At the end of a bad week, an operations lead looks at a dashboard showing delayed quotes, dropped packets and a vendor invoice that has climbed again. The feed itself may be healthy. The cost is scattered across exchange licences, connectivity, compute, support time and the occasional missed update.
A tick is not a standard unit of value. It might mean a trade, a best-bid-or-offer change, or an order-book message. The right comparison starts by deciding which updates the strategy or client actually needs, then measuring the full cost of getting them there.
Measure the useful update, not the headline price
Feed pricing usually combines several charges. Exchanges may price by venue, data product, subscriber, device, professional status and use.
A broker distributing data to clients can face different obligations from a prop firm consuming it internally. Non-display use, derived data and redistribution can each change the licence position. Confirm the definitions with the venue or authorised distributor before modelling costs.
For infrastructure planning, separate the invoice into four parts: rights to consume or redistribute data, the feed service, network and hosting, and the labour required to operate it. A low monthly feed fee can be poor value if it excludes depth, forces a costly normalisation layer, or leaves your team responsible for recovery and replay.
Useful measures include cost per active instrument, cost per licensed user, cost per million usable updates, and cost per unit of latency. The last two need care. A conflated feed may send fewer messages but discard intermediate book changes. A raw feed can carry more updates than the strategy can process. Count the updates your application consumes, and record whether the feed preserves sequence numbers, depth and timestamps.
Pay for the information your system can use, not the message count on the invoice.
For example, a sustained 20,000 updates a second over a 6.5-hour session is about 468 million updates a day, before adding bursts or other venues. That arithmetic helps estimate bandwidth and processing, but it is not a forecast of a particular market. Use captured peak traffic for capacity planning, and use the licensed product definition for billing.
Four ways to receive a tick
The main options differ in where the feed is decoded, who manages the connection and how much control remains with your team. The figures below are typical planning ranges for a modest deployment, not quotes. Monthly technical costs exclude exchange data licences and internal staff; actual terms vary by venue, location, redundancy and commercial agreement.
| Approach | Typical recurring technical cost | Typical delivery latency | Operational trade-off |
|---|---|---|---|
| Direct venue feed in colocation | £2,000-£15,000 per venue per month | 50-500 microseconds to application, in a well-tuned setup | Most control; your team owns handlers, recovery and venue-specific changes |
| Managed vendor feed over cross-connect | £500-£5,000 per service per month | 0.5-5 milliseconds | Less venue integration work; product scope and normalisation are vendor-defined |
| Vendor feed delivered through cloud | £300-£4,000 per month for hosting and delivery | 5-50 milliseconds, depending on region and path | Simple deployment and scaling; variable egress and jitter need monitoring |
| Broker or vendor API | £0-£1,500 per month for access in common entry tiers | 20-250 milliseconds, with wider variation under load | Fastest to integrate; limits, conflation and availability may constrain trading use |
These latency ranges describe typical paths from publication or provider delivery to an application, not a guaranteed service level. Measurement points matter.
A vendor's timestamp may mark receipt at its gateway, while your strategy records after decoding and queueing. Compare like with like, and report median, 99th percentile and worst observed latency separately.
Direct feeds are the usual choice where order-book detail, sequence control or low latency justifies the engineering and connectivity spend. They arrive in venue-specific formats, often over multicast, and may require redundant A and B lines, packet-gap detection, retransmission or snapshot recovery. A handler that silently accepts a gap can publish a plausible but wrong book.
A managed feed can consolidate venues, normalise symbols and handle some recovery. That reduces integration work, but check whether the service preserves exchange timestamps, order-level data and venue sequence information. Cloud delivery and APIs suit research, dashboards, monitoring and less latency-sensitive workflows. They are not automatically suitable for execution: quotas, throttling, region placement and burst behaviour can dominate the average latency.
Count the costs that do not sit on the feed line
Licence charges can exceed delivery costs, particularly when data is redistributed or exposed to many users. Keep a licence inventory by venue, product, consumer type and purpose.
Track who can view data, whether a client device counts as a subscriber, and how derived values are treated. A change from internal use to client display should trigger a review, not an assumption that the existing entitlement covers it.
