Algorithmic Trading Infrastructure: What You Need, What You Don't, and What It Costs

The headline number

Almost every retail algo trader builds a P&L model that accounts for spread and stops there. The real cost stack has six layers, and on a small account running a moderately active strategy, the total can exceed 30% of equity per year. That is your actual hurdle rate — the return your strategy must generate before you see a single dollar.

A representative setup: $10,000 account, 0.1 lots per position, 200 round-turn trades per month on EUR/USD.
That's a basic intraday strategy, not a high-frequency scalper.

Your backtested 34% returns just became roughly breakeven.

This is not a trick. It's the arithmetic that separates traders who survive year two from traders who conclude the market is rigged. Every line below is one you can attack — and two of them are lines you should be careful not to cut too far.

The other side of the ledger

The mirror-image mistake

A cost audit that only ever says "spend less" is a bad audit. Under-provisioning has a bill too; it just arrives irregularly, so nobody puts it in the spreadsheet.

What an outage actually costs

The relevant question is not "how often does my machine go down" but "what is open when it does". If your stops and targets sit on the broker's server, an outage is mostly survivable. If your EA manages positions in code — trailing stops, time-based exits, basket closes, re-entries — then every minute offline is a position with no manager.

What latency is actually worth

Here is the number nobody selling hosting will put in front of you. Suppose moving to a colocated VPS improves your average adverse slippage by 0.1 pips. That is a realistic, not heroic, improvement. What is it worth per year?

Annual value of a 0.1 pip slippage improvement on EUR/USD, where 1 pip on 1.0 lot is $10.

Against a $360/year specialised VPS, latency pays for itself at roughly 300 round turns a month at 0.1 lots — or just 60 a month at 0.5 lots.

Read that carefully, because it inverts the usual advice.Lot size drives the latency decision far more than account size does. A $50,000 account trading 0.1 lots twice a week gets almost nothing from colocation. A $5,000 account trading 0.5 lots two hundred times a month gets $1,200 of value from it. Nobody sizes their infrastructure this way, and they should.

Below those thresholds you are still buying a VPS — you are just buying uptime instead of speed, and you should buy it at the cheap tier accordingly.

Line item 01

Spread and commission

The largest single cost, and the one most people model incorrectly.

Standard vs Raw: the crossover math

Standard accounts bundle the broker's markup into the spread — no separate commission, typically 0.8–1.0 pips on EUR/USD. Raw or ECN accountspass through near-interbank spreads and charge a separate commission, typically 0.0–0.3 pips plus $4.50–7.00 per lot round turn.

Convert both to one number. On EUR/USD, 1 pip on 1.0 lot is $10, so on 0.1 lot it's $1.

30 cents. Trivial per trade, and $720 a year at 2,400 trades — 7.2% of a $10,000 account, for a decision you make once during signup.

Where your backtest lies to you

The MT4/MT5 strategy tester applies a fixed spread across the entire history by default. Live spreads are not fixed. They widen at rollover (5pm New York, often 3–10× for several minutes), during high-impact news, at the Sunday open, and permanently on exotics and indices.

Test this: run your backtest at your broker's average spread, then re-run at double it. If the strategy dies, it was never viable — it was living inside a modelling assumption.

Nine things to check that are not price

Price is the easy variable because it is published. These are the ones that actually break EAs, and almost none of them appear in a comparison table.

1. 01Stop level / freeze level. Some brokers refuse stops or targets placed within N points of market. If your EA uses a 5-pip stop and the broker enforces a 3-pip stop level, you will get error 130 and no trade. Ask for the number in points, per symbol.
2. 02Symbol suffixes. EURUSD.a, EURUSD.raw, EURUSDm — an EA hardcoded to "EURUSD" fails silently or throws on every tick. Check the exact symbol strings before you migrate a working EA.
3. 03Digits and point size. A 5-digit broker prices EUR/USD to 0.00001. An EA written in points for a 4-digit feed will set stops ten times too tight. Most modern EAs handle this; verify rather than assume.
4. 04Minimum lot and lot step. 0.01 minimum is standard, but not universal, and a strategy sized for 0.01 on a broker with a 0.1 minimum is a different strategy.
5. 05Scalping and hedging permissions. Some entities impose minimum hold times, restrict scalping, or forbid holding both directions. Read the account terms, not the marketing page.
6. 06Which legal entity you get. The same brand can onboard you to a Tier-1 regulated entity or an offshore one depending on your residence. That determines leverage, negative balance protection and whether any compensation scheme applies to you. It is often the single largest difference between two accounts at the same broker.
7. 07Number of servers and which one you are on. Brokers run multiple servers in different data centres. Your VPS decision depends on which one your account sits on, not on where the company is headquartered.
8. 08Execution model per account tier. A broker can be A-book on one tier and B-book on another. The answer that matters is the one for the tier you are actually opening.
9. 09Weekend and holiday trading hours. Session close and open times vary by broker and shift with daylight saving. An EA with time-based logic will drift by an hour twice a year if you have not handled it.

