I tried cross-exchange arbitrage in 2021. BTC was $200 cheaper on Kraken than Binance. I bought on Kraken, transferred, sold on Binance. By the time BTC arrived 25 minutes later, price had moved $350 against me. Lost $150. That's why retail arbitrage is dead.
Crypto arbitrage involves buying a digital asset on one exchange where the price is lower and simultaneously selling it on another where the price is higher—capturing the price difference as profit. But isn't arbitrage gone in 2025? Not at all. What has changed is how arbitrage is done. Manual arbitrage is nearly impossible now.
Here's why, and what infrastructure actually works.
1. Speed begins with information
Most retail systems continuously request price updates from exchanges. Through Apache Kafka event streaming architecture, professional infrastructure remains continuously connected to market data feeds.
You poll API every second. Pros get websocket tick-by-tick in <1ms. By the time you see the arb, it's gone.
2. Settlement latency kills you
We show that settlement latency—the delay until a transaction is validated on the blockchain—poses significant costs to cross-exchange arbitrageurs, who exploit price differences across exchanges. This latency exposes arbitrageurs to price risk, making arbitrage opportunities viable only if the price differences are substantial enough to offset this risk.
Your transfer takes 10-30 minutes (BTC) or 2-5 minutes (ETH). Price moves 0.3-1% in that time. Arb was 0.4%. You lose.
3. Capital fragmentation
You need capital on both exchanges. $10k on Binance, $10k on Kraken. That's $20k capital for $10k arb. Capital efficiency 50%.
Pros use cross-venue margin pooling. What CrossEX represents is infrastructure that matters for how professional capital deploys into crypto markets. Execution routing was the first generation, ensuring an order finds the best fill. Cross-venue margin pooling is the second, ensuring the capital behind that order is efficiently positioned across the venues a desk actually trades on.
Empirically, we analyze one year of transactions (September 2023 - August 2024) across nine blockchains and identify 242,535 executed arbitrages totaling 868.64 million USD volume. Activity clusters on Ethereum-centric L1-L2 pairs, grows 5.5x over the study period, and surges—higher volume, more trades, lower fees—after the Dencun upgrade (March 13, 2024).
242,535 arbitrages, $868M volume in one year. That's $3,580 average per arb. But retail captured maybe 2% of that.
Real data: 41% of market exhibits arbitrage. Research team analyzed data from April 2024 to April 2025: checked 17,218 conditions, 7,051 conditions exhibited single-market arbitrage (41%). Median price deviation: $0.60 (should be $1.00). 13 confirmed cross-market exploitable arbitrages. A median deviation of $0.60 means that the market regularly deviates by 40%.
Opportunities exist, but you can't capture them manually.
We investigate how exchange default risk and liquidity affect Bitcoin cross-exchange arbitrage opportunities. Analysing minute-level data from 16 cryptocurrency exchanges (April 2013–April 2024), we find arbitrage opportunities last longer when higher-risk exchanges have higher prices, as traders are cautious of default risks. There is a strong positive relation between capital flows from high-risk to low-risk exchanges and arbitrage opportunities, showing a preference for safer exchanges.
Translation: arb persists because traders won't move funds to risky exchange to capture it. You need to already have capital there.
What pros have that you don't:
1. Co-location
Servers in same data center as exchange matching engines
Latency: <0.5ms
Your latency: 50-200ms
They see price, trade, confirm before you receive websocket
2. Pre-funded accounts
Capital on 10+ exchanges
No transfer latency
Can arb instantly
3. Cross-exchange margin
Gate CrossEx: margin shared across venues
Don't need $10k on each exchange
$10k total margin covers $10k positions on multiple venues
4. Smart order routing
AI finds spreads in fractions of a second
Inter-Exchange Arbitrage: How AI Finds Market Spreads in Fractions of a Second
AWS Cloud Infrastructure, Apache Kafka streaming
Not polling, continuous connection
5. Inventory vs bridge
Hold inventory on both sides (no transfer)
Or use fast bridges (LayerZero, etc.)
Academic models show profitability of inventory against bridge arbitrages
Your process:
See BTC $67,000 on Binance, $67,200 on Coinbase (0.3% spread)
Buy on Binance (takes 2 seconds)
Withdraw BTC (takes 10 minutes)
BTC arrives Coinbase
Price now $66,900 on both
Sell at loss
Time elapsed: 12 minutes
Price risk: ±0.5% typical
Required spread to profit: >0.7% after fees
Actual spreads available: 0.1-0.3%
Math doesn't work.
