Oct 07, 2026 / Avik Arefin

How Long Does It Take to Make a Prototype? Hardware Timelines by Stage

A first works-like hardware prototype takes about four to five months and production takes longer, so a founder's first 1–2 weeks are best spent testing the one part most likely to fail.

Author

Avik Arefin

Published

October 07, 2026

How Long Does It Take to Make a Prototype? Hardware Timelines by Stage

A first works-like prototype takes months, production takes longer, and the first 1–2 weeks are best spent testing the one part most likely to fail.

It takes about four to five months to make a first works-like hardware prototype. That is the median of 19 weeks per prototype that Looney et al. found in interviews with 55 hardware-startup leaders, 20 of them in cleantech. The median held regardless of the prototype's complexity. Production then adds factory test builds, and in Instrumental's sample schedule the first one alone takes six to nine weeks. A 1–2 week sprint cannot replace those months; it tests the part most likely to fail before you pay for them. What those months cost is covered in how much it costs to get a prototype made.

Hardware prototype stages: what counts as the MVP

A hardware MVP (minimum viable product) is a works-like prototype: it does the core job but does not yet look or ship like the product. It usually passes through four stages:

  1. Dev-board prototype. A development board (dev board) is a ready-made circuit board, such as an ESP32 or Raspberry Pi board, that runs a chip without custom hardware.
  2. Custom PCB spin. A PCB (printed circuit board) spin is one cycle of designing, fabricating, assembling, and testing your own board. A re-spin is each repeat of that cycle after a fix.
  3. Firmware and app integration. Firmware is the code on the device. Integration makes the device, the cloud backend, and the phone app work as one system.
  4. Prototype enclosure. A 3D-printed case that holds the board, battery, and antenna in roughly the final shape.

The stages after the MVP prepare the product for a factory. Instrumental, a manufacturing-analytics company, defines them this way. EVT (engineering validation test) is the first build that combines the final look and function with production-intent materials. DVT (design validation test) checks factory yield on one production-worthy design made from hard tools, the production molds. PVT (production validation test) checks yield at mass-production speed, while certification and tooling run alongside these builds.

Custom PCB spin: days to fabricate, several spins to finish

A custom PCB replaces the stack of dev boards with one board shaped for your product, once you decide to switch from a dev board. The factory step is now fast. JLCPCB's September 2026 help page lists 24 hours to fabricate five simple two-layer boards and two to five days for four-layer boards. JLCPCB says 90% of its assembly orders finish within 24 hours, and 48 hours at most. Courier shipping comes on top, and JLCPCB warns that customs can delay it. In the US, OSH Park quotes 9–12 calendar days for standard two-layer bare boards.

The time goes into repeating that cycle. A 2018 Lifecycle Insights study, cited by Mentor, found an average of 2.9 re-spins per project, adding about 16 days of unplanned work. Bolt, a hardware venture fund, wrote in 2015 that a PCB often takes 5 to 10 revisions before mass production. Budget the PCB stage in weeks, and expect more than one spin.

Firmware and app integration: a 14-day app-store step

Firmware work ties the device, the cloud backend, and the phone app into one system, and publishing the app adds its own wait. On iOS, review adds little time. Apple says it typically reviews at least 50% of submissions in under 24 hours and 90% in under 48. Google Play adds a step for new developers. Personal developer accounts created after 13 November 2023 must run a closed test with at least 12 testers for 14 days before release. Start that closed test while firmware work continues, so the 14 days do not land at the end.

Enclosure: days to print, then the radio test

A prototype enclosure is 3D printed, and printing takes days: JLC3DP lists two days for resin (SLA) parts and three days for nylon (MJF) parts. The enclosure also changes how the radio performs. Espressif recommends at least 15 mm of clearance in all directions around an ESP32-S3 module's antenna inside the housing. Test the radio inside the real enclosure before you finalize either the case or the board.

EVT, DVT, PVT, certification, and tooling after the MVP

Each production build repeats the same cycle of building, testing, and iterating. In Instrumental's sample consumer-electronics schedule, the EVT cycle takes six to nine weeks, and Instrumental calls that schedule "optimistic."

Tooling means the molds that shape plastic parts, and it takes two to three months. Bolt wrote in 2015 that molded plastic parts typically take 8–12 weeks to develop and tune. Instrumental's schedule gives about 12 weeks to make hard tools. It advises waiting for EVT reliability results before starting them, because changing hard tools costs time and money.

Certification can take months when the radio is your own. SparkFun's 2019 FCC certification of its Artemis Bluetooth module took two and a half months, from 1 June to the FCC listing on 16 August. Test engineers Cory Bradshaw and Craig Brandt write that a pre-certified radio module significantly reduces development cost and time to market. The finished product still needs testing with the module installed. Pick a pre-certified module while the design is still on dev boards.

Europe has added dated rules. The European Commission says its cybersecurity rules for internet-connected radio equipment, Delegated Regulation (EU) 2022/30, entered into application in August 2025. The Cyber Resilience Act's reporting obligations apply from 11 September 2026, and its main obligations from 11 December 2027.

Hardware delays: the four usual causes

Hardware schedules run late more often than not. Of 471 funded Kickstarter design and technology projects due by July 2012, Mollick found only 24.9% delivered on time. Four causes recur in the sources.

Part lead times and shortages. Lead time is the wait between ordering a part and receiving it. In April 2026, Raspberry Pi said the LPDDR4 memory on its Pi 4 and Pi 5 cost seven times more than a year earlier. In February, it had blamed competition for memory factory capacity from the AI infrastructure roll-out. Check stock and price history for every chip your design depends on before you lay out the board.

