A dark pool is a private trading venue built mainly for institutions, where buyers and sellers match orders without showing them to the public before the trade happens. In the U.S. they are legal, registered alternative trading systems run by broker-dealers, and every trade they execute is reported to the public tape afterwards. What stays dark is the order book, not the trade.
That last distinction is the one most explanations skip, and it decides what dark pool data can and can't tell you.
Why dark pools exist
Imagine a pension fund that needs to buy two million shares of a company. If it posts that order on a public exchange, everyone sees a large buyer arrive. Other traders step in front of it, sellers raise their prices, and the fund ends up paying more than it planned. That cost is called market impact, and for large orders it can be bigger than any commission.
Dark pools exist to reduce it. The order rests out of sight, and it fills only when someone on the other side wants to trade at the agreed reference price, usually the midpoint of the public bid and ask. Neither side announces its size in advance, which lowers the risk of being front-run.
Are dark pools legal?
Yes. Dark pools that trade U.S.-listed stocks are registered with the SEC as alternative trading systems and run by broker-dealers. Since the SEC's 2018 rules, each must file a Form ATS-N describing how it operates, who runs it and how the operator's own trading interacts with it, and the SEC posts the filings publicly on its EDGAR system (SEC).
"Dark" means no pre-trade transparency. It does not mean unregulated, and it does not mean secret after the fact.
How much trading happens off-exchange
More than most people expect. Rosenblatt Securities' market-structure research puts off-exchange trading at 50.61% of U.S. equity volume in 2025, a second straight annual record, with dark pools accounting for 17.74% (Rosenblatt).
The gap between those two numbers matters. Off-exchange is not the same as dark pool: most of the rest is retail orders filled by wholesalers, plus trades brokers cross or fill from their own inventory. When a headline says "half of all trading is in dark pools," it is counting both.
What a dark pool "print" is
When a dark pool trade executes, the broker must report it to a FINRA trade reporting facility as soon as practicable, and during market hours no later than 10 seconds after execution (FINRA Rule 6380A). That report is the print: price, size and time, appearing on the consolidated tape marked as an off-exchange trade. The tape doesn't say which venue it came from, so a dark pool print looks the same as any other off-exchange trade.
FINRA also publishes weekly dark pool volume by venue and by stock, free, with a two-week delay for the largest stocks (FINRA's Tier 1, which includes the S&P 500 and Russell 1000) and four weeks for the rest (FINRA). That weekly data is where you can see which venues traded a stock. So the raw record is public. What takes work is reading it.
What dark pool data can and can't tell you
What it can show:
- Where size is concentrating. Off-exchange volume well above a stock's own normal level suggests large participants are active in that name.
- Where it printed. Block-sized prints clustering at a price level show where big trades were willing to transact.
- Persistence. Elevated activity across several sessions means more than a single large print.
What it can't show:
- Direction. A print doesn't say whether the initiator was buying or selling. Every trade has both.
- Intent. A block can be a new position, an exit, a hedge, an index rebalance or a fund moving shares between accounts.
- The whole picture. One stream of data, read alone, is ambiguous by design.
This is why "dark pool buying" headlines deserve suspicion. The print records that a trade happened at a size and price. Calling it buying is an interpretation, and usually an unstated one.
How AI changes the way you read dark pool data
The traditional way to follow dark pools is to watch a feed of large off-exchange prints and judge each one by eye. That made sense when only a few specialists looked. It doesn't scale now: about half of U.S. equity volume trades off-exchange, across thousands of stocks, every session. Reading that by hand is a volume problem, and volume is where software helps most.
Used well, AI changes four parts of the work:
- Baselines for every stock. "Unusual" means something different for a mega-cap that trades off-exchange all day and a small company that rarely does. AI can compare each stock's off-exchange activity to its own history, across the whole market, every day.
- Confluence. The real question is whether independent records point the same way. AI can line dark pool activity up against insider filings, 13F changes and options flow in the same window and describe where they agree and where they conflict.
- Memory. A single print says little. Activity that builds across several sessions says more. AI can track what happened yesterday and last week, so persistence shows up instead of getting lost in the daily noise.
- Plain English. It can turn tape data into a description a non-specialist can question: what traded, how unusual it is, what else happened around it, and what would change the read.
What AI still can't do is turn a print into a direction. Every trade has a buyer and a seller, and no model can read intent from a record that doesn't contain it. An honest tool says "unusual size traded here, and this is what else lines up." Be wary of any tool that claims dark prints tell it what institutions are buying.
You can try the approach with free data and any AI assistant. Take FINRA's weekly ATS volume for a stock you follow from its OTC Transparency data (free, published two to four weeks after the fact) and ask the assistant how the latest available week compares with the stock's own earlier weeks. Then ask what else happened in that window, such as filings, earnings or news, and what could explain the activity other than someone building a position. The point is the questions, not the answer.
Reading dark pool activity without chasing it
Dark pool data earns its place as context, inside a process, rather than as a trigger. The useful questions are the ones that test it against other evidence:
- Is the off-exchange activity unusual for this stock, or just normal for a heavily traded name?
- Do other records point the same way?
- What would have to be true for this activity to mean something, and what would show that it doesn't?
That is the same discipline behind building a framework rather than chasing forecasts: defined conditions, risk decided in advance, and your own reasoning about why an opportunity might exist.
It's also how we built EdgeForge: it reads off-exchange and dark pool activity against each stock's own baseline, lines it up with filings and options flow, tracks it across sessions, and describes in plain English what the records suggest together. It describes the evidence; the judgment stays with you. More on that split in what AI trading tools actually do.
Frequently asked questions
Are dark pools illegal? No. They are SEC-registered alternative trading systems run by broker-dealers, they file Form ATS-N disclosures with the SEC, and their trades are reported to FINRA within seconds.
Can individual traders see dark pool trades? Partly. Their trades appear on the consolidated tape after they happen, but marked only as off-exchange, mixed in with wholesaler trades. FINRA's free weekly data then shows volume by venue and stock. What no one sees is the order before it fills.
Is heavy dark pool activity bullish? Not by itself. A print doesn't show whether the initiator bought or sold, so it says that size traded, not which way. It becomes informative only next to other evidence.
Can AI predict prices from dark pool data? Not reliably. Prints record that size traded at a price, not which side started the trade or why. AI is useful for spotting what is unusual and what lines up with it, not for forecasting.
Why don't institutions just use exchanges? They do for much of their trading. For very large orders, showing the size in advance moves the price against them, and dark pools reduce that cost.
Researched and drafted with AI, and checked by an automated fact and compliance review.
Free guide
The AI Advantage
A plain-English walkthrough of the institutional data streams in this post — 13F filings, insider transactions, options flow, and dark pool prints — and where AI genuinely helps you read them.
- Every data source explained end to end
- How the streams combine into a picture
- Where AI helps, and where it can't
Educational content only. Nothing here is investment advice, a recommendation, or an offer to buy or sell any security. Trading involves substantial risk, including the possible loss of capital. Any figures or examples are illustrative only and do not represent actual or expected results.
Keep reading
What Is Smart Money? How Institutional Positioning Shows Up in the Data
Smart money is simply the market's largest, best-resourced participants — and while they don't announce their intentions, they do leave records.
Read the postWhat AI Trading Tools Actually Do (and What They Can't)
The most useful things AI does for traders have almost nothing to do with predicting prices. That's not a knock on the technology — it's a correction of the marketing.
Read the post