Report · October 2026
State of Open Hardware on GitHub 2026
We read the KiCad files of 15,056 public GitHub repositories — 29,305 boards from 10,477 designers — and counted what open hardware is built from: which microcontrollers, which parts, how many layers, under which licence. Three things changed more than anything else.
1. Microcontrollers: ESP32 took over
Of the 8,868 projects where we could tell which microcontroller a board is built on, the picture by year is a changing of the guard. In repos created in 2016–2019, about one in ten used an ESP32 and a third used an AVR — the chip behind the classic Arduino. Among projects started in 2026, ESP32 is at 40% and AVR has fallen from 37% (2020) to 10%. STM32 is the constant: between a fifth and a third of projects, every year.
RP2040: from zero to one project in five
Raspberry Pi's RP2040 came out in January 2021. In repos created before then it shows up in at most 4.4% (projects that were reworked later), and from 2022 it has held 20%–24% of new projects every year. Its successor is arriving: 176 repos already use an RP2350 or a Pico 2.
Within ESP32 projects the S3 is now the default: 43% of 2026 ESP32 repos use one, up from 23% in 2022, while the RISC-V C3 and C6 take about a quarter.
- ESP32-S3
- ESP32-C3 / C6
Show data
| Year | ESP32-S3 | ESP32-C3 / C6 | Repos |
|---|---|---|---|
| 2022 | 23% | 16% | 119 |
| 2023 | 27% | 13% | 188 |
| 2024 | 31% | 18% | 280 |
| 2025 | 37% | 23% | 430 |
| 2026* | 43% | 25% | 1,166 |
- ESP3228%
- STM3222%
- AVR/Arduino20%
- RP2040/RP235017%
- Raspberry Pi3.6%
- ESP82662.9%
- NXP2.8%
- Nordic nRF2.8%
- FPGA2.5%
- PIC2.1%
Show data
| Item | Share | Repos |
|---|---|---|
| ESP32 | 28% | 2,459 |
| STM32 | 22% | 1,964 |
| AVR/Arduino | 20% | 1,745 |
| RP2040/RP2350 | 17% | 1,536 |
| Raspberry Pi | 3.6% | 321 |
| ESP8266 | 2.9% | 261 |
| NXP | 2.8% | 251 |
| Nordic nRF | 2.8% | 245 |
| FPGA | 2.5% | 224 |
| PIC | 2.1% | 191 |
2. The parts everyone reaches for
Some choices are close to unanimous. 91% of projects with SPI flash use a Winbond W25Q, and 75% of those that protect their USB lines do it with a USBLC6. The AMS1117 is in 35% of projects with a linear regulator, ahead of the old 78xx series. Elsewhere it is a race: CH340 (42%) leads CP2102 as the USB-to-UART bridge, and three Li-ion chargers split the field.