Infrastructure costs also depend on traffic shape. Multicast can reduce duplicate network traffic for many consumers, but requires disciplined group management and packet-loss handling. Unicast simplifies some paths while increasing bandwidth and connection counts.
Depth feeds can produce sharp bursts around news and opens. Size network buffers and queues against captured bursts, not average messages per second.
- Network: cross-connects, ports, redundant circuits, internet transit and cloud egress.
- Compute: feed handlers, normalisation, book building, fan-out and replay storage.
- Resilience: secondary paths, snapshot recovery, clock synchronisation and disaster-recovery tests.
- Operations: venue changes, entitlement audits, incident cover and reconciliation of sequence gaps.
Staff time is often the hidden differentiator. A direct feed may need a specialist to maintain protocol handlers and test exchange changes.
A vendor service shifts some of that burden, but introduces a dependency on its release schedule, support process and outage reporting. For a broader view of platform dependencies and operating models, see Trading Technology's infrastructure coverage.
Choose by workload and failure mode
Start with the consumer, not the transport. A market-maker maintaining an order book has different requirements from a broker displaying indicative prices in a client app. Write down the required fields, depth, venues, update policy, latency budget and recovery behaviour. Then test whether each candidate meets those requirements during both normal traffic and a burst.
- Define the minimum data: list instruments, venues, depth, fields and whether every intermediate update is required.
- Set measurable service targets: specify end-to-end latency percentiles, maximum recovery time and acceptable packet-gap duration.
- Price the full path: include licences, connectivity, compute, support, user counts, egress and engineering effort.
- Replay captured traffic: test peak bursts, sequence gaps, reconnects and failover with the actual downstream consumers.
- Compare operating evidence: review timestamp semantics, incident notices, change windows, support response and entitlement reporting.
Use the same capture and measurement points for every option. A useful trial records exchange or provider timestamps, NIC arrival, decode completion and strategy consumption.
That separates network delay from queueing and handler time. It also exposes cases where a feed is fast on average but falls behind during a burst.
Many firms use more than one approach. A direct feed can serve latency-sensitive execution, while a normalised vendor feed supports risk, surveillance and client display.
That architecture can duplicate licensing and storage, so define which system is authoritative for each use. Document what happens when the primary feed recovers with a different sequence or timestamp history from the secondary.
Trading Technology has covered the platform and infrastructure choices that shape these operating decisions. The practical question is whether the extra control of a direct feed pays for itself in your workload, rather than whether it is technically faster in isolation.
Make the next decision reversible
Do not migrate a production feed on a latency chart alone. Run the candidate in parallel, compare books and timestamps, and reconcile gaps against a trusted source. Set a cutover gate that includes market-open behaviour, recovery tests, entitlement sign-off and an agreed rollback path.
Review the cost model quarterly and whenever you add venues, users, products or a new redistribution path. Track actual peak message rates, egress, handler CPU and incident hours beside licence counts. Those measures show whether a cheaper service remains cheaper after the workload changes.
If this were my operation, I would first map every feed to its consumers, licences and measured peak traffic. Then I would price the full path and run a parallel capture against the existing service before changing production.
Frequently Asked Questions
What affects the cost of a market data feed?
Costs include exchange licences, vendor service charges, connectivity, hosting and engineering support. Venue, product depth, user count, redistribution and non-display use can all change the total.
Are direct market data feeds always faster than vendor feeds?
A direct feed in colocation can reduce delivery latency, but the result depends on the venue, network path, handler and application queues. Compare measured end-to-end percentiles rather than headline figures.
When is a cloud market data feed suitable?
Cloud delivery often suits research, monitoring and client displays where millisecond-level latency is acceptable. Check regional placement, egress charges, burst capacity, jitter and the provider's recovery behaviour before using it for execution.
How should a firm estimate cost per tick?
Define a usable update for the workload, then divide the full cost of the feed path by the number of usable updates over the same period. Include licences, infrastructure and operations, and keep separate measures for users and instruments.
Platforms · Software · Providers · Markets