P2P vs On-Site Inventory: the Marketplace Split That Decides Everything
Strip away the branding, the fee schedules, and the Trustpilot scores, and every CS2 marketplace answers one architectural question: when you click buy, does the item come from another player's Steam inventory, or from stock the venue already holds? That single design choice — peer-to-peer versus on-site inventory — quietly determines the fees you pay, how fast you're delivered, when the trade lock bites, whose failure can hurt you, and whether a machine can trade on your behalf at all. Both architectures, at their reputable best, will sell you the same skin for less than Steam — the split just decides how. This is the structural explainer the rest of our marketplace coverage stands on.
The two architectures
Peer-to-peer (P2P): the seller keeps the item in their own Steam inventory. When you buy, the venue escrows your payment and instructs the seller to send you a Steam trade offer. The marketplace never touches the item; it's a matchmaker with an escrow account. CSFloat is the canonical western example — our review covers it — alongside a family of discount venues built the same way.
On-site inventory: the seller deposits the item into the venue's custody before or at listing. The item physically sits in marketplace-controlled storage; buying it is a database operation that moves ownership from seller's account to yours, instantly, with the proceeds landing in the seller's on-site balance. DMarket and SkinBaron are the on-site venues we know best; Steam itself is the model's giant, closed-loop cousin.
Everything else about a venue — pricing, speed, risk, tooling — is downstream of this choice. Take the dimensions one at a time.
Fees: P2P is cheaper because it does less
Custody costs money. An on-site venue runs bot fleets, storage infrastructure, instant-settlement plumbing, and support staff who own delivery problems end to end. A P2P venue outsources nearly all of that to its sellers — so it can charge low single digits where custody venues charge mid singles to low teens. Neither number is dishonest; they price different amounts of work. The full numbers live in our fee comparison, but the structural summary is: on P2P venues, you are part of the fulfillment machinery, and your compensation is the fee discount.
Delivery: instant versus "when the seller shows up"
On-site purchase settlement is measured in seconds — the item was already in the warehouse. P2P delivery is measured in "usually minutes, occasionally hours": the seller must be online, notice the sale, and send the trade before the fulfillment window expires. Most P2P trades complete uneventfully. But the variance is real, cancellations happen, and every buyer eventually meets the sale that silently times out. For a human buying one skin, that's an annoyance. For anything operating on a schedule, it's a design constraint.
Trade locks: paid now or deferred
Steam's roughly-7-day trade lock fires when an item moves into a new Steam inventory. P2P delivery is exactly that move — so every P2P purchase arrives locked. On-site purchases settle on the venue's books; the lock waits until you eventually withdraw the item to Steam, and until then you can hold or even resell inside the venue lock-free. We've mapped this per venue category in the trade-locks comparison; the one-line version is that on-site custody turns the lock from a recurring tax into a one-time, schedulable event.
Trust: counterparty risk versus custody risk
Here the trade-off genuinely cuts both ways, and honest people weigh it differently.
- P2P concentrates counterparty risk per transaction. You depend on an anonymous seller performing a manual step, wrapped in venue escrow, on top of Steam's reversibility rules. Escrow handles most failure modes; the residue is friction — timeouts, disputes, waiting.
- On-site concentrates custody risk at the venue. Your balance and items live on the marketplace's books. If the venue is hacked, freezes withdrawals, or dies, you're an unsecured creditor. This is not hypothetical: OPSkins was the biggest custody venue of its era before Valve's ban wiped out its model. Venue selection — track record, registration, payout reliability — is the entire risk control, which is why our safety tiering weights it above everything else.
Rule of thumb: P2P risk scales with how often you trade; custody risk scales with how much you leave on the venue. Frequent small buyers should mind the venue's balance sheet; occasional big buyers should mind the escrow terms.
Automation: the dimension where the split is absolute
On every dimension so far, the two models trade advantages. On automation they don't — and this is the part of the thesis we hold strongest, with the disclosure stated plainly: cs2stack is our tool, and it exists because of this exact asymmetry.
For software to buy skins unattended, every step must be executable and verifiable by a machine: query prices, place the order, confirm settlement, log the result. On-site venues with public APIs make each step a call — the purchase settles in custody instantly, so the bot can prove the item was acquired at the logged price, with no human in the loop. P2P venues structurally can't offer that: fulfillment runs through Steam trade offers into your own account, which means either you accept trades manually (no longer unattended) or you hand a bot your Steam credentials and trade authority (a security surface that every inventory-theft playbook exploits, and that no cautious person should automate). The on-site model is why hands-off automation works at all — the reason cs2stack runs on DMarket and SkinBaron rather than the cheaper P2P books, as the complete autobuy guide lays out step by step.
The scorecard
| Dimension | P2P | On-site inventory |
|---|---|---|
| Seller fees | Lowest (low single digits) | Higher (mid singles to low teens) |
| Delivery | Minutes to hours, seller-dependent | Instant settlement in custody |
| Trade lock | Every purchase, on arrival | Deferred until withdrawal |
| Main risk | Per-trade counterparty friction | Venue custody/solvency |
| Best for | Patient value buyers, max-margin sellers | Schedules, volume, resale agility |
| Unattended automation | Effectively no | Yes, where an API exists |
Choosing — and mixing
The split isn't a war with a winner; it's a matching problem. If you're a human buying occasionally and hunting the best cash price or a specific float, P2P's fee discount is real money and the delivery wait is irrelevant. If you're selling a grail at maximum margin and can wait for the right buyer, P2P again. If you're accumulating on a schedule, cycling inventory, running any kind of standing strategy — or simply unwilling to babysit trade offers — custody venues earn their fee many times over in settled-instantly certainty.
Most serious traders end up hybrid: automated accumulation on on-site venues, manual cherry-picking on P2P books, and a custody-risk budget that caps how much value sits on any single venue at once. Wherever you land, make the choice consciously — because as this page has tried to show, when you pick a marketplace, the architecture picks everything else.
And whichever mix you settle on, keep sight of the happy fact underneath the architecture: the reputable venues on both sides of the split are excellent places to buy below Steam prices. The lasting edge lives above the split entirely — comparing prices across the great marketplaces on every single purchase, because the cheaper book rotates daily. Architecture decides how you buy; comparison decides how well.