Calculate the cost of self-hosting shred delivery

9 cost lines replace one server price

The self-hosting calculator compares a complete 30-day operating model with a subscription. It includes traffic, ingest servers, receiver servers, storage, monitoring, engineering, on-call work, amortized setup, and expected failure cost.

The shredstream.sh reference is $100 for one IP and port for 30 days. Its published internal cost to serve one destination is about $12 to $17 under the stated transit and amortized-ingest assumptions. That is not a claim that a buyer can build an independent ingest service for $17.

Method

Let V be monthly traffic in TB, T be transit cost per TB, H be engineering hours, R be loaded hourly rate, O be on-call hours, A be amortized setup per month, and F be expected failure cost. Fixed monthly infrastructure categories are I.

traffic_cost = V * T
labor_cost = H * R + O * R
monthly_cost = traffic_cost + sum(I) + labor_cost + A + F

Expected failure cost is probability times impact across named failure classes:

expected_failure_cost = sum(probability_i * impact_i)

The tool reports cash infrastructure, allocated labor, expected risk, and total separately. It also reports the break-even engineering hours against a subscription price:

break_even_hours = max(0, subscription_price - nonlabor_cost) / hourly_rate

That value is an arithmetic threshold, not a forecast of how much work self-hosting requires.

Published relay arithmetic

The measured feed is 17.6 TB over 30 days. The published transit assumption is $0.40 per TB:

transit = 17.6 * 0.40 = $7.04
amortised ingest = $5 to $10
total = $12.04 to $17.04
published rounded range = $12 to $17

This is the marginal serving view for one subscriber on an already operating relay. It excludes the customer's receiver and does not represent the cost to establish network position, develop ingestion, run an organization, or absorb all shared failure risk.

The service price minus that rounded range is not pure profit. Gross margin and complete company profit are different accounting levels. The product source publishes an 85 percent gross-margin figure under its defined model.

Worked self-hosting example

Assume a team estimates these hypothetical monthly inputs:

ingest server and network services = $300
receiver and storage = $180
monitoring = $70
engineering = 8 hours * $120 = $960
on-call allocation = 2 hours * $120 = $240
setup amortisation = $100
expected failure cost = $150
traffic = 17.6 TB * $0.40 = $7.04

Total 30-day cost is $2,007.04. Cash infrastructure before labor and expected failure is $657.04 in this example. Labor is $1,200. The $100 setup allocation and $150 risk value complete the total.

These are user-entered example values, not claims about a particular cloud, engineer, or Solana network position. Replacing the hourly rate with zero because an employee is salaried would omit an opportunity cost. Replacing it with the employee's entire salary would overallocate work.

Define self-hosting scope

Three scopes are often mixed. Receiver self-hosting means operating the UDP consumer after a provider sends data. Relay self-hosting means maintaining an upstream feed and forwarding it. Independent network ingestion means obtaining and sustaining a position from which useful shreds arrive.

The $100 product already requires a customer receiver. A fair build-versus-buy comparison adds only costs that differ. If both options need the same decoder and storage, place those in a shared section rather than charging them to one side.

If a provider includes decoding, replay, several regions, or submission, its scope differs from raw delivery. Compare capabilities before totals.

Labor needs dated estimates

Separate initial implementation from recurring operations. Amortize setup over the expected useful period. Count upgrades, incident response, capacity review, security patches, key or credential work, and testing as recurring categories when applicable.

Use loaded hourly cost, not take-home pay. Loaded cost can include employer taxes, benefits, equipment, management, and overhead. The exact method depends on the organization, so the tool accepts the value rather than declaring one universal rate.

Record estimate owner and date. An integration that needed little work in a quiet month can require concentrated effort during a protocol or kernel change.

Failure cost needs a range

Some outages only delay research. Others miss trading opportunities or corrupt an index. Enter low, base, and high impact rather than one precise number. Keep revenue loss separate from speculative gross opportunity.

A raw UDP system can fail through upstream path loss, firewall changes, NIC drops, kernel overflow, application backlog, decoder changes, storage exhaustion, clock error, and missing monitoring. Redundancy changes probabilities and adds infrastructure.

Expected value does not describe worst-case tolerance. A rare loss can remain unacceptable under custody, oracle, or protocol duties even when its monthly expected dollar value is small.

Limitations

The calculator uses entered estimates and cannot price engineering quality, institutional knowledge, vendor support, contracts, regulatory duties, or strategic control. A team may self-host for research freedom even when arithmetic cost is higher.

The 17.6 TB traffic figure is a measured projection dated 2026-08-31, not a future maximum. Transit prices vary by provider and path. Ingress can be free while downstream services are metered.

Expected failure cost is especially uncertain and should remain visible as an assumption. The tool does not claim that subscribing removes receiver failures or that self-hosting guarantees independence.

Use sensitivity ranges. If the decision changes when engineering moves from four to six hours per month, the estimate needs better evidence before a long-term commitment.