Market data feeds: what every tick really costs

Trading screens displaying market data feeds and live price ticks
Photo by Rafael Minguet Delgado on Pexels

The wrong feed has a cash price

Get market data feeds wrong and a broker can waste $50,000 to $500,000 a year in licensing and infrastructure spend. That typical planning range covers duplicated subscriptions, unnecessary exchange ports, unused user entitlements and systems built for a latency target the business does not need. It does not include losses caused by stale prices, rejected orders or incorrect execution decisions.

The cost of a tick is not simply the exchange fee divided by message count. A tick has a licence, a delivery path, processing capacity, monitoring requirements and rules governing who or what may consume it.

A quote sent to a retail screen and the same quote consumed by an automated strategy can trigger different charges. Counting only the feed invoice misses the operational and contractual costs attached to each use.

Start with the use case. A retail broker may need broad, dependable prices for thousands of screens, with latency measured in tens of milliseconds. A prop firm running short-horizon strategies may need direct feeds, deterministic processing and recovery measured in microseconds or milliseconds. Those are different products, even if both teams ask for the same instrument list. Coverage of broker infrastructure often turns on this distinction between a data product and the systems required to deliver it.

What the invoice contains

Exchange charges are only one line. The contract may distinguish professional and non-professional users, display and non-display use, internal use and redistribution. A broker displaying prices to clients may owe per-user fees and reporting obligations.

A firm using prices in algorithms, valuation or risk may face separate non-display charges. Redistribution can add another layer, with obligations linked to the downstream audience and the way data is packaged.

A vendor can simplify the number of technical connections, but it does not automatically simplify the licensing. A consolidated feed may bundle transport and normalisation while passing exchange charges through separately.

Ask whether the quote includes exchange fees, recovery data, historical data, test environments, additional devices and permission to redistribute. Confirm whether a failover feed, backup site or second application counts as another use.

Typical planning ranges below are in US dollars. They are budget guides, not exchange tariffs or a quote. Actual charges vary by venue, user classification, region, product and contract.

  • Display entitlements: typically $10 to $150 per professional user, per venue, per month. Non-professional rates can be lower, but classification and evidence requirements matter.
  • Non-display use: roughly $100 to $10,000 or more per application, strategy or venue each month. Some venue schedules use other measurement units, such as devices, cores or enterprise agreements.
  • Vendor access and normalisation: around $1,000 to $20,000 a month for a modest feed set, before pass-through exchange fees. Broad multi-asset coverage can exceed this range.
  • Engineering and operations: one specialist role can represent about $120,000 to $250,000 a year in loaded employment cost, depending on location and seniority.

Maintain a rights inventory beside the technical inventory. It should name each venue, product, consuming system, user class, redistribution path and accountable owner. Reconcile it against billing and access logs at least monthly. Entitlement drift is easy to miss: a test terminal becomes a desk tool, a new pricing service reuses a licensed feed, or a failover site stays live after a migration.

Typical budget benchmarks

Cost itemTypical planning rangeBilling basisCommon cost driver
Professional display entitlement$10 to $150 per monthUser, venue and monthNumber and classification of viewers
Non-display exchange use$100 to $10,000+ per monthApplication, strategy or venueExchange rules and use case
Vendor feed and normalisation$1,000 to $20,000 per monthPackage, connection or serviceMarkets, depth, redundancy and support
Network ports and cross-connects$500 to $5,000 per monthPort, circuit and siteColocation, bandwidth and resilience
Feed handler and production infrastructure$2,000 to $20,000 per monthCompute, hosting and supportThroughput, latency target and location
Initial integration and certification$5,000 to $50,000 one-offProject or venueProtocol, test scope and vendor work

These are typical ranges for planning, not audited figures for a named provider. They exclude major exchange pass-through charges where those are invoiced separately, as well as internal salaries except where listed.

For a small broker, the user count and redistribution rights may dominate. For a prop firm, direct connectivity, redundancy and engineering time can outweigh display fees.

Build the budget by venue and consumer, not by supplier invoice. A vendor line item can hide several exchange products, while one exchange agreement can cover multiple applications with different rights. The Trading Technology infrastructure desk covers the platform decisions behind those costs, including the trade-offs between vendor aggregation and direct connections.