Which one is likely yours

These are starting points from published pricing, not endorsements. The right answer depends on your pairs, your entity and your measured fills.

Line item 02

Slippage and execution quality

The strategy tester fills you at exactly the price you requested. Reality does not.

What actually determines your fill

Execution model.Between your EA sending an order and the broker receiving it, the price often moves. What happens next depends on the account.

Market execution fills you at whatever the price is now. You always get the trade, sometimes at a slightly worse price — that's slippage.

Instant execution fills you only at the price you asked for. If the price moved, the broker sends back a requote asking whether you accept the new one, and your order doesn't execute in the meantime.

For an EA, that difference is bigger than it looks. Slippage is a cost you can measure in your trade history. A requote is a trade that never happened — your signal fired, your backtest counted it, and live it simply isn't there, with nothing in your results to flag it.
Worse, requotes cluster when price is moving fast: news, session opens, breakouts. So instant execution doesn't remove trades at random. It removes the volatile ones, often the exact trades your edge depends on. The result is a live curve that diverges from the backtest for reasons no amount of re-testing will surface.

A-book vs B-book. An A-book broker routes to liquidity providers and profits from commission. A B-book broker takes the other side internally. B-book isn't inherently predatory — it's how most retail volume is handled, and it often produces tighter pricing on small sizes. But your interests and the broker's are directly opposed, and B-book desks have discretion over fills. Ask any broker which model applies to your tier. One that won't answer clearly has told you something.

Latency. Every millisecond between decision and broker server is price movement you don't control. On a mean-reversion scalper that's the difference between edge and no edge. On a daily-timeframe swing strategy it's irrelevant. Know which strategy you have before spending money on latency.

Measuring your own

Don't estimate it. Export your trade history, compare requested price to fill price on every order, average the difference in pips. Most setups run 0.1–0.5 pips of average adverse slippage on majors. Above 0.5 and you have a latency or broker problem worth fixing.

Two refinements worth the extra ten minutes. First, split the sample by time of day — if your slippage is fine at 10am London and terrible at 8:30am New York, you have a news-handling problem, not a hosting problem, and no VPS will fix it. Second, look at the distribution, not just the mean. Slippage that averages 0.2 pips because it is usually 0.05 and occasionally 4.0 is a different diagnosis from slippage that is 0.2 every single time.

The asymmetry test

On a fair feed, slippage is noise. Price moves in both directions between your order leaving and your fill landing, so roughly half your orders should come back slightly better than requested and half slightly worse. Positive slippage is not a gift — it is the other half of a distribution that ought to exist.

Some brokers pass on only one half of it. The mechanisms vary: a bridge configured to apply negative slippage in full while capping or discarding positive, a dealing desk with discretion over which fills get improved, a last-look window that rejects the orders that moved in your favour and fills the ones that moved against you. The effect is identical in your trade history — a small, consistent tax that looks like bad luck and never once reverses.

Count, don't average. A mean of −0.2 pips tells you very little; "zero positive fills in four hundred orders" tells you almost everything, because on a symmetric feed that outcome is somewhere past astronomically unlikely.

So control for it. Split the sample into two buckets: limit-style entries during quiet sessions, and momentum or stop entries. The first bucket is your clean test. If even those fills never once come out in your favour, you have your answer.

What to do about it. Nothing technical. No VPS, no colocation and no EA parameter fixes a counterparty that only passes on one tail of the distribution. Ask the broker in writing whether positive slippage is passed through in full and what their execution policy says about price improvement. Firms that apply it symmetrically answer that plainly, and regulators in several jurisdictions have taken enforcement action over asymmetric slippage, so it is a question compliance teams expect. If the answer is vague or arrives as a marketing paragraph, that is the answer. Move your account.

Fixing it, cheapest first

Note where the last row sits: at the bottom, because it is the most disruptive, and in the list at all because it is sometimes the only fix. Everything above it addresses slippage you are causing. Asymmetric slippage is slippage being done to you, and no amount of infrastructure spend touches it.

