GB of egress to clients. On a colocated provider it is closer to USD 0.005 per GB. For high-traffic gateways, colocation usually pays for itself within a quarter.

Reputation and onboarding

A new gateway starts with zero reputation. The DHT picks gateways for routing partially by latency and partially by reputation. New gateways will see low traffic for the first 7-14 days while they earn reputation through correctness.

The fast path to reputation is to serve a small but reliable fraction of traffic. We recommend:

1. Run the gateway in single-replica mode for the first 24 hours. You will not be the primary path for any client, but you will be measured.
2. Enable consensus participation on day two. Sign attestations on the slots you do see.
3. Apply for a verified-operator flag once you have served 100,000 requests with no faults. The reputation boost is significant.

Do not rush. Operators who push hard on traffic before their reputation stabilizes end up with worse routing weights long-term.

Pricing your gateway

Gateways set their own per-query price, subject to a network-wide floor. The current floor is 0.0001 SOL per 1,000 queries. Most production gateways price between 0.0002 and 0.0008 SOL per 1,000 queries depending on region and performance tier.

The pricing is exposed to light clients and shows up in their gateway selection logic. Cheap gateways get more traffic and earn less per query. Premium gateways get less traffic but earn more.

Our advice: start at the midpoint (0.0004 SOL/1k) for the first month. Look at your traffic, your costs, and your latency percentiles. Adjust monthly. Do not chase traffic by undercutting — the bottom of the pricing tier is a race to the floor.

Revenue at a glance

A gateway serving 100 queries per second at 0.0004 SOL/1k queries earns about:

• 100 qps × 86,400 s/day = 8.64M queries/day
• 8.64M / 1k × 0.0004 SOL = 3.46 SOL/day in gross revenue
• After the 15% network split: 2.94 SOL/day net to the operator

At a conservative USD 140/SOL that is about USD 410/day or USD 12,300/month gross to the operator. Subtract:

• USD 300-500/month for the reference hardware (colo or reserved cloud).
• USD 1,500-3,500/month for egress depending on cloud vs colo.
• USD 100-200/month for monitoring, alerting, and reserve margins.

Net margin: roughly USD 8,000-10,000/month for a single gateway at 100 qps. The math improves nonlinearly with scale because the fixed costs do not grow proportionally.

What can go wrong

The honest list:

• Soft faults pile up after a network blip. Set up alerting on your soft-fault rate. We recommend paging if it exceeds 0.5% over a 5-minute window.
• Hot cache eviction patterns matter. If you serve a large historical workload alongside a tip workload, the tip can evict useful historical pages. Run two cache tiers.
• Egress bills surprise people. On a major cloud, a successful 200-qps gateway can bill USD 3,000-5,000/month in egress alone. Colocation is much cheaper here.
• SOL price volatility moves your USD revenue significantly. Plan for it. Some operators denominate their reserve margins in USD, others in SOL.

A minimum-viable operator playbook

If you want to start in the next 30 days:

1. Apply for an operator slot via the operators@cryptuon.com channel. You will get reviewed and admitted in 3-5 business days.
2. Stand up the reference hardware in your favorite region. Colo strongly preferred.
3. Run the gateway in single-replica mode for 24 hours. Watch the metrics.
4. Enable consensus participation. Set your initial price at 0.0004 SOL/1k.
5. Apply for the verified-operator flag at 100,000 served requests.
6. Reach 100 qps within 30-45 days. The DHT will route you there if your latency and correctness hold.

If you are operating at this kind of scale already on other networks, the SolanaVault economics are favorable. We are happy to walk through the math with you. The operators who have moved from running validators or hosted RPC nodes have generally been positively surprised by how clean the revenue accounting is — the on-network payment channels do not need an invoicing department.

The network needs more operators. If you have the hardware and the operational discipline, come run a gateway. The economics are real.