Series context: This is the 13th article in ManThink's "LoRaWAN Technical Reading" series (3rd in the Industry Trends track). It builds on the previous 12 articles and forms a coherent "macro → policy → supply chain" narrative with Article 11 ("LPWAN Market Landscape 2026") and Article 12 ("China CN470 Spectrum Rules Explained").
Sources: LoRa Alliance public reports; ChirpStack / The Things Stack GitHub repository statistics (August 2026); RISC-V International membership data; SupplyICs 2026 mid-year RISC-V Industrial IoT readiness report; Edgepilot and AESTECHNO public TCO analyses; Semtech SX1302/1303 datasheets; ThinkLink GD6 open-source repository (github.com/ManThink/gd6-open).
Disclaimer: This article synthesizes publicly available industry data and technical documentation. All prices, GitHub metrics, and community-size figures are timestamped. Procurement decisions should rely on the latest vendor quotes and local distributor pricing.
In August 2026, ManThink posted a short note on LinkedIn:
What does US$44 buy in a LoRaWAN gateway? With the GD6 base version, it buys more than a closed box: • Hardware design files and a versioned BOM • ESP-IDF firmware and Web console source • Ethernet and Wi-Fi backhaul • Standard connections to ThinkLink, ChirpStack, and The Things Stack.
What does $44 buy? An 8-channel indoor LoRaWAN gateway based on ESP32-S3 + Semtech SX1302, plus hardware design files, a versioned BOM, ESP-IDF firmware, the Web console source code, and standard connectivity to ChirpStack / The Things Stack / ThinkLink. In 2018, that price wouldn't even cover a single SX1301 concentrator chip.
Zooming out: in 2015, an 8-channel commercial LoRaWAN gateway retailed for $1,200-$2,500 (Kerlink iFemtoCell, Browan Femto, Multitech Conduit). By 2026, the entry-level price for comparable hardware has dropped to $44-$150, with open-source firmware and open-source NS as the default option. Ten years of price compression to 1/20 of the original cost, with community-driven development replacing vendor lock-in — this isn't a single company's price war, but a maturing infrastructure supply chain.
This article answers three questions:
Mapping LoRaWAN gateway evolution over the past decade into three waves clarifies why this is happening now.
Representative products: Kerlink iFemtoCell, Browan Femto, Multitech Conduit, Ubisense.
Characteristics: - 8-channel SX1301 concentrator + ARM SoC (typically Freescale i.MX or Marvell 88W8997); - Closed-source firmware, closed-source Web UI, configuration options locked by the vendor; - Unit price: $1,200-$2,500; - Sold to operators and enterprise integrators; primary battlegrounds were smart-city pilots and meter-reading projects.
This wave solved the problem of "taking LoRaWAN out of the lab into the field," but didn't solve "how to change the configuration, manage the devices, or upgrade the firmware" — three tasks that integrators had to script themselves every time.
Representative products: RAK2245/RAK2287 Pi HAT + Raspberry Pi, Dragino PG1302, IC880A. On the NS side: loraserver's evolution into ChirpStack (in 2018 Orne Brocaar renamed the project from LoRa Server to ChirpStack). In parallel, The Things Network community surpassed 100,000 gateways.
Characteristics:
- Hardware went modular: SX130x HAT + general-purpose SBC (Pi, MT7628 SoC);
- Configuration carried by Packet Forwarder's global_conf.json + community scripts;
- Mature open-source NS options emerged (ChirpStack, TTN Stack);
- Price range dropped to $100-$300;
- Deployment depended on "DIY engineers + forum help + GitHub issues."
This wave opened the door to "open source," but fragmented the user experience — the same gateway model configured differently in two integrators' hands; the same NS running different plugin versions in two projects. Production-environment pain was still borne by integrators themselves.
Representative products: ManThink GD6, RAK WisGate Edge Lite 2 (partially open-source firmware), Dragino LPS8N (OpenWrt-based), TTIG. On the NS side: ChirpStack v4 rewritten in Rust with full containerization; The Things Stack commercial and community editions running in parallel; ThinkLink embedding NS directly into gateway hardware.
Characteristics: - Hardware open-source (design files + BOM + firmware + Web UI all released); - ESP32-S3 / RISC-V SoC running Web console + TLS + MQTT is no longer a stretch; - 8 MB PSRAM + 8 MB Flash lets "out-of-the-box + deeply customizable" coexist; - Price range dropped to $44-$150 for entry-level, $200-$400 for outdoor models; - NS expanded from "standalone deployment" to "embedded in the gateway" (ThinkLink in-gateway NS).