Line item 03

Swap and rollover

The quiet account killer, because it accrues while you're not looking.

Hold past 5pm New York and you pay or receive interest on the rate differential between the two currencies. Wednesday carries triple swap to cover the weekend.

Who this destroys

Grid and martingale strategies. These hold for days or weeks, often many positions at once, frequently on the wrong side of the carry. Take eight positions at 0.1 lots held three weeks. Swap bills seven times a week, not five — Wednesday charges triple to cover the weekend — so three weeks is 21 charges, not 15.
At an example negative carry of $6.25 per lot per night, that's 0.8 lots × $6.25 × 21 = $105. On a basket targeting $150 gross, financing has taken 70% of the profit before you have counted a single pip of spread.

Any strategy on a negative-carry pair. Check the swap table for your instrument in both directions before deploying. Swap rates vary enormously between brokers on specific pairs, and almost nobody shops on it — which is exactly why it is one of the few remaining places where shopping still pays.

How to actually shop swap

1. 01Pull the number in account currency, not points. Broker swap tables publish in points, pips, currency units or percent depending on the broker. Convert everything to "dollars per lot per night" before comparing anything.
2. 02Get both directions. A pair that costs you $6 long may pay you $3 short. If your strategy is directional, only one of those numbers is yours.
3. 03Check which day is triple. Wednesday for most FX. Friday for many indices, metals and crypto CFDs. Getting this wrong understates a weekly-hold strategy by 40%.
4. 04Multiply by your real holding period. Not the average from your backtest — the 90th percentile. Swap costs are worst precisely on the trades that go wrong and stay open.
5. 05Re-check after policy changes. Swap follows the rate differential. A central bank move can turn a carry-positive pair carry-negative within a quarter, and your EA will not notice.

The backtest problem

MT4/MT5 applies the current swap rate across your entire historical test. Backtest 2015–2025 and you're applying today's rate to a decade in which rate differentials moved enormously. For any strategy holding over 24 hours, treat backtested swap as unreliable and stress-test it manually — re-run at 2× and 3× the current rate and see what survives.

Line item 04

VPS and hosting

Your trading bot needs uptime during market hours. A home setup fails on power cuts, ISP outages, Windows updates rebooting at 3am, and laptop lids closing — each leaving a position open with no manager, no trailing stop, no exit logic running. The question isn't whether to use a VPS. It's which tier, and that is a question with a real answer rather than a matter of taste.

The three tiers

Which tier is yours

Work down this list and stop at the first row that describes you. This is the section the rest of the internet skips.

$30/month is $360/year. On a $3,000 account that's a 12% drag before you place a trade.

If your account is under roughly $3,000, a specialised VPS is hard to justify unless your broker gives you one. Paying for 2ms latency on a $1,500 account running a swing strategy is spending real money to solve a problem you don't have.

Sizing it

MT4 ≈ 200–300MB RAM per instance, MT5 ≈ 400–600MB, plus ~2GB for Windows. Match the VPS to your broker's server location, not your own — you're optimising the EA-to-broker hop.

Two things eat resources more than people expect: open charts with many indicators, and MT5's built-in strategy tester if you leave it running. Neither shows up in the per-terminal estimate above.

The four providers, compared

Four things the marketing pages won't tell you

1. 01

   Every latency number in this category is self-measured.

   Providers publish their own benchmarks, under undisclosed conditions, on environments nobody can replicate. There's no independent auditor for retail forex VPS the way STAC Research audits institutional infrastructure. Treat published millisecond figures as directional marketing, and use the money-back window to measure it yourself against your actual broker — that's what it's for.

2. 02

   Jitter beats raw ping.

   A stable 6ms connection outperforms one swinging between 2ms and 25ms, because fill quality depends on predictability more than peak speed. When you test during the refund window, run the ping continuously through a news event, not once on a quiet Tuesday.

3. 03

   The uptime SLA covers the host, not your terminal.

   A 99.9% guarantee means the hypervisor was reachable. It does not mean MetaTrader was still running after a Windows update, that your EA reattached to its chart, or that your account stayed logged in. The compensation for a breach is normally a service credit worth a few dollars, which will not cover one unmanaged position. Monitor the terminal itself — an EA that pings a heartbeat endpoint is twenty lines of MQL and worth more than the SLA.

4. 04

   The money-back window is the real product spec.

   Latency claims are unverifiable, so the number that carries actual information is how long you get to test before you are committed. A provider offering a genuine no-questions refund period is making a costlier promise than one publishing a 0.7ms figure. Use the window fully rather than treating it as a formality.

