You’d think answering that would be straightforward.
After all, a spot Bitcoin is a spot Bitcoin. Fungible across venues, tradable across regions. And yet, even for seasoned market participants, simply asking “what’s the price?” opens the door to a thicket of nuance that most don’t see.
Same asset, different places
In traditional finance, trading tends to be more neatly packaged. You co-locate in a server rack near the matching engine, the venue tells you where that is, and your edge is in how fast and smart your systems are.
In crypto? The rules change. Matching engines sit inside opaque AWS regions, scattered across the globe. Binance may run out of Tokyo, Bybit out of Singapore, Coinbase out of Virginia. Trades route over the public internet, and the difference in latency—35ms here, 120ms there—is not trivial. It’s an eternity when your internal code executes in microseconds.
You’re not just trying to be fast. You’re trying to make decisions using incomplete, time-lagged information across venues that don’t speak in a consistent format.
So when someone says “Bitcoin is trading at $69,420,” it depends: where? And when? And how many milliseconds late is that quote, and how deep is the liquidity at that price? Is it real, or has it already moved? Is that the price to buy, or the price to sell?
You can’t arbitrage uncertainty
There’s this romantic idea that if one venue quotes $100 more than another, you just scoop it up and pocket the spread. But the real world doesn’t play that nicely (anymore!).
Not only do you need to account for latency, fees, and slippage, you also need to model whether those price differences are meaningful.
And even if you could arbitrage, you’re still left with the question: how do you know what the real price is?
You’re trading something that’s meant to be the same. But you’re receiving delayed information, routing it over the internet, and trying to act on it in time. The entire game becomes one of inference.
That’s where the models come in. You reach for your maths toolkit. You say, right, I know this quote is slightly ‘stale’ given I’m a machine making decisions in a different location. What’s the best estimate I can make of what’s going on right now? Sometimes, your best guess is just to assume the price hasn’t changed. Buit, you can do better. You start building models to improve your view of the world.
This isn’t just about latency between venues either. You also need to factor in depth. Price is a simplification. The real object is the full order book. Two venues might both show $69,420, but one has size behind it and the other doesn’t. Suddenly that quote means something different.
Then you get to DeFi
The crypto trading space originally built up around centralised exchanges. They were basically copies of tradfi limit order books. But from 2019 onwards, with Uniswap and others, you got a new model entirely.
Now you’re trading in a system where there’s no matching engine at all. You’re connected to a node, which broadcasts to a network, and you hope your trade goes through in the next block.
That block might be constructed by a builder who has their own rules. Maybe they prioritise gas. Maybe they prioritise a specific searcher. It’s not time-based anymore. You don’t know where your transaction will land in the sequence.
Understanding how these transactions interact, and when they’ll confirm, is another layer of modelling entirely. It’s not just a case of writing fast code. It’s writing smart code that knows it might be operating with stale or partial information, and still makes the right call.
It’s not just dumping on retail
This is what makes market making in crypto hard. You’ve got to manage risk across multiple venues, protocols, and regions. You’re dealing with incomplete data, stale data, and noisy data. You’re building models that constantly need refining. You’re taking trades on infrastructure that was never built for co-location. And you’re doing it all in real time.
So when a market maker gets dragged on Crypto Twitter for “dumping,” it’s not always a fair shot, but it’s not always baseless either. Some teams are bad actors. Some are structuring deals that leave projects and communities exposed. There’s no smoke without fire.
But that’s not the full picture. There are also firms operating across 80 venues in multiple paradigms across many different types of asset, managing risk responsibly, and doing hard technical work to provide real liquidity. They’re building infrastructure, maintaining orderly books, and making markets in the way the term was originally meant.
When done properly, this work sits at the intersection of distributed systems, low-latency engineering, statistics, and game theory. And some teams out there are genuinely excellent at it.
🎙️ If this sort of thing interests you, the full conversation goes even deeper. This article is adapted from episode three of The Local, LO:TECH’s new podcast about trading, infrastructure, and crypto’s weird edges. You can find it on Youtube, Spotify, Apple, or wherever you get your podcasts.
