Who Advertises on Facebook in Malaysia (2026)

12,943 distinct Malaysian brands were running at least one Meta ad at their last check. The median one had 4 ads live. That median is the number to hold on to, because the totals hide it. 23.1% of Malaysian advertisers run exactly one live ad and between them hold 1.3% of all the advertising. At the other end, the 438 brands running a hundred or more ads hold 39.9% of it. The concentration curve says the same thing more precisely. The busiest 1% of advertisers, 129 brands, hold 20.1% of every running Malaysian ad. The busiest 10% hold 63.3%, and the busiest 25% hold 83.1%. Anything built on ad counts is therefore describing the top of that curve. A statement like "Malaysian advertisers favour video" is, in practice, a statement about a few hundred brands running industrial volumes of creative, and the small business reading it has a different problem entirely.

An ad id is not a creative. This archive is full of one image re-uploaded across many ad objects, which is what happens when an advertiser duplicates an ad set, runs the same visual against several audiences, or lets a dynamic format generate variants. So the same brands were counted twice: once by how many ads they hold, and once by how many visually distinct creatives. 6,388 brands hold five or more live ads. 5,671 hold five or more distinct live creatives. The second number is the honest one and it is 43.8% of the advertiser base. The difference between those two figures is worth a sentence of its own, because it is the size of the duplication problem in any Malaysian Meta dataset, including ours. Roughly one brand in fourteen looks data-rich when counted by ad and thin when counted by creative, and any study that ranks advertisers by ad count without saying which it means will over-rank whoever duplicates most. The median advertiser holds 4 distinct live creatives, which is to say the typical Malaysian brand on Meta is running a handful of images, not a creative programme.

Counted by advertiser rather than by ad, the largest Malaysian verticals are services with 6,645 brands and retail and marketplace with 6,455. The more useful column is the last one: the share of each vertical's advertisers holding five or more distinct live creatives. Healthcare and consumer goods run high, above three in five, while real estate sits far lower. A vertical with many advertisers and few of them data-rich is a vertical of one-ad businesses, and it behaves differently from one with fewer, busier brands. Verticals are model-assigned and multi-label here, so a brand carrying two labels is counted in both and the rows deliberately do not sum to the total. That is correct for a per-vertical rate and wrong for an addition, and the CSV says so on every row.

The obvious way to count advertisers is to ask the index for a distribution over the brand field. It does not work, and the way it fails is quiet enough to be worth publishing. That index caps a facet at a hundred values. A field with more than a hundred distinct values comes back truncated ALPHABETICALLY, with no error and no flag, so a naive count would have returned a confident, wrong, alphabetically-ordered slice of the Malaysian advertiser base. With more than twelve thousand brands there is also no partition that fixes it: the vertical labels overlap, so partitioning by them double-counts. So this study does not use a facet. It reads the documents, a thousand at a time, and counts brands here rather than asking the index to. Every running Malaysian ad in the archive was walked exactly once: 222,809 of them, against 222,921 counted by the archive's own totals at the same snapshot, a difference that is ordinary ingest during the minutes the scan took. Nothing is sampled and nothing is estimated. Identity is the advertiser's stable id rather than its display name, because names change and several brands share one. The two counts are close, 12,943 ids against 12,928 distinct names, which is itself the answer to whether name-based counting would have been good enough.

It cannot tell you how many Malaysian businesses advertise on Meta. It tells you how many were running an ad our archive had seen and last checked as still running. A business that advertises in bursts is absent between bursts, and every count here is a floor for that reason. It cannot tell you anything about money. There is no spend, no budget, no cost and no return in this dataset, and an advertiser holding a hundred live ads may be spending less than one holding three. And it cannot tell you who any of these advertisers are. That is deliberate rather than a limitation: a page that named the top hundred would be a company list dressed as a study, and this one is about the shape of the base rather than its members.

The whole dataset is one CSV under CC BY 4.0, free to reuse including commercially with attribution to AdPlay.ai. It carries the advertiser totals, both bucket distributions, the concentration curve and the per-vertical counts, with the method note on every row. It contains no brand names.