Telegram’s gold rushes, seen through 400 channel niches
Estimated creation intervals reveal unusually large cohorts, long silences and younger channels among today’s large ones. Explore the data by topic and language.
Telegram’s numeric channel IDs leave a rough creation clock behind. We joined that clock to a semantic map of 451,369 public channels, grouped the non-adult sample into 400 niches, and asked which periods left unusually large cohorts.
The rush has a shape
A large cohort alone is not unusual: Telegram itself grew. For each niche, we compare its dated channels with dated channels in the same languages. A lift of 4× means four times as many channels in that creation interval as the language mix would predict. It does not mean four times the monthly growth rate.
nft / crypto
11.9×2025-08-04 – 2025-09-25
86 observed vs 7.2 expected
blockchain / cryptocurrency
2.9×2024-05-16 – 2024-08-24
277 observed vs 95.1 expected
vpn / proxy
3.4×2022-09-27 – 2023-03-09
147 observed vs 43.7 expected
french / series
3.8×2023-03-09 – 2023-09-26
67 observed vs 17.8 expected
These are cluster labels inferred from frequent, distinctive tags. Some clusters combine adjacent subjects. The three largest channel examples in each detail view make that grouping inspectable; the boundaries are exploratory, rather than a definitive taxonomy of Telegram.
Gifts and giveaways
The gifts cluster contains a concentration of young channels, including channels labelled around NFTs and giveaways. A rush can remain visible in IDs even after some of its participants stop posting.
928 channels; 57.7% silent among 887 with usable activity timestamps. Recent entrants account for 72.1% of 147 dated channels with 10K+ subscribers.
Web3 and game communities
This broad blockchain cluster includes Hamster Kombat among its largest examples. Its creation spike is visible in May–August 2024, but the cluster is broader than tap-to-earn games alone.
1,772 channels; 65.9% silent among 1,647 with usable activity timestamps. Recent entrants account for 38.7% of 1,294 dated channels with 10K+ subscribers.
Persian VPN and proxy channels
One Persian proxy cluster has an unusually large September 2022–March 2023 cohort. A separate proxy cluster peaks later. Adjacent clusters can describe different waves inside the same subject; the timing alone does not identify their cause.
883 channels; 23.5% silent among 843 with usable activity timestamps. Recent entrants account for 9.1% of 679 dated channels with 10K+ subscribers.
French-language series
A French entertainment cluster has an unusually large spring–summer 2023 cohort. Most channels in this cluster were silent at observation. We cannot tell from these data why they stopped, or whether any were removed.
299 channels; 89.8% silent among 274 with usable activity timestamps. Recent entrants account for 10.4% of 201 dated channels with 10K+ subscribers.
Explore all 400 niches
Choose a niche to compare its creation intervals with the language baseline. Sort by recent entrants to see where younger channels appear among those that currently have at least 10,000 subscribers. This describes the present sample, not the probability that a new channel will reach that size.
400 niches · recent entrants use dated channels with 10K+ subscribers
| Niche | Language | Channels | Recent entrants | Silent |
|---|---|---|---|---|
| ar 93% | 4,362 | 3.8%n = 740 | 26.0%n = 3,844 | |
| ar 66% | 4,156 | 2.0%n = 50 | 30.1%n = 3,436 | |
| ar 67% | 3,341 | 0.0%n = 32 | 33.6%n = 2,968 | |
| ar 81% | 3,260 | 4.8%n = 21 | 35.4%n = 2,827 | |
| en 40% | 3,219 | 30.7%n = 75 | 55.9%n = 3,080 | |
| ru 27% | 3,214 | 13.2%n = 38 | 30.5%n = 2,957 | |
| uk 74% | 3,189 | 4.3%n = 1,386 | 19.9%n = 2,774 | |
| ru 99% | 2,876 | 6.7%n = 1,214 | 17.3%n = 2,433 | |
| ru 46% | 2,847 | 14.9%n = 276 | 33.5%n = 2,201 | |
| ru 94% | 2,803 | 6.6%n = 1,054 | 22.6%n = 2,261 | |
| ru 93% | 2,741 | 6.0%n = 1,338 | 25.2%n = 1,926 | |
| ru 91% | 2,719 | 9.9%n = 182 | 31.9%n = 2,408 | |
| fa 88% | 2,704 | 7.3%n = 1,334 | 14.0%n = 2,133 | |
| ru 99% | 2,617 | 12.2%n = 1,153 | 19.6%n = 1,716 | |
| ru 93% | 2,610 | 10.5%n = 76 | 25.1%n = 1,949 | |
| fa 78% | 2,431 | 8.7%n = 69 | 33.6%n = 1,905 | |
| ru 96% | 2,429 | 5.4%n = 682 | 37.1%n = 2,042 | |
| ar 48% | 2,410 | 2.7%n = 146 | 41.1%n = 2,151 | |
| ru 24% | 2,407 | 23.1%n = 39 | 41.3%n = 2,113 | |
| en 24% | 2,386 | 26.7%n = 30 | 43.5%n = 2,154 |
Silence is an observation, not an obituary
364,559 channels have both a last-post timestamp and an observation date. Of these, 33.5% had not posted for more than 180 days when observed. Missing post timestamps stay unknown. We do not count them as silent, and a silent channel may resume posting.
Observations span 2026-04-15 to 2026-10-06; this is not a live activity census. The index snapshot is dated 2026-10-10. Different niches can have different observation coverage. The detail view shows the denominator and observation range for each one.
An old channel can join a new rush
The clock dates a channel, not its current subject. A channel about a new game may carry an older numeric ID after a rename or change of topic. Our study of renamed channels documents this using current titles and public introduction dates. An early ID in a new niche is not proof that the niche already existed then.
Try your own subject
The Telegram niche checker takes a topic or @channel and reports a close semantic sample, language mix, dated entrants and silence. It uses similarity and shared tags, so its sample differs from the 400 fixed clusters in this article.
The channel map places channels about similar subjects next to each other. Use it to explore the subjects behind these cohorts.
Method and reproducibility
- Start with 451,369 channels in the multilingual-e5-small index from 2026-10-10. Exclude adult categories, adult tags and explicit titles; retain one row per known numeric Telegram ID. Rows without a known ID remain separate. 23,982 index rows are excluded or duplicates.
- Use the existing normalized 384-dimensional topic/description vectors, weighted 0.6/0.4. MiniBatchKMeans, 400 clusters, seed 16, five initializations, batch size 4,096.
- Map numeric IDs to empirical ID intervals. Merge November 2017–June 2021 into one cohort because ID blocks overlap. IDs beyond the calibrated upper bound and missing IDs stay undated. Intervals have unequal lengths; the chart uses counts, not rates.
- For each dated channel, use its language’s cohort distribution as the expected distribution. Sum those expectations within each cluster. Display lift only where at least five channels are expected.
- Recent entrants: creation interval starts on or after May 16, 2024; denominator: dated channels with 10,000+ current subscribers. Silence: last post more than 180 days before the later of analyzed_at and checked_at. No timestamps, or a post after observation, means unknown.
Build script: rnd-research/build_gold_rush_snapshot.py. The downloadable snapshot contains all 400 aggregates, cohort definitions, expected counts, tag labels and public channel examples. Download the JSON data →