The honest answer is "it depends," but the real constraint isn't the spread - it's the fixed costs. Crypto arbitrage has a cost floor that doesn't shrink with your position: a network withdrawal fee is a flat dollar amount whether you move $100 or $100,000, and some venues add a minimum taker fee. On a small turn that flat fee is the whole game - a $2 network fee on a $100 round trip is 2% gone before any spread shows up. The spread itself scales with size, the fixed fee does not, so the same 0.5% gap that loses money at $100 prints clean profit at $10,000. Below we work the math at three tiers, show the second ceiling that bites from the top (book depth and slippage), and find the practical range where fixed fees stop being the whole spread but slippage hasn't started eating it yet.
The honest answer: it's about fixed costs, not the spread
Most "how much to start" answers give you a number and stop. The useful answer is a structure, because two very different costs hang on every arbitrage turn and they behave in opposite ways as your capital grows.
Fixed cost - the network withdrawal fee. To move an asset between exchanges you pay the network fee, a flat amount in dollars. It is roughly the same whether you send $100 or $100,000 of the token. As a percentage of your turn it is brutal when small and negligible when large. Some venues also enforce a minimum taker fee per order, another flat floor.
Variable cost - the taker fees. Each leg charges a taker fee as a percentage (commonly around 0.1% a side, so about 0.2% on a buy-then-sell round). This is a constant share of your turn at any size - it does not get worse as you scale, and it does not get better either.
The whole "how much capital" question is really: at what size does the fixed cost stop dominating? You need enough that the flat network fee is a small slice of the spread, not the whole spread. That threshold, not some magic minimum, is the real floor.
Why $100 barely works: the fixed-fee math
Take a clean 0.5% spread - a perfectly realistic gross gap. On $100 of capital, 0.5% is 50 cents of gross spread. Now subtract the costs that do not care how small you are:
- Taker buy 0.1% of $100 = $0.10
- Taker sell 0.1% of $100 = $0.10
- One network withdrawal to move the asset = $1 to $6 depending on the chain
Even on the cheapest possible network at $1, you have spent $1.20 to capture $0.50. You are net negative 70 cents on a winning spread. On a typical $2-3 network fee you are down more. The spread was real, the trade was correct, and you still lost money - because the fixed network fee alone was 1-6% of a $100 turn, and no 0.5% spread survives that.
This is the single most important thing a beginner gets wrong. They see a 0.5% or even 1% gap, do the percentage in their head, and forget that the flat fee is denominated in dollars, not percent. At $100 the flat fee is the dominant term. To make a 0.5% spread profitable, the network fee has to fall under roughly 0.2-0.3% of your turn - which on a $2 fee means a turn of at least $700-1000.
A small capital does not make small profit on each trade - it often makes negative profit, because the fixed network fee is a larger share of a small turn than the entire spread. The danger isn't slow growth, it's bleeding out one flat fee at a time. Size the turn so the network fee is a fraction of the spread, never a multiple of it.
The three tiers: $100 / $1,000 / $10,000
Hold everything constant - the same 0.5% gross spread, the same 0.1% taker each side, the same $2 network withdrawal to move the asset once - and only change the capital. Watch the fixed-fee drag collapse.
| Capital | Gross spread 0.5% | Taker fees (0.2%) | Network fee (flat) | Net profit | Net % | Fixed-fee drag |
|---|---|---|---|---|---|---|
| $100 | $0.50 | −$0.20 | −$2.00 | −$1.70 | −1.70% | 2.00% of turn |
| $1,000 | $5.00 | −$2.00 | −$2.00 | +$1.00 | +0.10% | 0.20% of turn |
| $10,000 | $50.00 | −$20.00 | −$2.00 | +$28.00 | +0.28% | 0.02% of turn |
The spread did not change. The skill did not change. The only thing that changed is that the $2 flat fee went from 2% of the turn to 0.02% of the turn. At $100 it sinks the trade. At $1,000 it is a tax that barely lets the spread through. At $10,000 it has all but vanished and you keep almost the full 0.3% net.