How to test a VPS in the refund window

Seven days of deliberate testing separates a provider that suits your broker from one that suits their own marketing page. Do this before the window closes, not after.

1. 01Day 1 — Establish the baseline. Install MT4/MT5, log into your live account, and record the ping figure in the bottom-right status bar. Then run a continuous ping to the broker server address and log it to a file. You want a distribution, not a screenshot.
2. 02Day 2 — Compare against home. Run the identical ping from your home machine at the same times. If the difference is under 20ms and your strategy is not latency-sensitive, you have your answer already and can request the refund.
3. 03Day 3–4 — Run the EA in parallel. Same EA, same settings, on a demo account on the VPS and on a demo at home. Identical signals should produce near-identical fills. Divergence is the thing you are shopping for.
4. 04Day 5 — Test through a release. Pick a high-impact event on the calendar. Watch ping and jitter across the release window. This is the only test that matters for news-sensitive strategies, and it is the one nobody runs.
5. 05Day 6 — Break it on purpose. Reboot the VPS. Time how long until the terminal is back up, logged in, and the EA is running with the smiley face. If that is not automatic, you have found a real operational cost and should fix it now rather than at 3am.
6. 06Day 7 — Check the resource ceiling. Open every chart and terminal you plan to run, plus one more. Watch RAM and CPU under that load for a full session. Providers size their entry tier for one quiet terminal.

If a provider's refund window is too short to run this, that is itself a data point.

Line item 05

Data and tooling

Tick data. Default MT4/MT5 history is interpolated from M1 bars and produces modelling quality around 90%. Real tick data gets you to 99.9%. On strategies with intrabar entries, stop-hunting logic, or tight stops, that difference routinely turns a profitable backtest into a losing one. If you've never run your strategy on real tick data, you don't yet know whether it works.

This is the one line item where the cheap option and the correct option are usually the same. Free historical tick data exists for major pairs, and MT5's own tester will accept real ticks from your broker if the broker stores them. Paid tick data is worth it when you need variable spread modelling, long histories on crosses, or a feed independent of the broker you are about to evaluate. Buying it once and re-using it across every strategy you ever test is a better shape than a subscription.

The EA itself. MQL5 Market EAs run from about $30 to $2,000+, often with a rent-or-buy choice. Rent is the right call while you're evaluating, but check the crossover: $30/month against a $500 purchase pays for itself in 17 months, and subscriptions tend to outlive the attention anyone gives them.
On price as a quality signal — there isn't much of one. An EA costs nothing to copy, the vendor is frequently pseudonymous, and the only evidence you see before paying is a backtest the vendor produced themselves. Price is set by positioning, not by anything you can verify. Expensive tells you the vendor wanted to look serious; it doesn't tell you the strategy works.
That cuts both ways once you've bought. Money already spent is gone whether you keep running the EA or not, and it's a poor reason to keep a losing system on a live account. Judge it on what the trades have actually done since you deployed it.

Line item 06

Everything uncounted

Currency conversion

Depositing SGD, EUR or GBP into a USD account typically costs 0.5–1.5% each direction, and you pay again on withdrawal. Open in your home currency where supported, or use a multi-currency transfer service rather than your bank’s retail rate.

Withdrawal fees

Some brokers are free; others charge $20–30 per wire. Withdraw monthly and that’s $360/year. Batch withdrawals quarterly and you have saved most of a VPS.

Inactivity fees

Common after 3–12 months dormant, typically $10–15/month. Relevant if you rotate strategies and leave accounts idle. Close accounts you have stopped using rather than leaving a float in them.

Prop firm challenges

Model the fee as a recurring cost with an expected number of attempts, not a one-off. A $500 challenge with a realistic three attempts before passing is a $1,500 cost of capital.

Track record subscriptions

A monthly platform fee is a fixed cost like a VPS, and it bills every month regardless of whether you qualified for an allocation that month. Model it in qualifying months, not against your equity — there is a section on this below.

Your time

Not cash, but reconciliation, tax records and monitoring are real hours. Price them at whatever your hour is worth and the number usually exceeds your VPS bill.

Tax and record-keeping

Depends entirely on your jurisdiction, but the cost of reconstructing two years of trades under pressure is far higher than exporting statements monthly from the start. Automate the export the day you go live.

