Getting this wrong can burn $1 million to $3 million before the first live trade, or lock a firm into venue and routing costs that erase its execution margin. A typical venue connection may cost $100,000 to $400,000 to launch. Building and operating a matching engine can take several times that amount, before regulatory, liquidity and ongoing support costs.
The choice is not simply software versus exchange fees. A venue supplies a rulebook, order book and counterparties. An owned engine supplies control over matching logic and the customer experience, but the operator must provide the surrounding controls, resilience and market access. Compare the costs over three to five years, at realistic volumes, rather than treating the build estimate as the total.
The bill starts before the first fill
A venue route still needs more than a FIX session. The broker or prop firm has to map order types, time-in-force instructions, rejects, fills, cancels and corrections into its own order management and risk systems. It also needs session recovery, sequence handling, drop copy, trade reconciliation and a process for venue notices. Each integration adds test environments, certification work and support coverage.
For a single venue and a conventional product set, a typical implementation takes three to nine months. A small team may need two to six engineers across connectivity, order management, testing and production support.
Loaded annual staff costs can run from $250,000 to $900,000, depending on location and seniority. These figures exclude the business cost of delayed launch and any specialist vendor licence.
An owned matching engine is a wider programme. Engineering covers the matching core, order entry, market data, client gateways, admin tools, surveillance hooks and recovery. Operations needs deployment controls, monitoring, incident procedures and reconciled records. Risk needs pre-trade limits and kill switches that behave consistently during partial failures. A typical build can require eight to 20 full-time roles across engineering, SRE, QA, product and operations.
Scope changes the estimate sharply. A central limit order book with price-time priority is not the same product as a multi-asset system supporting auctions, pro-rata allocation, implied orders or complex order types. A firm should also check whether it is operating an internal broker facility or creating a venue that needs authorisation. That question can change both the project plan and the cost base.
What venue access actually costs
Venue access converts a large share of the build into recurring and usage-based charges. Common items include onboarding, certification, cross-connects, market data, order-entry ports, drop copy and transaction fees. The schedule varies by asset class, venue, message rate and whether the firm is a member, sponsored participant or customer of an intermediary.
Budget $2,000 to $25,000 a month per venue for connectivity and data as an initial planning range, then validate it against the actual fee schedule. Some venues charge much more for data distribution, additional users or premium connectivity. Transaction costs are even less portable: equity fees may depend on shares, notional and liquidity role, while futures and options often price per contract. FX venues may combine access charges with commissions or liquidity-provider terms.
Venue costs also rise with the number of destinations. Each connection has its own certification cycle, session limits, market-data format and incident path. A smart order router may need venue-specific throttles and rules for duplicate orders, stale quotes and partial fills. The operational cost of a route is therefore not just the monthly port charge.
- Include market-data redistribution rights, not just the cost of receiving a feed.
- Model order and cancel rates during volatile periods, when message limits and throttles matter most.
- Price the backup route, disaster recovery site and drop-copy feed separately.
- Check whether fees apply to rejected, modified or cancelled messages as well as executions.
Venue access can still be the cheaper choice when the firm needs external price discovery, established liquidity or a recognised market structure. It is also easier to add or remove destinations than to recreate a liquid market internally. The venue does not guarantee execution quality, though. Measure fill rate, rejection rate, slippage and market impact by instrument and time of day.
What an owned matching engine adds
An owned engine makes the operator responsible for the matching rules and the state of every order. The core must preserve ordering, reject invalid instructions, allocate fills deterministically and recover without losing or duplicating trades. Those properties need tests under load and after faults, not just a fast-path benchmark from a clean test environment.
Typical engineering and implementation costs range from $750,000 to $2.5 million for a production-grade system with a defined product scope. Annual staff costs can reach $1.2 million to $4 million for an eight to 20-person team.
Hosting, observability, disaster recovery and test environments may add $10,000 to $75,000 a month. These are planning ranges, not quotes, and exclude major regulatory work, market-data rights and liquidity incentives.
The engine does not create external liquidity. If customers expect to trade against an aggregated external book, the firm still needs venue connections, routing, hedging or liquidity-provider agreements.
If it matches customer flow internally, it needs controls for conflicts, execution policy, exposure and post-trade records. A lower per-trade venue bill can be offset by the cost of carrying risk or sourcing a hedge.