This is why "how much to start" has no single number but a clear shape: below roughly $700-1000 a single flat network fee can eat an entire ordinary spread, and the more turns you do the faster you bleed. Around $1,000 the same spread starts clearing the costs. By $10,000 fixed fees are a rounding error and your net percentage is close to the gross spread minus taker fees.
Same 0.5% spread, same $2 network fee, only capital changes: $100 turn: +0.50 spread − 0.20 taker − 2.00 network = −1.70 (lost on a real gap) $1,000 turn: +5.00 spread − 2.00 taker − 2.00 network = +1.00 (barely positive) $10,000 turn: +50.0 spread − 20.0 taker − 2.00 network = +28.0 (≈0.28% net kept) Fixed-fee drag = network fee ÷ turn: $100 → 2.00% | $1,000 → 0.20% | $10,000 → 0.02% The spread is constant. The drag is what scales - downward, with size.
The other ceiling: book depth and slippage
If fixed fees only argue for going bigger, why not arbitrage $1,000,000 a turn? Because a second cost works in the opposite direction and gets worse with size: slippage against a finite order book.
A quoted spread - "buy at 0.500, sell at 0.509" - exists only at the top of the book, for the size resting there. Lift more than the top level and you walk up the asks paying progressively more, then walk down the bids receiving progressively less. The effective spread you actually capture shrinks as your size grows past the available depth. What lives at $200 does not live at $20,000: a gap that is clean and fillable in small size can be half-eaten or gone entirely once you try to push real money through a thin book.
So the real picture is a curve, not a line. At the left, fixed fees keep you underwater. The curve climbs steeply as those fees fade. It peaks across a broad middle - the sweet spot - then bends over and falls on the right as your turn outgrows the depth available at the quoted price. The exact location of that bend depends entirely on the asset: a top pair like BTC/USDT swallows large size with little slippage, a thin mid-cap rolls over at a few thousand dollars. Depth, not your bankroll, sets the right edge. That is also why a screener must read depth, not just the headline gap - the topic of the spread screener.
Don't size to your whole bankroll - size to the book. Before committing, ask how much rests near the top of each side at the quoted price. If $3,000 lives there, a $3,000 turn captures close to the quoted spread, a $30,000 turn does not. The right turn size is per-route and per-moment, set by depth, not by how much you happen to hold.
The practical sweet spot and how to scale
Putting the two ceilings together gives a workable range rather than a magic number:
- Under ~$500. Learning territory only. You can and should run real routes here to build the workflow - place orders, move an asset, watch a withdrawal status flip - but expect the network fee to eat most or all of the spread. Treat the cost as tuition, and prefer cheap networks (TRC-20, low-fee L2s) and the rare wider gaps where 0.5% becomes 1.5%.
- ~$1,000 to ~$5,000. The honest entry zone. Here an ordinary 0.4-0.6% spread clears its costs with room to spare, fixed fees are a small tax rather than the whole bill, and almost every liquid route has enough depth that slippage is not yet the binding constraint. Most pairs fill near their quoted spread at this size.
- ~$10,000 to ~$50,000. Comfortable scale. Fixed fees are a rounding error and net percentage sits close to the gross spread minus taker fees. You now start picking routes partly by depth, skipping thin books that would slip, and favoring pairs that can absorb your size.
- $100,000 and up. Depth is the whole game. The constraint is no longer "is the fee small enough" but "can the book absorb this without eating the spread." You split size across routes, lean on the deepest pairs, and accept that the very large gaps you see on thin tokens are mostly unfillable at your size.
To scale up, grow the turn only as fast as two things allow: that the fixed fee stays a small fraction of the spread (it falls away quickly, so this frees up early), and that the book depth on your routes can absorb the larger size without slippage (this is the binding limit as you get big). In practice you raise size per route until you notice your fills drifting from the quoted price - that drift is the book telling you that you have reached its edge for that asset.
You do not need a large bankroll to learn the craft - a few hundred dollars teaches the full workflow. But you do need enough that the network fee is not the entire spread, which in practice means about $1,000 before the math reliably works in your favor on ordinary gaps. Below that you are mostly paying flat fees to practice. Above it the spread starts keeping you, not the fee.