Delivery adds cost beyond the subscription

Direct feeds usually provide greater control over depth, sequencing and latency, but require more operational ownership. The team must handle multicast, packet sequencing, gap recovery, snapshots, redundant paths and protocol changes.

A vendor-normalised feed can reduce that workload and simplify symbol mapping, though it may add a hop, impose a schema or limit access to venue-specific fields. The right choice depends on the value of control relative to the cost of building and operating it.

Throughput is not just an average message-rate problem. A feed averaging 20,000 updates per second may burst to several times that rate during volatile periods. Engineers should size buffers, CPU and network paths for bursts, then test recovery while the primary stream remains busy.

Measure p50 and p99 processing delay, queue depth, sequence gaps and time to restore a complete book. A low average latency says little if a queue grows during the opening auction. A practical delivery checklist is:

  • Record latency at feed receipt, book update, pricing output and client or strategy consumption. Keep clocks synchronised and document whether measurements include network transit.
  • Track gaps, duplicates, out-of-order messages, stale instruments and snapshot recovery time by venue and line.
  • Test primary and backup paths independently, then test failover under load. A warm standby that has not been exercised is an assumption, not a control.
  • Set alert thresholds around business impact. A brief gap in an inactive instrument is different from stale best bid and offer data across a heavily traded symbol.
  • Keep protocol versions, symbol maps and venue calendars under change control. Mapping errors can look like a feed failure while returning plausible but wrong prices.

Conflation can reduce bandwidth and downstream processing, particularly for screen display, but it discards intermediate updates. That may be acceptable for a retail interface and unacceptable for a strategy using order-book changes. Splitting consumers across products or processing paths can preserve detail where it has value without paying to deliver every update to every user.

Measure cost per usable tick

A useful internal measure is cost per million usable updates, calculated separately for each consumer class. Include exchange and vendor fees, connectivity, compute and an allocated share of engineering operations. Define usable: a message must be authorised, correctly mapped, timely enough for its purpose and available to the consuming system. This measure will not replace contract review, but it can expose a feed that is expensive because it is over-specified or one that appears cheap because recovery work is excluded.

Run the calculation against alternatives. Compare direct exchange access with a vendor feed on the same instruments and depth.

Include two sites, a recovery stream and the expected user population. Model a 25 to 50 per cent rise in users or message volume, then check whether licensing and infrastructure scale linearly, by tier or at a contract threshold. A low entry price can become costly if every new application creates a separate entitlement.

Also assign a price to operational failure, without pretending it can be forecast precisely. Estimate the duration of a detected gap, affected symbols, orders exposed and time to manual intervention. Use controlled failure tests to establish recovery performance. These figures help set investment priorities, but they are not a substitute for venue rules or an agreed risk limit.

Review the architecture when the business changes: a new asset class, a new client segment, a move into algorithmic execution or a second operating site can alter both rights and delivery requirements. Keep the contract map, feed topology and consumer inventory in one review process. Market data is one of the few infrastructure costs where the packet and the permission to use it arrive on separate invoices.

Frequently Asked Questions

How much do market data feeds cost a broker?

Typical planning budgets range from a few thousand dollars a month for limited coverage to tens of thousands or more for multi-asset services. Exchange fees, user entitlements, redistribution rights, connectivity and support all affect the total.

What is the difference between display and non-display market data?

Display data is shown to a person, such as on a trading screen. Non-display data is consumed by systems for functions such as automated trading, pricing or risk, and exchanges may charge for those uses under separate terms.

Is a vendor feed cheaper than a direct exchange feed?

A vendor can reduce integration and operations work by aggregating and normalising data. Direct access may provide more control and venue-specific detail, but adds connectivity, feed handling and recovery responsibilities. Compare total cost for the same instruments, depth and rights.

How should a trading firm measure feed latency?

Measure timestamps at receipt, book update and consumption by the client or strategy. Track percentiles such as p50 and p99, along with queue depth and recovery time, and document clock synchronisation and which network segments each measurement includes.

What causes market data licensing costs to rise unexpectedly?

New users, applications, venues or redistribution paths can trigger additional fees or reporting obligations. Costs can also increase when a firm changes a system from display use to automated consumption, or keeps duplicate entitlements active after a migration.