Show data
| Part | Share | Repos |
|---|---|---|
| AMS1117 | 35% | 1,170 |
| 78xx | 20% | 663 |
| AP2112 | 15% | 496 |
| XC6206 | 8.8% | 297 |
| LM1117 | 8.6% | 293 |
- CH34042%
- CP2102 / CP210434%
- FT23214%
- FT2317.8%
- CH3433.4%
Show data
| Part | Share | Repos |
|---|---|---|
| CH340 | 42% | 421 |
| CP2102 / CP2104 | 34% | 345 |
| FT232 | 14% | 137 |
| FT231 | 7.8% | 78 |
| CH343 | 3.4% | 34 |
- USBLC675%
- TPD2E / TPD4E13%
- SRV0510%
- PRTR5V0U2X3.8%
Show data
| Part | Share | Repos |
|---|---|---|
| USBLC6 | 75% | 1,071 |
| TPD2E / TPD4E | 13% | 193 |
| SRV05 | 10% | 150 |
| PRTR5V0U2X | 3.8% | 54 |
- BQ24 / BQ2536%
- MCP7383135%
- TP405626%
- TP40543.2%
- IP53061.4%
Show data
| Part | Share | Repos |
|---|---|---|
| BQ24 / BQ25 | 36% | 394 |
| MCP73831 | 35% | 373 |
| TP4056 | 26% | 277 |
| TP4054 | 3.2% | 35 |
| IP5306 | 1.4% | 15 |
- Winbond W25Q91%
- GigaDevice GD25Q4.0%
- AT253.9%
- ISSI IS253.1%
Show data
| Part | Share | Repos |
|---|---|---|
| Winbond W25Q | 91% | 1,086 |
| GigaDevice GD25Q | 4.0% | 48 |
| AT25 | 3.9% | 46 |
| ISSI IS25 | 3.1% | 37 |
- MPU-6050 / 925031%
- LSM6DS31%
- BMI23%
- ICM-4268815%
- ICM-209484.8%
Show data
| Part | Share | Repos |
|---|---|---|
| MPU-6050 / 9250 | 31% | 188 |
| LSM6DS | 31% | 187 |
| BMI | 23% | 138 |
| ICM-42688 | 15% | 90 |
| ICM-20948 | 4.8% | 29 |
- BME280 / BMP28057%
- SHT3x / SHT4x24%
- DHT12%
- AHT20 / AHT215.9%
- BME680 / BME6885.1%
Show data
| Part | Share | Repos |
|---|---|---|
| BME280 / BMP280 | 57% | 215 |
| SHT3x / SHT4x | 24% | 91 |
| DHT | 12% | 45 |
| AHT20 / AHT21 | 5.9% | 22 |
| BME680 / BME688 | 5.1% | 19 |
Each part links to the open designs that use it. Download all part families (CSV)
3. What people build
Power electronics, IoT nodes and sensor boards lead, followed by development boards, displays and mechanical keyboards. Retro computing is a small but stubborn corner: hundreds of boards for machines from the 1980s.
- Power & battery14%
- IoT & wireless14%
- Sensors12%
- Dev boards & breakouts11%
- Displays & LEDs11%
- Keyboards9.9%
- Audio9.0%
- Robotics & motion8.0%
- RF & radio6.7%
- Retro computing4.3%
- USB & debug tools4.1%
- Test & measurement3.0%
- Wearables2.4%
- Science & space0.6%
Show data
| Item | Share | Repos |
|---|---|---|
| Power & battery | 14% | 2,125 |
| IoT & wireless | 14% | 2,075 |
| Sensors | 12% | 1,785 |
| Dev boards & breakouts | 11% | 1,720 |
| Displays & LEDs | 11% | 1,700 |
| Keyboards | 9.9% | 1,486 |
| Audio | 9.0% | 1,358 |
| Robotics & motion | 8.0% | 1,198 |
| RF & radio | 6.7% | 1,001 |
| Retro computing | 4.3% | 654 |
| USB & debug tools | 4.1% | 612 |
| Test & measurement | 3.0% | 455 |
| Wearables | 2.4% | 355 |
| Science & space | 0.6% | 92 |
4. Board anatomy
Four-layer boards used to be for the ambitious. They are now routine: 31% of projects started in 2024–2026 are on four or more copper layers, against 18% in 2016–2020, as four-layer prototypes got cheap.
Show data
| Year | 4+ layers | Repos |
|---|---|---|
| 2016 | 20% | 230 |
| 2017 | 17% | 345 |
| 2018 | 18% | 459 |
| 2019 | 19% | 649 |
| 2020 | 18% | 859 |
| 2021 | 23% | 1,007 |
| 2022 | 21% | 1,118 |
| 2023 | 26% | 1,256 |
| 2024 | 32% | 1,755 |
| 2025 | 31% | 2,258 |
| 2026* | 31% | 4,837 |
5. Half of open hardware has no licence
Publishing a design on GitHub is not the same as letting others use it: without a licence, the default is all rights reserved. 49% of the repos have no licence GitHub detects (some state one only in a README, so the true figure is a little lower). Those that do choose overwhelmingly reach for software licences — MIT and GPL-3.0 — rather than the CERN Open Hardware Licence written for exactly this.