Performance targets must describe the whole path. A matching decision measured inside one process says little about client-to-acknowledgement latency, which includes gateways, risk checks, network hops and persistence. A venue route adds the distance to the venue and its own queueing. A useful benchmark separates median and tail latency, measures under expected peak load, and records behaviour during failover. For broader platform and execution coverage, Trading Technology tracks the infrastructure decisions that shape these systems.
Benchmark the full operating model
The ranges below are typical planning benchmarks in USD for a single product group and a production setup. They are not audited results. Geography, asset class, membership status, service levels and regulatory scope can move the figures materially.
Typical benchmark ranges
| Measure | Venue route | Owned matching engine | What to include |
|---|---|---|---|
| Initial implementation | $100,000-$400,000 | $750,000-$2.5 million | Engineering, integration, certification and test work |
| Annual team cost | $250,000-$900,000 | $1.2-$4 million | Typical team of 2-6 FTE versus 8-20 FTE |
| Monthly connectivity and infrastructure | $2,000-$25,000 per venue | $10,000-$75,000 for engine stack | Ports, data, hosting, monitoring and recovery |
| Execution and liquidity charges | Venue schedule, often usage-based | External hedging and liquidity terms still apply | Compare by instrument and executed volume |
| Typical technical latency | About 50 microseconds-2 milliseconds from a colocated gateway; 2-30 milliseconds from a remote gateway | About 10-200 microseconds for an in-process match; 100 microseconds-5 milliseconds end to end | Product, network path, load and measurement point change results |
The table is a starting point, not a break-even answer. Put the same assumptions into both cases: expected orders and executions, peak message rate, supported instruments, service hours, staffing location, data entitlements and recovery objectives. Include the cost of the control plane around the engine, not only the matching process.
Then calculate cost per executed million and cost per active account at several volume levels. For a venue route, include variable charges and any intermediary spread. For an owned engine, include build amortisation, ongoing release work, liquidity and hedging. A simple model should show a low, base and high volume case, plus the threshold at which the engine becomes cheaper. If that threshold depends on immediate growth, the model is fragile.
Choose by flow, then test the failure path
Venue access usually fits firms that need broad external liquidity, want to launch quickly or have flow below the engine's break-even point. An owned engine can make sense where the firm has sustained volume, a distinct matching model or a need to control the order lifecycle closely. Some operators use both: internal matching for a defined flow, with external routing for residual orders and price discovery.
Do not use headline latency as the deciding metric unless the business can monetise the difference. A few hundred microseconds may matter to a colocated strategy. It may be irrelevant to a workflow dominated by client network distance, risk checks or human decision time. More useful operational measures include peak orders per second, reject and duplicate rates, recovery time, reconciliation breaks and the percentage of orders that reach the intended destination.
Run a failure exercise before committing. For a venue route, test a dropped session, stale market data, sequence gaps and a venue-wide reject. For an owned engine, test process failure during a partial fill, lost persistence, a failed risk service and recovery from a known state. In both cases, confirm that kill switches stop new risk without preventing the firm from cancelling or reconciling live orders.
Build a cost model with three- to five-year totals and a volume sensitivity, then validate each line with vendors, venues and internal owners. For an owned engine, include the people who will carry overnight and weekend incidents.
For a venue route, include the cost of every connection needed for resilience and product coverage. The cheaper option on a launch budget can become the more expensive one after staffing, liquidity and support are counted.
This week, export 30 days of order, execution and venue invoices and put both routes into a volume-tiered cost model.
Frequently Asked Questions
How much does it cost to build a matching engine?
A typical production build can cost $750,000 to $2.5 million before major regulatory work, market-data rights or liquidity costs. Annual staffing and operating costs are separate and can exceed the initial build over a three- to five-year period.
Is it cheaper to connect to a trading venue?
Venue access often has a lower initial cost, with typical implementation ranges of $100,000 to $400,000. Recurring connectivity, data and transaction fees vary by venue and product, so compare total costs at expected trading volumes rather than comparing launch budgets alone.
What should a broker benchmark before choosing an execution model?
Measure total cost per executed volume, peak order and cancel rates, fill and reject rates, and end-to-end latency at both median and tail levels. Test recovery from session loss, stale data and partial fills before relying on either route in production.
Platforms · Software · Providers · Markets