The part that decides your spend

Which of these costs are actually yours

Six cost layers, but no strategy pays all six equally. The dominant one for a swing EA is swap; for a scalper it is spread; for a news strategy it is slippage. Spending evenly across all six is how people end up with a colocated VPS and a swap bill they never looked at.

Find yourself below. The "skip" line is as important as the "spend" lines.

Order of spend

The provisioning ladder

Infrastructure should be bought in a sequence, and the sequence is driven by equity, because fixed costs are a percentage and percentages are what kill small accounts. Do not buy a rung before you are standing on the one below it.

Measurements, not vibes

The upgrade triggers

Every one of these is something you observe in your own account, not something you read on a spec sheet. If none of them has fired, you do not need to spend anything.

Note what is absent from that list: account balance, on its own, for anything except the last row. Balance determines what you can afford; measurements determine what you need.

Where the math inverts

The prop firm inversion

Everything on this page has assumed your infrastructure cost is measured against your own equity. On a funded account it is measured against someone else's, and every conclusion flips.

What changes on a funded account

• →You do not choose the broker. The firm does. Your entire broker-selection lever is gone, which means the spend you can still control — hosting, data, monitoring — carries more weight, not less.
• →Rule breaches are binary. Exceed a daily drawdown limit by one dollar and the account is gone. An unmanaged position during an outage is no longer a bad trade; it is the end of the account and the challenge fee with it.
• →The challenge fee is the real cost of capital. A $500 challenge at three expected attempts is $1,500. Against that, refusing to spend $30/month on hosting that protects the attempt is not frugality.
• →Consistency rules punish gaps. Many firms require trading activity across a minimum number of days. Downtime does not just cost you a trade — it can cost you eligibility.
• →Scaling changes the slippage math. Funded lot sizes are multiples of personal ones, and the slippage table above scales linearly with lots. At 1.0 lots, 0.1 pips of improvement is worth $10 per hundred trades.

On a funded account, infrastructure stops being a cost and becomes insurance on a challenge fee you have already paid.

The other route to capital

Renting a track record instead of buying a challenge

The section above assumed the challenge model: pay a fee, hit a profit target without breaching a drawdown limit, get funded, keep most of the profits. There is a second model that inverts almost every part of that, and it suits systematic EA traders considerably better than it suits discretionary ones.

Instead of a one-off fee and a pass/fail test, you pay a monthly subscription and trade a virtual account with no profit target and no time limit. Your trade history is converted into a verifiable, third-party-certified record. Investors then allocate real capital against that record, and you take a cut of the profits it generates. Nothing you do can fail an evaluation, because there isn't one.

The two models, side by side

Look at the "your share" row before anything else. 15% against 80% looks indefensible until you notice the two numbers are percentages of very different things. A prop firm's scaling plan tops out at a few hundred thousand; allocation here starts as seed capital and opens onto uncapped third-party investor money, and it does not evaporate the first time you have a bad Tuesday.

The row that surprises people, though, is "how you get funded". This is not a beauty contest where you post a record and hope somebody notices. Seed allocation is rules-based: hit a defined score and the capital is committed to you. That makes it modellable in a way discretionary allocation never is, which is the whole reason it belongs on a page about arithmetic.

Darwinex Zero, specifically

The main operator of this model is Darwinex Zero(opens in a new tab), run by Tradeslide Technologies, the company behind the Darwinex brokerage. You subscribe, trade a virtual account on MT4 or MT5, and after a calibration period your strategy becomes a DARWIN — a tradeable index tracking your positions, which investors can buy. Allocation comes through their DarwinIA programme and from third-party investors, and you receive a performance fee on the profits your DARWIN generates for them.

Published subscription pricing, which varies by region and asset class and has moved more than once. A reset costs one month. The 3-year pack includes a booster credit. Verify current pricing before subscribing.

How the allocation actually works

This is the part that changes the arithmetic, and it is worth being precise about. Seed allocation runs through DarwinIA, which has two levels.

The rating runs 0–100 and is dominated by trailing performance: cumulative return over the previous five months plus the current one carries the most weight, current-month return next, and maximum drawdown across that same six-month window the rest. Track record length adds a small bonus that grows past 6, 12 and 18 months. Clear 75 at month end and the minimum allocation is committed — you are not competing with anyone for it. Rank higher and the allocation scales up substantially.

Two eligibility details that quietly disqualify people: the account needs to hold a minimum equity threshold throughout the month, and your DARWIN must not correlate too closely with another user's, or with your own other DARWINs. If you run three variants of the same EA, expect two of them to be excluded.

