You export every French company created in 2026 from the Sirene open data to build a list of new businesses to contact. Your file shows one total. The French statistics office announces another, much higher one. Neither is wrong — they simply don't count the same population. We measured the gap on our local copy of the registry, and it changes how a "recently created companies" file should be read.
Two counters, two populations
The official figure first. INSEE counted 1,170,000 company creations in France in 2025, an all-time annual record, up 4.9% on 2024. The momentum carried into 2026: over the twelve months from July 2025 to June 2026, creations grew another +11.1% versus the previous twelve months. Since August 2025, monthly creations have crossed the 100,000 mark for the first time since the series began in 2000.
Now our counter. On a local copy of the Sirene registry — the official stock of August 1, 2026, plus the daily update feed applied through August 13 — we count 473,230 legal units created between January 1 and July 31, 2026 that are still active and publicly listed (non-restricted). That's a visible average of roughly 67,600 per month, while the official pace runs above 90,000.
Month by month: a slope that isn't real
| Creation month (2026) | Legal units still active and listed on 2026-08-13 |
|---|---|
| January | 95,516 |
| February | 65,107 |
| March | 70,128 |
| April | 77,182 |
| May | 55,153 |
| June | 62,371 |
| July | 47,773 |
Read naively, this table tells a story of collapsing company creation through 2026. INSEE measures the opposite: creations rebounded strongly in May before a mild dip in June. The downward slope in a registry snapshot is a measurement artifact, not an economic fact. A recent month is structurally under-represented, for three cumulative reasons.
The three mechanisms that hide a recent creation
1. Non-disclosure rights. Some entrepreneurs — mostly individuals, micro-entrepreneurs first — exercise their right to keep their data out of public extracts. Those units exist in the registry and INSEE counts them, but they never appear in open data. In our measurement, only 39% of January's visible creations are individuals, while official publications put micro-entrepreneurs at the majority of all creations: that's where the missing mass is concentrated.
2. Registration lag. A company founded in late July may only be registered in August or September. The most recent month is always incomplete at snapshot time — which is why January (95,516) looks twice as "productive" as July (47,773): it simply had seven months to fill in.
3. Early closures. Our perimeter only keeps units still active. A company created in January and already closed by June has left the count, even though INSEE recorded it as a creation.
Order of magnitude of the combined effect: comparing our visible monthly average (~67,600) with the official pace (above 90,000 in recent trend), roughly a quarter to a third of recent creations do not appear in a public registry extract. The comparison mixes a raw count (ours) with seasonally adjusted series (INSEE's): it bounds the order of magnitude, it doesn't give a decimal.
Which figure counts what
| Figure | What it counts | Where it lives |
|---|---|---|
| 1,170,000 (2025) | Every creation registered in the year, including non-disclosed units and companies that have since closed | INSEE publications |
| 473,230 (Jan–Jul 2026) | 2026 creations still active AND publicly listed at snapshot date | Open-data registry extract |
| 26,862,604 | All active, publicly listed legal units, whatever their creation year | Open-data registry extract |
It's the same family of gaps we documented for the stock of establishments in our French B2B prospecting barometer: "36 million establishments" (historical cumulative), "14.3 million" (active and listed) and "5.9 million" (economically active) are three different perimeters — all correct.
What this means for a prospecting file
If you target recently created French companies, two practical consequences. First, a "last month's creations" export will always look thin: it fills up over the following weeks as registrations come through — re-running the same export later returns additional rows for the same creation month. Second, micro-entrepreneurs are structurally under-represented: a "new companies" file built from open data is de facto tilted toward incorporated companies.
It's also why a list should be verified at send time, not at export time: some rows have moved in between. Crossing your list with up-to-date legal records for each company or re-enriching SIRET numbers before a campaign avoids contacting structures that were already struck off — or missing the ones that just appeared. To build the list itself, a free outsend alpha access lets you start from the live registry rather than a frozen export.
Frequently asked questions
How many companies were created in France in 2026?
The definitive figure doesn't exist yet. INSEE publishes a monthly report; the trend as of June 30, 2026 is +11.1% over twelve rolling months, after a record 2025 at 1,170,000 creations. In an open-data extract of the Sirene registry taken on August 13, 2026, 473,230 legal units created between January and July 2026 are visible — a subset: active and publicly listed only.
Why doesn't my Sirene export match INSEE's numbers?
Because a public extract excludes three populations INSEE counts: entrepreneurs who opted out of data disclosure (frequent among individuals), creations not yet registered at export time, and companies already closed since their creation.
Is French company creation slowing down in 2026?
No. The downward slope seen in a registry snapshot (95,516 visible creations for January versus 47,773 for July) is an artifact: recent months aren't complete yet. INSEE's seasonally adjusted series show growth over twelve months.
What share of French creations are micro-entrepreneurs?
In INSEE publications, micro-entrepreneur registrations make up the majority of creations. In a public extract their apparent share is far lower — 39% of January 2026's visible creations are individuals in our measurement — because non-disclosure is concentrated on that status. Not a contradiction: a visibility bias worth knowing.