Don't do cross-exchange. Do:
Triangular arbitrage on single DEX (same block)
Or use bots with pre-funded capital
Or accept you're providing liquidity to pros
An arbitrage opportunity arises when the exchange rate of two cryptocurrencies in a DEX differs from that in a CEX. The exchange rate of each trading pool is independent, which introduces arbitrage opportunities with trading through different cryptocurrencies: traders are able to trade cryptocurrencies cyclically.
But even cyclic arbitrage requires MEV bots, not manual.
1. Cross-chain arbitrage (with inventory)
Hold USDC on Arbitrum and Base
When price deviates, swap on both chains simultaneously
No bridge latency
Requires $50k+ capital per chain
Returns: 5-15% APY
2. CEX-DEX arbitrage (with bots)
Run bot monitoring Binance and Uniswap
When spread >0.5%, execute both sides
Requires: co-located server, $100k+ capital, low fees
Returns: 20-40% APY for pros
3. Funding rate arbitrage
Not cross-exchange, but similar
Long spot on Binance, short perp on Bybit
Capture funding
Requires capital on both, but no transfer
Returns: 10-25% APY
I do funding arb via Binance and Bybit. No transfer latency, pure spread capture.
Minimum viable setup:
$200k capital ($20k on 10 exchanges)
Co-located servers ($5k/month)
Kafka streaming infrastructure ($2k/month)
Smart order router (build or buy: $50k+)
Cross-exchange margin account (Gate CrossEx or similar)
Total cost: $100k+ setup, $7k/month
Expected returns:
0.05-0.1% per arb
10-20 arbs per day
$200k capital
Gross: $100-400/day
Net after costs: $0-200/day
ROI: 0-36% annually
Not worth it for retail.
Speed begins with information. Most retail systems continuously request price updates. Pros use Apache Kafka event streaming architecture, remaining continuously connected to market data feeds.
Through Kafka, they receive every trade, every order book update, in real-time. You poll every second, miss 999 updates.
This is why manual arbitrage is nearly impossible now.
Activity clusters on Ethereum-centric L1-L2 pairs, grows 5.5x over study period, and surges—higher volume, more trades, lower fees—after the Dencun upgrade (March 13, 2024).
Dencun reduced L2 fees 90%. Made cross-chain arb viable. But still requires inventory on each chain, not bridging.
Instead of cross-exchange arb, do:
1. Provide liquidity
Market make on single exchange
Earn spread + rebates
No cross-exchange risk
Use 3Commas grid bots
2. Statistical arbitrage
Trade correlated pairs (BTC/ETH ratio)
Mean reversion, not cross-exchange
Works with retail latency
3. Use professional tools
Don't build infrastructure
Use platforms like Aonica, etc.
They provide edge, you provide capital
Revenue share model
I stopped cross-exchange arb in 2022 after losing $2,400 to latency. Now I:
Keep $50k on Binance, $50k on Bybit
Do funding rate arb (no transfer)
Use Coinigy to monitor spreads
When spread >0.3%, I manually execute both sides
Takes 10 seconds, no transfer
Returns: ∼15% APY
This works because I'm not moving capital, just taking offsetting positions.
What CrossEX represents is infrastructure that may not draw retail attention but does matter for how professional capital deploys. Execution routing was first generation. Cross-venue margin pooling is second.
The edge isn't finding arbs (everyone sees them). The edge is:
Having capital pre-positioned
Executing in <10ms
Paying 0.01% fees (vs your 0.1%)
Sharing margin across venues
Retail can't compete on any of these.
Cross-exchange arbitrage: why it has become nearly impossible for retail, and the infrastructure edge that makes it viable.
Why retail fails:
Settlement latency (10-30 min) exposes to price risk
Need capital on both exchanges (50% efficiency)
Polling API vs Kafka streaming (1000x slower)
50-200ms latency vs <0.5ms co-location
Manual execution vs AI in fractions of second
What pros have:
Pre-funded accounts on 10+ exchanges
Co-located servers, Kafka streaming
Cross-venue margin pooling (Gate CrossEx)
Smart order routing, AI spread detection
0.01% fees, <10ms execution
The data:
242,535 arbitrages, $868M volume (2023-2024)
41% of markets exhibit arbitrage
Activity grew 5.5x, surged after Dencun
But manual arbitrage is nearly impossible now
What works for retail:
Don't do cross-exchange with transfers
Do funding rate arb (capital on both, no transfer)
Or triangular arb on single DEX
Or provide liquidity, earn spread
I learned losing $150 on a "sure" arb. The spread was real, but latency killed me. Now I only arb where I have capital on both sides and can execute simultaneously. That's the infrastructure edge — and without it, you're just donating to pros.