Arcus has 179 equities live and 85,000 traders queued for perps. In its closed beta, five markets still take 83% of trades. What it puts in front of that queue decides the rest.
Centaur · Research
September 23, 2026 · 10 min read
Arcus launched on 1 July 2026, the same day Robinhood Chain went live, built by the dYdX Labs team in partnership with Robinhood Crypto. Ten weeks in, it runs two products at two different stages.
Spot is open to anyone. Stock tokens, no commission, trading around the clock, with 179 equities live in the API today alongside indices and commodities. Their own line for it is trade while the world sleeps, and for equities that is a genuinely new thing to offer.
Perpetuals are still in closed beta. The product page promises 125+ markets and up to 50x leverage, and the way in is a waitlist.
Underneath both sits infrastructure most venues never finish. A full REST and WebSocket API with mainnet and testnet on identical paths and signing. Programmatic order placement, batch operations, a dead man's switch, self-served API keys, published latency guidance. Real-world-asset perpetuals with their own margin treatment around market hours. A live vaults product. This is an exchange built to be traded by people and by machines.
So the hard engineering is done, and a queue of users is waiting on the other side of a gate. The question is what they trade when it opens.
Closed betas are a useful preview, because the people inside them are the most motivated users a venue will ever have. Arcus says more than a hundred traders are shaping its roadmap. These are not passive signups.
On 16 September, the beta ran 241,737 perpetual trades across 58 live markets. Five of them took 88% of that activity, and all five were crypto.
| Market | Trades, 24h | Share |
|---|---|---|
| Bitcoin | 89,061 | 36.8% |
| Solana | 56,940 | 23.6% |
| Ethereum | 31,385 | 13.0% |
| Zcash | 23,150 | 9.6% |
| Hyperliquid | 11,489 | 4.8% |
| Top five combined | 212,025 | 88% |
Thirty of the live perp markets are equities, and together they saw under eighteen thousand trades. Microsoft traded twenty-six times in the day. Marvell traded seven. Source: Arcus markets API, 16 September 2026.
Nothing is wrong with those markets. They are live, priced and clearing, and access is restricted, so low counts are expected. The signal is in the distribution, not the totals. Given a menu spanning equities, commodities, indices and crypto, the most engaged cohort Arcus will ever have still put nearly nine tenths of its attention into five crypto tickers.
A public queue will not behave better than that. It will behave worse, because it will arrive with less conviction and less context.
This is the industry's unsolved problem, and it sits after acquisition.
| Signal | Figure |
|---|---|
| Crypto owners, August 2026 | 774 million, an all-time high |
| Top-ten exchange spot volume | Fell more than half in two quarters |
| Coinbase sales and marketing, FY2025 | Up 62% against 9% revenue growth |
| eToro registered vs funded accounts | 40 million vs 4.02 million |
More holders, less trading. The step nobody owns is the one after the deposit clears, when a user has money in the account and nothing chosen to put it in. The window to answer that is short: two thirds of UK investors aged 18 to 40 finalise a decision within 24 hours, and for many the first session is the whole funnel.
A waitlist concentrates that moment. Instead of arriving in a trickle, a cohort lands at once, all of them new to the product, all of them facing the same question on the same day.
Figures per Why New Exchange Users Never Place a First Trade.
The instinctive response is to list more, and the perps page already promises 125+ markets against the 58 live now.
MEXC ran that experiment, listing 879 new perpetual contracts in sixteen months, more than any competitor, and fell from second place to seventh while its volume more than halved.
To a user who does not know what to trade, a longer list is a harder question. Going from 58 markets to 125 doubles the menu without touching the decision. What keeps someone on a venue is whether it made them feel capable of choosing.
Arcus already computes a ranking of its own traders by volume, profit and fees paid. It exists only as an API response, with no page anywhere in the product, and of the top hundred by all-time volume, twenty-three are in profit. Ranking by activity surfaces the most active, which is a different group from the most worth following, and on a ten-week-old venue it can only ever describe the people already inside.