Late design problems. Design problems found after the MVP cost weeks. In Instrumental's example, a major design problem found in EVT reliability testing adds a 6–9 week build plus 2–4 weeks of tool changes.

Radio and certification issues. Rhein Tech, a test lab, cites Intertek's finding that nearly 50% of products fail electromagnetic compatibility (EMC) testing the first time. Plan time for at least one retest.

Scope creep. Scope creep is the addition of features after the plan is set. PMI's 2018 Pulse of the Profession survey found it in 52% of the projects it covered, across industries. Finalize the MVP's feature list before the first PCB spin.

If one of these risks sits inside your device, send your idea in three sentences through cortextech.dev. We will reply with the part we would test first.

AIoT timelines: when the data takes longer than the hardware

AIoT devices run AI models on the device itself, and whether the model sets the schedule depends on whether data for your feature exists.

Some features have public data. Google's Visual Wake Words dataset, built from the COCO photo collection, labels images as "person" or "not person." Chowdhery et al. report that models under 250 KB, a size chosen to fit microcontrollers, reach 85–90% accuracy on it. A person detector can therefore start from a trained model on the first day.

Other features need new data. Pete Warden's Speech Commands dataset holds 105,829 one-second recordings of 35 words from 2,618 speakers. A custom wake word or a product-specific sound needs a collection effort of its own.

Data from the real device matters too. In a Google study, 92% of 53 AI practitioners had hit at least one "data cascade," a data problem that compounds downstream. Problems from real-world noise, such as dust, shadows, and moved sensors, took up to two to three years to appear. If your AI feature needs custom data, start collecting it with the device's own sensor while the first prototype is built.

1–2 week prototype sprint: what it can and cannot prove

A 1–2 week sprint can answer one risky question with a measured number. Three questions fit that window:

  • Can the on-device model hit its latency target? Espressif reports 54 ms for its person-detection example on an ESP32-S3, but only for inference. We time the full chain on the target chip, from wake-up to alert.
  • Can the battery last? We measure current on the chip's own supply, in sleep and in each active step, not the dev board's extra parts.
  • Can the radio hold its link from the install site? The next section walks through this test.

A sprint cannot produce a finished custom PCB, since that takes several spins. It also cannot produce a finished enclosure, a certification, a trained model for data that does not exist yet, or reliability data from EVT builds. Before a sprint starts, write its question as a number with a pass or fail threshold.

CortexTech builds works-like prototypes as one system: firmware, electronics, backend, and companion app, on prototype boards that integrate existing chips and modules. We built a non-technical founder's connected home-entry device with video calling, on-device AI, and a companion app. When a prototype needs industrial design or high-speed boards, we bring specialists into the project. We harden a validated prototype to the point where a DFM (design for manufacturing) and manufacturing partner takes over.

When a founder's device must hold a wireless link from where it will be installed, the link is one of the first things we test. This applies to Wi-Fi, BLE (Bluetooth Low Energy), LoRa, and cellular alike. A link that works on a desk says little about a basement plant room or a metal cabinet. Espressif says the same in its design guidelines: test the final product for throughput and communication range.

The founder cannot see this risk from outside, because walls and enclosures change the link. In 1997, NIST's William Stone measured signal loss through building materials at 0.5–2 GHz and 3–8 GHz. A 1 mW signal through three layers of masonry block, 610 mm thick, arrived at about 0.05 mW, one twentieth of its power. Concrete absorbed the most, and loss generally rose with frequency. Semtech's North American LoRa kits run at 915 MHz, inside the tested range. Wi-Fi and BLE on 2.4 GHz chips such as the ESP32-S3 fall in the gap NIST did not test.

Metal near the antenna causes a second problem. Bradshaw and Brandt, test engineers at D.L.S. Electronic Systems, describe a Wi-Fi module mounted in an all-metal appliance. Reflections from the cabinet pushed the radio's harmonic emissions over the limit during compliance testing. A plastic spacer that moved the antenna away from the metal fixed it.

The obvious fix for a weak link is a bigger external antenna, but that can cost certification time. A pre-certified module's integration instructions list the antennas approved for use with it. Bradshaw and Brandt note that an unapproved antenna means more extensive testing. It may also mean a Class II permissive change filed with a TCB, the FCC-recognized body that issues radio grants. SparkFun estimated that adding antenna models to its 2019 Artemis certification would cost about $2,000 each in testing, plus $2,500 to $6,000 in filing fees. If the fix moves the antenna on the board, it also means a re-spin.

In a 1–2 week sprint, we build the link on the module maker's dev kit with an antenna from its approved list. We set it up at the target distance, through the same wall or enclosure material the product will face. The firmware logs every disconnect, its reason, the time to reconnect, and the signal strength over several days. In Espressif's ESP-IDF, the usual response to a disconnect, such as lost beacons, is to reconnect, so drops go unseen without a log.

The founder then decides with numbers: drops per day and the longest outage. If the link holds with margin, the custom board follows, with the antenna clearance designed in. If it fails, the design changes: a new placement, another approved antenna, a repeater or gateway, or a lower-frequency radio. If no option holds, the founder drops that install location from the product's promise. The reusable method: find the part that could force a re-spin or a retest, and measure it before building the rest.

Send your device idea in three sentences through cortextech.dev. We will name its riskiest part and how a 1–2 week sprint would test it.

Hardware prototype timeline: what to plan

Plan about four to five months per works-like prototype, and expect more than one PCB spin. After the MVP, plan for six-to-nine-week EVT cycles, two to three months of tooling, and months of certification for a custom radio. Plan custom AI data collection as its own project. Spend the first 1–2 weeks measuring the part that could stop the product.

Sources

Want to read more?

More engineering deep dives from the Cortex R&D team.

All Articles