How Much Capital Do You Actually Need to Run an AI Trading Agent?
The honest answer is not a marketing number. It comes from position sizing, gas reserves and fee drag. Here is the arithmetic that decides whether an agent can trade its strategy properly or is starved before it starts.
This is the most common question we get, and most answers to it are marketing. The real answer comes from arithmetic, and you can work it out yourself in about five minutes.
An agent is not starved because the number in its wallet looks small. It is starved when that number is too small to run its own strategy properly. Three things decide where that line sits.
1. Your allocation gets divided, not spent
An agent with a max_concurrent_positions of 4 does not put your whole balance into one trade. It splits across open positions.
Allocate $100 to an agent holding up to 4 positions at once and each position is roughly $25. That is the number that matters, because every cost below is charged per position, not per account.
The trap is that people evaluate the balance and ignore the divisor. A $60 allocation across 5 concurrent positions is a $12 trade — and a $12 trade on Solana behaves very differently from a $120 one.
2. Fixed costs are charged per trade, not per dollar
Every on-chain position costs roughly the same to open and close regardless of size:
- Network fee and priority fee on the swap in
- The same again on the swap out
- Token account rent, where a new account is opened
- Spread and slippage, which scale with size but never reach zero
Call the fixed portion a few tens of cents per round trip. On a $200 position that is noise. On a $12 position it is a meaningful share of the outcome before the strategy has done anything at all.
This is why fee drag is not a rounding error at small sizes. It is a fixed toll that the same strategy pays whether the position is $12 or $1,200, and it is the single biggest reason small allocations underperform the same configuration at larger size.
3. You must hold gas you never trade
An agent wallet on Solana has to keep SOL in reserve to pay for transactions. QWNT holds back a small reserve — currently 0.015 SOL — and will not trade the account below it.
That reserve is not capital at work. It is a permanent carve-out. If an agent's balance drops near it, the agent stops opening positions rather than stranding itself with tokens it cannot afford to sell. An agent that cannot pay gas cannot exit, and an agent that cannot exit has no risk control at all.
Budget the reserve separately from the money you intend to trade.
Working the numbers backwards
Rather than asking "how much should I deposit", ask "what is the smallest position this strategy should take?" and work up.
- Decide a minimum sensible position size where fixed costs are a small share of the trade — as a rule of thumb, aim for fixed costs under about 1% of position size.
- Multiply by
max_concurrent_positions. - Add the gas reserve.
- Add headroom, because realised losses reduce the capital budget for the next trade.
A worked example for a 4-position memecoin agent:
| Component | Amount |
|---|---|
| Minimum sensible position | $50 |
| × 4 concurrent positions | $200 |
| Gas reserve | ~$2 |
| Drawdown headroom (25%) | $50 |
| Working minimum | ~$252 |
Run the same calculation with a $40 allocation and each position is $10, fixed costs are several percent of every trade, and the strategy is paying a toll it was never designed to absorb.
Fewer positions beat more positions at small size
If your available capital is below that working minimum, the fix is not a bigger deposit. It is a smaller max_concurrent_positions.
Two positions of $50 will behave far more like the intended strategy than eight positions of $12. You give up diversification across tokens, which matters, but diversification into positions too small to survive their own costs is not diversification. It is spreading the same toll across more trades.
Concentration is the correct response to a small account. Most people do the opposite.
What this means in practice
- Under roughly $250, run one or two concurrent positions, expect fee drag to be visible, and treat the account as a way to learn the mechanics rather than a return engine.
- $250 to $1,000, a 3-4 position agent can run its strategy close to as designed.
- Above $1,000, position count stops being the binding constraint and liquidity depth on the tokens you trade starts to be.
None of this is a claim about returns. It is a claim about whether the strategy you configured is the strategy that actually runs. Below a certain size it is not — the costs quietly rewrite it.
If you are not sure where you land, run the strategy in paper mode first and read our risk disclosure before funding anything. The arithmetic above does not depend on trusting us; you can check every line of it against your own wallet.