The alternative is a record that exists independently of Arcus. Centaur reads what public traders post on X and Telegram and turns it into trade discovery data: what was opened, closed or resized, by whom, and when, each event linked back to the post it came from. Every record is a position with a direction and a date, which is a different object from a mention count or a sentiment score.
On 16 September, 61% of Centaur's open positions sat in assets Arcus already lists, and 113 of Arcus's 200 non-crypto markets had at least one tracked trader holding a live position. The equities book that barely trades in the beta is not empty of interest. That interest is happening somewhere Arcus cannot see.
The simplest way to describe it is wallet tracking, for socials. Arcus can put it in front of a user in two ways, and Centaur powers both. One is copy trading, where a user follows a tracked trader's positions directly.
Arcus already has the vault machinery this needs, NAV, rebalancing and deposits, and its current vaults track a fixed multiple of one asset rather than a person, so the step is a new vault type rather than a new system.
The other is a discovery layer, where the individual record and the aggregate do different jobs. A single trader's history is what a user follows, checks and forms a view against, and it is the most direct thing to put on a market page.
The aggregate is what a competitor watching the same accounts cannot reconstruct, because the value sits in the population rather than any one record. Both run off the same feed, and Arcus could ship either on the pages it already has.
| What Arcus could show | Where it goes, and what the user sees |
|---|---|
| Who is in this market | A panel on the Nvidia page: the tracked traders holding it, long or short, with the date each opened and a link to the post |
| Which markets are waking up | A list of the Arcus markets picking up new positions this week, so a quiet ticker surfaces while the interest is live |
| A cohort trading one theme | The traders currently spread across the semiconductor and AI-infrastructure names Arcus lists, as one group rather than ten separate tickers |
| Traders worth following | A board ranked on record over a stated window, populated from day one because it does not depend on anyone having traded on Arcus |
| Where everyone is on one side | Markets where tracked positioning is almost entirely long, shown as a crowding read next to the order book |
One more is worth noting because of how Arcus is built. Centaur follows the same person through crypto and equities, so a trader holding a Bitcoin position and a semiconductor position shows up as one book. On a venue listing one of those legs that is background. Arcus lists both, in one wallet, under one collateral, so it is a view a user there can act on.
Everything above is a surface, and surfaces are Arcus's to build. The module on a market page, the discovery feed, the ranked view, those are frontend work on top of infrastructure that already exists.
The layer underneath is a different proposition.
There is a second cost that rarely makes the estimate. All of it has to be right before the feature ships at all. A discovery surface built on a noisy extraction layer does not degrade gracefully. It shows wrong positions to people who are about to act on them, which is worse than shipping nothing.
If engineering time is the constraint on the roadmap, and ten weeks after launch with perps still gated it probably is, that is the argument for licensing this layer and spending the team on the parts users actually see.
Arcus chose to be programmable. Testnet and mainnet share identical paths and signing, keys are self-served, the WebSocket carries positions, orders and fills, and the documentation covers latency. Third parties can build real products on this exchange.
What they cannot build is anything needing a view of the world. Every market-data channel Arcus publishes is price: orderbook, best bid and offer, oracle prices, funding, lending rates, candles. A developer building on Arcus today has execution and no signal, which confines the output to execution tooling.
Add trade discovery data and the buildable set widens. Screeners ranking Arcus markets by what tracked traders are doing instead of by price change. Research surfaces comparing traders across the venue's own tickers. Alerts when someone a user follows enters a market.
The same holds for automated strategies. An agent trading Arcus already has everything it needs to execute and nothing to form a thesis from, and typed trade events are the right shape for that gap. An agent cannot place an order from a number describing how bullish a timeline feels. It can act on an event saying a named trader opened a position in a specific market on a specific date. Every product built that way is volume Arcus did not have to acquire.
Discovery layers get built once, and the cheapest time to build one is before the users arrive rather than after a cohort has already bounced off the order book.
Arcus has ten weeks of beta data telling it exactly what happens when a motivated trader meets a long menu without guidance. The waitlist is the second chance at that moment, at much larger scale.
Get access and pull the current open book against your own market list.
All Arcus and Centaur figures were pulled or browsed on 16 September 2026.