Three streams, not one

This is the distinction that decides whether the subscription is worth paying, and it is the one every summary of this product blurs. Money reaches you through three separate channels, and they behave completely differently.

1. 01Seed capital, SILVER. Rules-based and recurring. Clear the rating each month, get an allocation, hold it three months. Miss a month and that month simply produces nothing — the stream pauses rather than closing.
2. 02Seed capital, GOLD. Larger, held twice as long, and competitive rather than guaranteed. You drop back to SILVER if you stop meeting the criteria, and can climb back later.
3. 03Third-party investor capital. The one that actually matters. Reaching GOLD once publishes your DARWIN to Darwinex-Classic investors — and it stays published even if you subsequently fall out of GOLD. Investors allocate on their own schedule, there is no expiry, no monthly re-qualification, and no ceiling. You earn the same 15%, under the same high-water mark.

Reaching GOLD is a one-time unlock, not a subscription within a subscription. That is the real objective; the seed allocations are what pays the bills on the way there.

Where this belongs on a cost page

Right here, because a subscription is a fixed cost and this entire page is about what fixed costs do to small accounts. $50 a month is $600 a year — larger than the VPS line, larger than the data line, and it accrues every month whether you clear the rating threshold or not.

But because the seed trigger is a rule rather than someone's judgement, the floor is an actual calculation rather than a hope. Work it in qualifying months, and treat investor capital as upside on top rather than as part of the break-even.

What can go wrong, precisely. The performance fee applies to return achieved under a high-water mark, so a losing quarter earns nothing and you have to recover the ground before earning again. And the SILVER rating leans on your trailing six-month return, so a flat run does not merely fail to earn — it drops you below 75 and pauses the seed allocations while the subscription keeps billing. The failure mode is not blowing up. It is mediocrity, sustained, at $50 a month.

What that does not do is close the door. There is no bust-out event here: a bad six months costs you the seed stream until the trailing numbers recover, and if you had already reached GOLD, your DARWIN stays in front of investors throughout. That asymmetry — downside is a pause, upside is a permanent unlock — is the actual argument for the model, and it is a very different shape from a challenge fee you can lose in an afternoon.

Against the challenge model: a $500 challenge at three realistic attempts is $1,500 spent up front. At $50 a month you reach $1,500 after thirty months — so for the first two and a half years the subscription is cheaper in absolute terms and cannot be lost to a rule breach. The difference is that a challenge pays out in a lump once you pass, while this pays in small, rules-based instalments for as long as you keep clearing the bar.

Honest fit

The infrastructure footnote

A virtual account is still a live account for every purpose this page cares about. The orders meet real market conditions, so slippage, spread and execution quality all still shape the record — and the record is the entire product. Two consequences worth spelling out.

You still need hosting, and the case is stronger than usual. An outage does not just cost you a trade; it puts a gap in the record you are paying monthly to build. Match the VPS to their server location, not to any broker you also trade with.

Your broker costs do not disappear if you run the same EA live in parallel, which most people do. Budget both, and expect the two accounts to diverge on fills — that divergence is itself information about your execution, and worth measuring rather than ignoring.

The 20-minute infrastructure audit

Work through this in order. Every step is free.

1. 01Count your trades per month and your average lot size. Both. Volume decides the broker question; lot size decides the latency question.
2. 02Calculate all-in cost per round turn — spread + commission + measured slippage. Multiply by annual trade count, divide by equity. That’s your hurdle rate.
3. 03Compare against your backtested return. If the hurdle exceeds half your expected return, the strategy isn’t viable at your account size regardless of infrastructure.
4. 04Check your average holding period. Over 24 hours, pull the swap table for your pair in your direction and model it explicitly.
5. 05Find your broker’s server location. Status bar, bottom right. Then check whether your VPS is in the same data centre.
6. 06Measure actual slippage from trade history. Above 0.5 pips on majors means a fixable problem. Check whether it is symmetric while you are there.
7. 07Ask whether your EA manages positions in code. If it does, uptime is a risk control and you need hosting regardless of everything else on this list.
8. 08Run the latency break-even. 0.1 pips × your lot size × your annual trades. If that is under $360, you are buying uptime, not speed — buy the cheap tier.
9. 09Re-run your backtest at 2× spread. If it dies, it was a modelling artifact.
10. 10Stress-test the equity curve before spending anything on steps 1–9. Resample the trade sequence and look at the distribution of outcomes, not the single path history happened to hand you.

Step 10 is last on the list and first in importance. Costs only matter if there's something worth protecting.

Loose ends

Common questions