The key turning point of this wave was 2023-2024: the maturation of ESP32-S3 + 8 MB PSRAM combined with ChirpStack v4's Rust rewrite. The former compressed the runtime cost of Web console + TLS + MQTT to a sufficiently low level; the latter compressed NS deployment difficulty to "spin it up with one docker compose."
Open-source gateways aren't something a company can "decide" to build. Their viability requires three foundational variables to converge. By mid-2026, all three have finally crossed the "industrial-grade" threshold simultaneously.
In 2018, the default open-source LoRaWAN gateway build was Raspberry Pi 3B + SX1301 HAT. The problems:
The 2026 options are completely different:
| MCU | Architecture | Clock | RAM | Flash | Industrial Temp | Price Tier |
|---|---|---|---|---|---|---|
| ESP32-S3-WROOM-1 (N8R8) | Xtensa LX7 dual-core | 240 MHz | 8 MB PSRAM | 8 MB QIO | -40 ~ 85°C | $2.50-$3.50 |
| ESP32-C6-WROOM-1 | RISC-V single-core | 160 MHz | 512 KB SRAM + 4 MB Flash | 4 MB | -40 ~ 85°C | $3.00-$4.50 |
| ESP32-C3 | RISC-V single-core | 160 MHz | 400 KB SRAM + 4 MB Flash | 4 MB | -10 ~ 70°C | $1.80-$2.50 |
| Raspberry Pi CM4 (RAK7394) | ARM Cortex-A72 quad-core | 1.5 GHz | 1-8 GB | 8-32 GB eMMC | 0-50°C (industrial -25-80°C) | $45-$75 |
| MT7628 (RAK7268V2 / Dragino LPS8N) | MIPS 24KEc | 580 MHz | 64 MB DDR2 | 16 MB | -10 ~ 70°C | $5-$8 |
The critical metric is: "stable concurrent execution of Web/TLS/MQTT/Packet Forwarder over an 8 MHz SPI bus". ESP32-S3 is the first solution to hit this at the $3 price tier. GD6, Heltec HT-M2802, and other open-source gateways all use it.
RISC-V's industrial readiness also formally crossed the line in 2026. SupplyICs' mid-2026 assessment report cites three quantitative milestones:
The report's direct conclusion: for non-safety-critical industrial IoT applications (smart sensors, environmental monitoring, edge AI, predictive maintenance, wireless connectivity gateways), RISC-V is ready for volume procurement at 50K+ units/year in mid-2026. From an engineering perspective, this means open-source gateways no longer carry the tail risk of "MCU supply discontinuity" within a 5-10 year lifecycle.
If hardware determines "can the gateway be cheap," open-source NS determines "can the network avoid vendor lock-in."
Two de facto standards in 2026:
The following comparison table comes from pistack.xyz's May 2026 review (data cross-verified with official repositories):
| Capability | ChirpStack | The Things Stack | lorawan-server (lightweight) |
|---|---|---|---|
| License | MIT | Apache-2.0 | Apache-2.0 |
| Implementation | Go + Rust (ChirpStack v4) | Go + Erlang | Erlang |
| Containerization | Official docker compose | Official docker compose | Community images |
| Multi-tenancy | Organizations | Clusters | Single-tenant |
| Web Console | Full-featured | Full-featured | Lightweight |
| MQTT integration | Built-in | Built-in | Built-in |
| HTTP integration | Webhooks | Webhooks | Webhooks |
| LoRaWAN version | 1.0 / 1.1 | 1.0 / 1.1 | 1.0 / 1.1 |
| Class | A / B / C | A / B / C | A / C |
| FUOTA | Built-in | Via plugin | No |
| Multi-region | Region profiles | Frequency plans | Region config |
| GitHub stars (Aug 2026) | ~1.1K | ~1.2K | ~1.0K |
The Things Network community (TheThingsNetwork) operating metrics further demonstrate scale:
Source: The Things Network official site, 2026 public data.
These numbers aren't evidence of technical superiority — they're evidence of ecosystem maturity, meaning integrators who hire, troubleshoot, or look up reference designs will find community members who have likely encountered the same pitfalls.
The most painful part of building open-source LoRaWAN gateways in 2018 wasn't technology — it was "how to navigate different countries' compliance." FCC / CE / MIC / NCC / MIIT each required separate pre-research, with R&D costs routinely reaching millions of dollars.
