Receiving shreds without a provider: what it actually takes

Every provider on the pricing index, us included, is doing something you are technically allowed to do yourself. Turbine is not gated. Run a node, join gossip, and the network will send you shreds for free. Since we sell the alternative, you should treat this page skeptically, so it is written to be checkable: here is the path, each real cost on it, and the two situations where doing it yourself genuinely wins.

The path

You cannot point a bare UDP socket at the network and collect shreds. Turbine delivers to participants it knows about. Becoming one means running a Solana node that joins gossip, publishes a TVU address, and holds its place in the Turbine tree. That is the origin of the shredstream-proxy architecture Jito open-sourced and everyone copied: a node subscribes to the firehose legitimately, a proxy forwards raw packets onward to the machines that trade. Building the receive side yourself means becoming that first machine.

The node does not need to vote. It needs to be seen. It also needs to keep up: a box that joins gossip but cannot hold the packet rate becomes a dead branch, and dead branches get routed around.

The position problem, which is the actual product

Here is the part the DIY arithmetic usually misses. Turbine's tree is stake-weighted. High-stake validators sit near the root and hear shreds first; an unstaked listener sits in the canopy, downstream of however many relay hops the tree assigns it, and its position reshuffles as the tree does. Position is latency, and stake buys position.

So the unstaked-node path hands you the median experience of the network, not the front of it. Your own node will trail a well-positioned tap by a spread that moves around and is, on bad reshuffles, exactly the kind of tens-of-milliseconds gap this whole market exists to close. It will also simply miss packets, which is what the repair protocol exists to fix, and repair is a request round-trip, which is to say: latency, again, at the worst moments.

Closing the position gap yourself means stake, peering agreements with validators, or both. At that point you have not avoided the shred-provider business. You have entered it.

The arithmetic

The monthly bill for the honest version of DIY: a machine that keeps up with mainnet's packet rate plus headroom for bursts, which is real hardware in real colocation, low hundreds to around a thousand a month depending on the facility; the bandwidth of the full firehose arriving whether or not you use it; and the engineering time, which is the number everyone rounds to zero and is the largest line. Somebody has to notice when a network upgrade changes shred variants, when the box falls behind, when gossip health degrades quietly on a Saturday. Call it a slice of one engineer, priced however your engineers are priced.

Against that, published managed-feed prices run from our $100 a month to $1,000-plus per region elsewhere, delivery included, position included, upgrades absorbed by someone whose whole job that is. The economics page shows our own cost structure if you want to audit the comparison from the other side.

When DIY genuinely wins

Two situations, and they are real.

You already run validator infrastructure. Then the incremental cost of tapping your own nodes is close to zero, your stake already bought you tree position, and paying a third party would be strange. Most serious market-making shops on Solana are in this category, which is worth knowing before you assume your competition buys what you buy.

You need something no feed sells: a tap in a metro no provider serves, custom filtering fused into the receive path, packet-level research on Turbine itself. Products have shapes; when your requirement leaves the shape, build.

Everyone else is choosing between a monthly invoice and a part-time infrastructure job with worse latency. That is the honest comparison. It happens to favour the thing we sell, which you were right to be suspicious about; the numbers above are the defense.