Example: the same route at three sizes
Trace one liquid mid-cap token XYZ, quoted to USDT on both legs, with a clean 0.5% gross gap (ask 0.500 on exchange A, bid 0.5025 on exchange B), taker 0.1% a side, and a single $2 network withdrawal to move the asset. Only the size changes.
Route: buy XYZ on A @ 0.500, move on-chain, sell XYZ on B @ 0.5025 Gross spread 0.5% · taker 0.1%/side · network withdrawal $2 flat — $100 turn — buy: $100 / 0.500 = 200 XYZ, taker −$0.10 move: 200 XYZ on-chain, network −$2.00 sell: 200 × 0.5025 = $100.50, taker −$0.10 result: 100.50 − 100 − 0.10 − 0.10 − 2.00 = −1.70 → real spread, lost money — $1,000 turn — buy: $1,000 / 0.500 = 2,000 XYZ, taker −$1.00 move: 2,000 XYZ on-chain, network −$2.00 sell: 2,000 × 0.5025 = $1,005, taker −$1.005 result: 1,005 − 1,000 − 1.00 − 1.005 − 2.00 = +0.99 → barely clears — $10,000 turn — (assume the book holds ~$10k near top) buy: $10,000 / 0.500 = 20,000 XYZ, taker −$10.00 move: 20,000 XYZ on-chain, network −$2.00 sell: 20,000 × 0.5025 = $10,050, taker −$10.05 result: 10,050 − 10,000 − 10 − 10.05 − 2.00 = +27.95 → ≈0.28% net kept ────────────────────────────────────────────────────────────── Same spread, same fee. The $2 flat network cost is 2% of $100, 0.2% of $1,000, and 0.02% of $10,000. That single line decides whether a correct trade makes or loses money.
The numbers are illustrative, but the structure is exact: the flat network fee is the term that flips the sign at small size, and the same layer-by-layer accounting - spread minus taker minus network, then a depth and D/W check - is what a route reader applies on every gap. The mechanics of that check are in how to read a route, and the costs that vary by network in withdrawal windows and fees.
FAQ - starting capital for crypto arbitrage
What is the minimum to start crypto arbitrage?
There is no hard minimum, but below roughly $1,000 a single flat network fee can eat an ordinary spread whole. You can learn the workflow with a few hundred dollars - just expect most of the spread to go to the network fee until your turn is large enough that the flat fee is a small fraction of it.
Why does $100 lose money on a real 0.5% spread?
Because 0.5% of $100 is only 50 cents, while one network withdrawal costs $1-6 flat. The spread scales with size, the fee does not, so at $100 the flat fee alone is 1-6% of the turn and swamps the half-cent-on-the-dollar gap. The trade is correct and still negative.
How much do I need for arbitrage to be profitable?
For ordinary 0.4-0.6% gaps, around $1,000 per turn is where the math reliably clears costs, and $10,000 is where fixed fees become negligible and you keep close to the full net spread. The exact threshold depends on the network fee and your taker rate.
Can a large capital arbitrage any spread it sees?
No - the opposite ceiling appears. Big size cannot fill at the top-of-book price, so it walks the book and eats its own spread. The wide gaps you see on thin tokens are mostly unfillable at large size. Above roughly $100k, book depth, not the fee, is the binding constraint.
Where do I see spreads already net of fees?
In the spread scanner: the gap is computed net of taker fees and network cost, with honest deposit/withdrawal statuses, so you can judge whether a route clears at your size before committing. Sort by net spread and check the depth, not just the headline percentage.
Not financial advice. The math here shows why small capital struggles and large capital hits depth limits, but it does not make any spread safe. A flat fee can turn a correct trade negative, a thin book can eat a clean gap, and a withdrawal can close while the asset is in transit. The decision to trade, and its consequences, are yours.
Read on: the full picture starts in the pillar Crypto arbitrage guide. First steps and setup - getting started. The errors that cost beginners money - arbitrage mistakes. Why network fees vary and when a withdrawal is closed - withdrawal windows and fees. How a screener reads depth, not just the gap - the spread screener. Live routes net of cost - in the web dashboard.