This path has simplified in 2025-2026, for three reasons:
GD6 goes further in this wave — it open-sources hardware design files, versioned BOM, ESP-IDF firmware, and Web console together under MIT. Integrators can clone the entire design to obtain their own model certification, or directly purchase GD6 whole devices and use ManThink's pre-tested RF chain. This path was previously available only to large vendors.
Putting the top four "open-source-friendly" 8-channel indoor LoRaWAN gateways side by side in 2026 reveals the positioning differences:
| Dimension | ManThink GD6 | RAK WisGate Edge Lite 2 (RAK7268V2) | Dragino LPS8N | TTIG (The Things Indoor Gateway) |
|---|---|---|---|---|
| MCU | ESP32-S3 dual-core 240 MHz | MT7628 MIPS 580 MHz | MT7628 MIPS 580 MHz | ESP32 dual-core 240 MHz |
| LoRa chip | SX1301 / SX1302 | SX1302 | SX1302 | SX1302 |
| Channels | 8 freq / 65 sub-channels | 8 freq | 8 freq | 8 freq |
| Bands (per model) | CN470/EU433 (GD614), EU868/AS923/AU915/US902 (GD618) | EU868/US915/AS923/AU915/IN865/CN470 (per SKU) | EU868/US915/AS923/AU915/CN470 (per SKU) | EU868/US915 (per SKU) |
| Backhaul | Ethernet + Wi-Fi + 4G (optional) | Ethernet + Wi-Fi | Ethernet + Wi-Fi | Wi-Fi only |
| Protocol stack | Semtech UDP / MQTT Bridge / Basic Station | Semtech UDP / Basic Station | Semtech UDP / Basic Station | Semtech UDP |
| OS | ESP-IDF (FreeRTOS) | OpenWrt | OpenWrt | ESP-IDF |
| Web UI | Custom bilingual, open source | WisGateOS (partially open) | OpenWrt LuCI | Minimal config (TTN only) |
| Network server | Any (supports ThinkLink embedded NS) | Any | Any | TTN/TTI only |
| REST API | ✅ Complete | Limited | Limited | No |
| Remote MQTT config | ✅ Dedicated channel | No | No | No |
| OTA | A/B dual partition | Single partition | Single partition | Mandatory cloud |
| Hardware design files open | ✅ (KiCad source + schematics) | ❌ | ❌ | ❌ |
| Firmware source | ✅ ESP-IDF + Web | Partial | Partial (OpenWrt packages) | ❌ |
| Price (2026-08 entry) | $44 (base, no 4G/PoE) | ~$135 | ~$115 | ~$90 |
| Ideal audience | Developers, integrators, education, customization | Mid-scale commercial | Budget-conscious projects | TTN community users |
Note: TTIG is strictly speaking not "open-source hardware," but it represents the opposite path from local open-source — $90 price, extremely simple, but fully bound to The Things Network ecosystem.
The differences between the four products boil down to "freedom" vs "ease of use" trade-offs:
"Open-source gateways are cheap" is an imprecise statement. Open-source gateways have cheaper hardware unit prices, but the engineering time cost isn't cheap. Edgepilot's 2026 TCO analysis provides a conservative estimate (5-year lifecycle, per gateway, in EUR):
| Cost Item | Self-Managed (including open-source hardware) | Managed Service |
|---|---|---|
| Hardware | €300-€600 | Included |
| 4G SIM / data plan | €60-€120/year (5 years: €300-€600) | Included |
| Configuration time | 2-4 hours × engineer rate | Included |
| Monitoring infrastructure | Build or license (€50-€200/year) | Included |
| Firmware updates | Build OTA system or site visits (€200-€500 each) | Included |
| Troubleshooting / support | Internal engineering time | Included |
| Hardware replacement | At your cost | Included |
| 5-year TCO (per unit) | €1,500-€3,500+ | €960-€1,310 |
Source: Edgepilot, "The True Cost of Running LoRaWAN Gateways: Build vs Buy vs Managed" (2026 public analysis).
Edgepilot's core conclusion: for a single gateway over 5 years, self-managed TCO is typically 30%-150% higher than managed service, with the gap coming not from hardware but from engineering time and operations tooling.
