Between 2019 and 2021, IoT Analytics tracked 188 IoT cloud platforms shutting down, with 26% simply going out of business. In the last four months, Particle was acquired by Digi (January 2026), Losant by SUSE (February 2026), and PTC ThingWorx with Kepware went to TPG (March 2026 close, announced November 2025). Platform mortality is happening inside a rising market (Gartner’s $44.5B forecast by 2026), not a collapsing one. For founders turning devices into revenue, the question is no longer “custom or low-code?” – it is which vendor survives the decade.
Before the vendor question, the revenue question.
Three Monetization Models With Real Numbers Behind Them
Three disclosed models cover most of the startup-stage playing field, and the one that looks most like software is usually the right call.
Samsara’s FY26 10-K reports $1.9B ARR at 30% year-over-year growth, with 3,194 customers paying more than $100K ARR, generating $1.2B (61% of total ARR) on three-to-five-year per-asset SaaS contracts. Whoop crossed $1B in bookings in 2025 with 2.5M members on a subscription-only model where hardware sits inside the $199-$359 yearly membership, and turned cash-flow positive the same year.7 Oura posted $1B in revenue in 2025 (2x year-over-year) across 5.5M rings, with the disclosed 2024 split at roughly 80% hardware and 20% subscription – the consumer-hardware shape, not the software shape.
Usage-based pricing is common across software suppliers (Revenera’s 2026 Monetization Monitor surveyed 501 suppliers and found 74% using usage-based models at least moderately), but outcomes-based pricing specifically has retreated from 60% to 38% adoption year over year. John Deere’s See & Spray is the rare case where the outcomes shape actually works: $1 per fallow acre, $5 per in-crop acre across 5M+ acres in 2025, because the physical outcome is deterministic and the counterfactual is measurable on equipment the farmer already owns.
Then the reality check, from the Michelin Fleet Solutions tires-as-a-service case: after 3 years, expansion was far below expectations and profitability was terrible.11 Outcomes pricing rewards deterministic physical outcomes and punishes everything else.
Why Low-Code Wins at the MVP Stage
A traditional IoT MVP runs three to four months with five-to-ten engineers at roughly $90K-$150K each, while a low-code IoT platform collapses the cycle to around 30 days with a two-to-four-person team. On a seed-stage runway, that delta is the difference between shipping a hypothesis test and burning a pre-series hire round before product-market fit lands.
Adoption is no longer in question. Gartner (Wong and Davis, July 2022, document 4350499) puts 70% of new applications developed by enterprises on low-code or no-code by 2025, up from less than 25% in 2020 – a shift driven in large part by engineer-supply constraints that hit IoT startups especially hard.
The speed claim has limits that must be declared aloud, because it assumes off-the-shelf hardware, no custom firmware, and no regulated workload – protocol-level control, sub-100ms latency, or compliance-gated data flows break the 30-day number.
Rule of thumb: low-code wins when the hypothesis is unproven and the hardware is commodity. Custom wins when the hypothesis is proven and the hardware is the moat.
The Vendor Is the Risk, Not the Category
Category adoption is one story, and individual vendor survival is a different one that startups systematically underprice when they pick a platform from a demo.
Three of four hyperscaler IoT services have exited or scared customers out. Google IoT Core retired in August 2023 with 12 months of notice, IBM Watson IoT retired in December 2023 with zero direct replacement, and Azure IoT Central ran a February 2024 retirement scare later walked back by Microsoft within two days. Only AWS remains committed, and even AWS charges data-gravity and egress fees that shape startup economics before a founder signs.
Commercial scale did not protect Particle – more than $20M of ARR pre-acquisition, absorbed into Digi in January 2026. Samsung ARTIK gave paying customers less than 30 days of notice when it shut in 2019, the pattern every long-horizon hedge plans against.
The mechanical hedge is to commit to open-standard telemetry (MQTT, Sparkplug, OPC-UA) so only the dashboard layer is proprietary, and to reserve one quarter of engineering runway every three to five years for a possible migration. An IoT data orchestration platform that speaks those protocols natively is cheaper to keep than a vendor-specific pipeline is to rebuild.
Where It Actually Breaks: Traeger, Michelin, Particle
The best cautionary tales are the ones with numbers attached, and three of them bound the honest limits.
Traeger Grills migrated 111,000 devices and 700,000 mobile users onto AWS IoT Core inside three months after their platform vendor sunsetted, with per-device cost moving from $6 to $2 to $0.50 – a twelvefold delta, and the whole case for keeping the data layer portable. Michelin’s tires-as-a-service line holds up as the honesty bomb it is: after 3 years, expansion was far below expectations and profitability was terrible. Particle’s January 2026 exit to Digi at more than $20M of ARR bounds the upper end, where commercial traction did not translate into platform independence.
Iotellect’s own documentation caps single-server scale at “several thousand devices” before a distributed install is required – a vendor declaring its own envelope, not a marketing claim. CISIN’s five-year analysis places low-code at roughly 35% higher total cost than custom on heavily-customized mission-critical apps past year three – though CISIN is a custom-software services vendor and the estimate carries that bias. The directional point stands: low-code ceilings are real, even when the exact multiplier is vendor-flavored. Most startups never hit these ceilings. The point is to know where they are before scale forces the question.
Stage-Matched Tooling, Not Tribal Warfare
Seed to Series A is the low-code zone for validation on sub-10K-device fleets. Series A to Series B is the hybrid zone, where UI and configuration stay low-code and the data plane gets owned. Series B and beyond is the managed-cloud-native or full-custom zone that produced the Traeger pattern. Stage transitions are messy by construction, and rebuilds are a sign of success, not failure – the goal is to avoid a surprise rebuild.