A common claim is that tech companies hold the world’s largest databases, and that science is a rounding error. The fact is that “largest” has no shared unit, and the counted public stock is a physics archive. In December 2025 CERN said it had stored 1 exabyte of experimental data from the Large Hadron Collider, after more than 15 years of running. An exabyte is a million terabytes. Most of it sits on magnetic tape, because analysis can happen years later. CERN said this stock is about 10% of what the next decade will require. The High-Luminosity upgrade, expected in the middle of the 2030s, is planned to generate about 10 times more data.

Biology publishes a different unit. GenBank’s 2025 update reported 34 trillion base pairs from more than 4.7 billion nucleotide sequences, for 581,000 formally described species. The archive exchanges records daily with the European Nucleotide Archive and the DNA Data Bank of Japan. A base pair is not a byte in this table. Adding 34 trillion to 1 exabyte is a unit error. The Worldwide LHC Computing Grid is a different object again: about 160 computing centres in more than 40 countries, built to move and analyse the collider’s output, not to be a single disk.

Private platforms can be larger. They have not published the stock. A search index, a video library, and an ad profile are not the same database, and none of them releases a figure next to CERN’s 1 exabyte. Older papers guessed at YouTube’s growth from upload rates. A guess is not a census. The claim that science is a rounding error treats an unpublished company total as a fact and a published archive as small. The ranking does not exist until the units match.

CERN counts experimental data sent to its storage system. Raw collisions are filtered before that step. The 1 exabyte is the stored remainder, not every proton collision. Tape is the archive. Disk is the working copy. The grid figure of about 160 centres is a collaboration count, not a second exabyte.

GenBank counts nucleotide sequences and base pairs. Daily exchange with two partner archives is meant to avoid a split record, not to triple the total. A video hour and a gene sequence cannot be ranked against an exabyte without a conversion the archives do not publish.

It is settled that CERN had stored 1 exabyte of LHC data by December 2025, and that GenBank held 34 trillion base pairs. It is not settled that a tech company is larger, or that science is a rounding error. Use 1 exabyte for the counted physics stock. Use 34 trillion only for base pairs. Use “not published” for company totals. Do not rank them on one chart.