How to Open a CSV File That Excel Will Not Open

There is a specific moment this page is for. The file is 900 MB, Excel has been showing a progress bar for four minutes, and you are starting to suspect it is not going to finish.

The short answer

Open it in a viewer that queries the file on disk instead of loading it into memory. In DataTray, open Data Explorer and drop the file in: it detects the delimiter and encoding, shows you what it detected so you can correct it, counts the rows, and starts showing you data. Filtering and sorting re-run against the file rather than against a loaded copy, so they work at any size. It is free, and nothing is uploaded.

Why Excel gives up

Two separate limits, and people usually hit the second one first.

The row ceiling is 1,048,576. A spreadsheet has that many rows and no more, so a file with two million rows cannot be represented at all — Excel will open it and silently show you the first million, which is worse than refusing.

Memory is the real constraint. Excel loads the whole file into RAM and builds a full object model over it. A 900 MB CSV becomes several gigabytes of working set, and long before the row ceiling matters the machine is swapping.

A viewer that pages against the file avoids both. It never holds more than a screenful, so the file's size is a property of your disk rather than a limit on what you can do.

The one-column problem

The single most common CSV complaint, and it is not really a size problem.

A CSV does not record its own delimiter. If the file was written with semicolons — normal across much of Europe, where the comma is the decimal separator — and your machine expects commas, every row parses as one enormous field. Same for a file written as UTF-8 opened as Windows-1252, which is where Café becomes Café.

The fix is for the reader to show you its guess rather than make it silently. DataTray states what it found — "semicolon-delimited, UTF-8 with BOM, header on row 1" — with every part overridable before anything is read. When the guess is wrong you change it in a click instead of wondering why the file looks broken.

Filtering that works on a big file

The distinction that matters in a viewer: does filtering search the file, or search what has been loaded?

Loaded-copy filtering is why so many viewers feel fine on a small file and useless on a large one. DataTray compiles filters to a query against the file and sorts the same way, so narrowing two million rows to the eleven you want takes about as long as narrowing two thousand.

Paging is by key rather than by offset, which is the technical reason scrolling stays fast deep into a file instead of degrading the further down you go.

Look at the columns, not just the rows

Once the file is open, the more useful question is usually not "what is in row 40,000" but "what is in this column at all". A second tab profiles every column: inferred type, what it actually holds, null rate, distinct count, ranges, the top ten values, and a distribution.

That is where you find out the identifier is not unique, a date column has four formats in it, and a field you were about to build a report on is 60% empty.

Formats other than CSV

The same viewer opens Excel workbooks, Parquet files and folders of them, JSON, NDJSON and XML. Parquet is worth calling out because Windows has no built-in answer for it at all — a .parquet file has no icon, no preview and no double-click behaviour, and every guide tells you to install Python.

Nothing uploads

Online CSV viewers exist and some are good, but they all share one property: the file goes to somebody else's server. For a public dataset that is fine. For an export containing customers, salaries or patient records it is a decision somebody should have made deliberately.

A desktop viewer removes the question. There is no upload step to opt out of, no size cap because nobody is paying for bandwidth, and no queue.

---

Every other free way to do this asks you to upload the file first. Opening a file should not require that.