- MIT37%
- GPL-3.019%
- CERN-OHL-S-2.05.9%
- Apache-2.05.3%
- Other (unrecognised)4.5%
- CC-BY-SA-4.04.4%
- CC-BY-NC-4.04.2%
- CERN-OHL-P-2.04.0%
- CC0-1.02.0%
- CERN-OHL-W-2.01.9%
Show data
| Item | Share |
|---|---|
| MIT | 37% |
| GPL-3.0 | 19% |
| CERN-OHL-S-2.0 | 5.9% |
| Apache-2.0 | 5.3% |
| Other (unrecognised) | 4.5% |
| CC-BY-SA-4.0 | 4.4% |
| CC-BY-NC-4.0 | 4.2% |
| CERN-OHL-P-2.0 | 4.0% |
| CC0-1.0 | 2.0% |
| CERN-OHL-W-2.0 | 1.9% |
6. The JLCPCB effect
LCSC part numbers are how JLCPCB's assembly service knows what to place, so a BOM that carries them was made to be assembled there. Before 2020 almost nobody bothered: 2.5% of projects. By 2021 it was 12%, and 22% of projects started in 2025 carry them. Manufacturer part numbers rose alongside — about half of new projects now name the exact part, not just "10k".
- Manufacturer part numbers
- LCSC part numbers
Show data
| Year | Manufacturer part numbers | LCSC part numbers | Repos |
|---|---|---|---|
| 2016 | 30% | 2.6% | 230 |
| 2017 | 20% | 1.7% | 345 |
| 2018 | 22% | 2.8% | 459 |
| 2019 | 26% | 2.6% | 649 |
| 2020 | 28% | 4.7% | 859 |
| 2021 | 34% | 12% | 1,007 |
| 2022 | 38% | 17% | 1,118 |
| 2023 | 47% | 20% | 1,256 |
| 2024 | 49% | 20% | 1,755 |
| 2025 | 51% | 22% | 2,258 |
| 2026* | 52% | 22% | 4,837 |
How we counted
- The source is the HardLabs board gallery: public GitHub repositories with KiCad files (a layout, schematic or netlist), found through GitHub search and read on 2026-10-10. Designs made in other tools (Altium, Eagle, EasyEDA) and private repos are not in it.
- Everything is counted per repository: a repo with a dozen revisions of one board counts once. A repo counts once for each microcontroller family or part any of its boards uses.
- A project's year is the year its GitHub repo was created, not when the board was designed. Because the gallery searches recently updated repos, recent years — 2026 most of all, which is also partial (*) — hold more repos than earlier ones. That is why the charts compare shares within a year, not counts.
- The microcontroller is recognised from part names on the board; we could tell it for 59% of repos (many boards have none: power supplies, breakouts, keyboards' daughterboards).
- Licences are the ones GitHub detects for the repo; LCSC and manufacturer part numbers come from the footprint, symbol and BOM fields. Both are lower bounds.
Use this data
The figures are free to use under CC BY 4.0 — for articles, talks, theses or your own analysis. Please credit “HardLabs, State of Open Hardware on GitHub 2026” with a link to this page. Every chart above has a PNG download and a link of its own.
- mcu-share-by-year.csv — Microcontroller family shares by year
- mcu-overall.csv — Microcontroller families across all repos
- part-families.csv — Parts within each family
- categories.csv — Project categories
- layers.csv — Copper layer counts
- licences.csv — Licences
- by-year.csv — Layers, licences and part numbers by year
Every project behind these numbers is in the gallery, with its KiCad files, parts list and a render — by chip, by part and by what it is for. And HardLabs can turn one into a working simulation, so you can run its firmware before you build it.
Open the board gallery