But this conclusion has a scope — it assumes only 1-50 gateways. Once volume rises above 100+ units, the open-source TCO advantage reverses. AESTECHNO's 2026 study provides a 1,000-node 5-year TCO comparison (EUR):
| Item | Private LoRaWAN | Operator NB-IoT | Sigfox (decommissioned) |
|---|---|---|---|
| Radio modules | €6,000 | €9,000 | €3,000 |
| Gateways + NS | 10 × €800 = €8,000 | €0 (operator) | €0 |
| Network server | €2,000/year × 5 = €10,000 (or free with open-source ChirpStack) | €0 | €0 |
| Subscriptions | €0 | €10/node/year × 5 years × 1,000 = €50,000 | €7/node/year × 2 years × 1,000 = €14,000 |
| Forced migration | €0 | €0 | €15,000 |
| 5-year TCO | €24,000 | €59,000 | €32,000 + migration |
| Per node per year | €4.80 | €11.80 | €6.40 + migration amortized |
Source: AESTECHNO, "LPWAN networks compared: LoRaWAN, NB-IoT, Sigfox" (2026).
Critical crossover point: over a 5-year lifecycle, the break-even between private LoRaWAN and operator NB-IoT sits at 500-800 nodes — below that scale, NB-IoT has lower marginal cost (because there's no infrastructure to build); above that scale, self-built LoRaWAN is significantly cheaper (because cumulative subscriptions exceed infrastructure investment).
Combining these two independent studies yields a deployment-decision table:
| Node Scale | Single Gateway Coverage | Recommended Approach | 5-Year Savings |
|---|---|---|---|
| < 50 nodes / single gateway | Building / campus pilot | Managed LoRaWAN service or NB-IoT | Avoid self-build operations |
| 50-500 nodes / 2-10 gateways | Campus / factory | Open-source LoRaWAN (GD6/RAK) + self-hosted ChirpStack | 40%-60% vs NB-IoT |
| 500-5,000 nodes / 10-100 gateways | City / industrial park | Open-source LoRaWAN + commercial NS (ThinkLink / TTN Cloud) | 60%-70% vs NB-IoT |
| > 5,000 nodes / > 100 gateways | Large city / nationwide | Open-source LoRaWAN + embedded NS + dedicated ops team | 70%+ vs NB-IoT |
Key takeaway for integrators: the real TCO advantage of open-source gateways only materializes from "50+ nodes" upward; below 10 nodes, managed services are actually more economical.
Open-source LoRaWAN gateway engineering teams frequently underestimate three "invisible costs":
Closed-source gateways have vendors responsible for security patches; open-source gateways have integrators responsible.
Budget recommendation: reserve at least 0.3 FTE engineer per project for firmware tracking, regression testing, and security update releases. Equivalent to roughly €30,000-€60,000 over a 5-year lifecycle.
Open-source hardware can be "cloned by design," but whole-device model certification cannot be reused — model certification is tied to a specific vendor + specific model + specific RF chain.
Budget recommendation: at least €10,000-€30,000 one-time certification cost per target market + annual spot-check maintenance. This is especially critical for integrators planning multi-country deployments.
One gateway can be configured via Web UI; 10 can use REST API; 100 need a dedicated gateway fleet management system:
Budget recommendation: building this toolchain in-house requires 0.5-1 FTE frontend + backend engineers for 6-12 months, equivalent to €50,000-€120,000. Purchasing managed platforms (ThinkLink / ChirpStack commercial / The Things Stack Cloud) reduces this to €0-€20,000/year.
Pulling GD6 out for a separate look is because in 2026 it took a "full-stack open-source but not GitHub-hosted" middle path — a position distinct from RAK WisGate, Dragino LPS8N, and TTIG.
What GD6 open-sources (as of 2026-08):
What GD6 does not open-source:
This open-source boundary is intentionally designed as "hardware and gateway layer fully open to developers, network layer delegated to existing open-source/commercial NS" — avoiding direct competition with ChirpStack / ThinkLink and other NS vendors.
| Dimension | GD6 | RAK7268V2 | Dragino LPS8N | TTIG |
|---|---|---|---|---|
| Entry price | $44 | $135 | $115 | $90 |
| All hardware design files open | ✅ | ❌ | ❌ | ❌ |
| All firmware source | ✅ | Partial | Partial | ❌ |
| Web console source | ✅ | ❌ | ❌ | ❌ |
| REST API | ✅ | Limited | Limited | ❌ |
| Remote MQTT config channel | ✅ | ❌ | ❌ | ❌ |
| Embedded NS capability | Via ThinkLink | Via RAK WisGateOS | No | No |
| OTA | A/B dual partition | Single partition | Single partition | Mandatory cloud |
| Default compliance params | CN470 calibrated per MIIT Announcement 52 | Manual config | Manual config | TTN only |
GD6 CN470 default parameter details: 50 mW ERP (17 dBm ERP ≈ 19.15 dBm EIRP), default disable of channels 6-38 and 45-77 (China State Grid channels), default enable of CAD channel detection, default enable of LDRO, default enable of CFList. See Article 12, "China CN470 Spectrum Rules Explained," for details.
GD6's choice of A/B dual-partition OTA is a critical engineering trade-off for industrial deployments — upgrade interruption or failure won't brick the device. For multi-gateway fleet operations, this is a qualitative change.
GD6's choice of a dedicated MQTT channel for remote configuration is physically isolated from the data-forwarding MQTT Bridge — preventing the remote config and data channels from blocking each other, and preventing the remote config channel from being misused to modify the Broker itself.
GD6's choice of default compliance parameters per MIIT Announcement 52 is an engineering "one step ahead for the customer" — aligned with the compliance boundaries from Article 12.
For fairness, GD6 isn't suitable for these scenarios:
The "last mile" of open-source gateways in production environments is an engineering challenge. Three issues that cannot be avoided:
Open-source ESP-IDF + FreeRTOS gateways have dramatically improved 24/7 stability compared to 2018, but still lag behind closed-source commercial firmware (which has undergone months of vendor long-term stability testing).
Engineering practices:
The compliance paths for the same gateway model in different countries vary:
Engineering practices:
ChirpStack / TTN Stack GitHub issue response time is typically 1-7 days, with severity affecting priority; commercial support (The Things Industries, Orne Brocaar commercial edition) provides SLA response (typically within 24 hours).
Engineering practices:
Returning to the three opening questions:
1. Why are gateways collectively going open-source in 2025-2026?
Because three foundational variables have matured simultaneously: ESP32-S3 + 8 MB PSRAM makes "$3 MCU running Web + TLS + MQTT" reality; RISC-V passed industrial IoT volume verification in mid-2026; ChirpStack v4's Rust rewrite brought NS to the "one docker compose" state. None of these three were in place in 2020.
2. Is open-source LoRaWAN gateway 5-year TCO actually lower than closed-source? By how much?
For single gateways, no (closed-source managed services are actually 30%-150% cheaper); at 50+ units it breaks even; above 500 units it's significantly lower than closed-source (40%-70% savings vs NB-IoT). The critical scale is 50 units / 500 nodes.
3. "Open source ≠ free" — what hidden costs must enterprise deployments budget for?
Firmware maintenance (0.3 FTE), compliance certification (€10K-€30K / market), gateway fleet management system (self-built €50K-€120K or managed €0-€20K/year). These three aren't on the hardware BOM, but they must be on the project budget.
Final judgment:
Open-source LoRaWAN gateways crossed the "usable" phase in 2026 and entered the "production-ready" phase. But they aren't an endgame — they're a new entry ticket. What determines project success isn't "open source vs closed source" but "whether the team has the ability to continuously invest engineering time."
Teams that can do this can, starting in 2026, build private LoRaWAN networks with $44 GD6s, $135 RAKs, or $115 Draginos achieving 5-year TCO 40%-70% lower than operator NB-IoT.
Teams that can't do this should still choose closed-source managed services or operator NB-IoT — outsourcing operations to specialists, focusing on data and application layers.
Open source is an entry ticket, not an endgame — this statement holds true for individual developers, integrators, and ManThink itself.
| Term | Full Name | Meaning |
|---|---|---|
| NS | Network Server | LoRaWAN network server, handles device authentication, deduplication, ADR, forwarding |
| AS | Application Server | Application server, handles payload decoding, application-layer encryption, forwarding to business systems |
| LNS | LoRaWAN Network Server | Same as NS; terminology used in LoRa Basics Station protocol context |
| CUPS | Configuration and Update Server | Semtech Basic Station's server-side configuration protocol |
| FUOTA | Firmware Update Over The Air | LoRaWAN remote firmware upgrade protocol |
| TCO | Total Cost of Ownership | Total cost of ownership, including hardware, software, labor, operations, compliance |
| PSRAM | Pseudo Static RAM | Pseudo-static RAM, provides additional large memory buffer for MCU |
| FOTA | Firmware Over The Air | Remote firmware upgrade (different from FUOTA, which is a LoRaWAN standard protocol) |
| LTS | Long Term Support | Long-term maintenance version |
| RISC-V | Reduced Instruction Set Computing - V | Open instruction set architecture |
Copyright notice: This article is an original ManThink technical analysis, released under CC-BY 4.0. When citing third-party data, please retain the source link.