The $60,000 Mould Problem: Why MVP Fails Hardware and What Replaces It

97% of Hardware Startups Die or Go Dormant, and the Framework Deserves Part of the Blame

Hardware startups fail at rates that make software's mortality look mild. According to MacroFab's analysis of industry data, 97% of hardware ventures either die outright or become zombie companies — shipping nothing, raising nothing, persisting only on paper.¹ CB Insights and Hemar Group put a finer point on the failure mode: more than 70% never reach mass production at all.² The single largest cause, cited in 42% of post-mortems, is protracted development — teams that run out of money before they finish building.¹

The lean startup methodology, formalised by Eric Ries in 2011, prescribes a specific remedy for this: ship a minimum viable product, measure customer response, iterate. The prescription works when iteration is fast and cheap. A software team can deploy a new build in hours. A hardware team cannot. Instrumental, the manufacturing-analytics firm, estimates that a single hardware iteration runs roughly 100 times longer than a software one — weeks for simple parts, months for anything requiring work from multiple vendors.³ The build-measure-learn loop does not shrink that cycle. It merely names it.

A Single Mould Costs More Than Most Seed Rounds Allocate to Product

The asymmetry is financial before it is philosophical. Jaycon's 2026 injection-moulding pricing report breaks tooling into five classes.⁴ A prototype aluminium mould (Class 105), good for fewer than 10,000 units, costs $1,500–$8,000. A production-grade hardened steel mould (Class 101), rated for 250,000 to over one million shots, costs $60,000–$150,000. A family mould combining multiple parts runs $80,000–$200,000.

Most consumer hardware products contain not one moulded part but many. Bolt's portfolio data puts the range at 5–50 injection-moulded components per product.⁵ A mid-complexity device with 15 moulded parts, each requiring its own Class 103 mould at $25,000–$60,000, faces $375,000–$900,000 in tooling alone — before a single unit ships.

That figure sits awkwardly beside what hardware teams typically raise. The average hardware crowdfunding campaign brings in $92,000, per Bolt's analysis.⁵ Even a priced seed round, whose mean pre-money valuation Bolt pegs at $4.5 million, does not leave room for six-figure tooling if the capital must also cover salaries, certification, and a first production run.

The MVP framework offers no guidance on how to sequence those commitments. It says: build the minimum product a customer will pay for. It does not say: the minimum product requires $375,000 in irreversible tooling before anyone's willingness to pay has been established.

The Iteration Gap Runs from $2 to $150,000

What makes hardware distinct is not that iteration is hard. It is that iteration cost spans five orders of magnitude depending on which layer of the stack a team is testing.

At the bottom: a PCB prototype from JLCPCB starts at $2 with 24-hour production turnaround.⁶ A desktop 3D printer from Formlabs or Prusa can produce a functional enclosure overnight for the cost of resin. At this layer, the build-measure-learn loop works exactly as advertised.

At the top: a production steel mould costs $60,000–$150,000 and takes 10–14 weeks to cut.⁴ FCC certification for a device with custom RF runs $8,000–$20,000 and two to three months; a cellular device requiring carrier approval can reach $200,000.⁷ These are not iteration costs. They are commitment costs — capital deployed against a specific bill of materials that cannot be recouped if the design changes.

Between those extremes lies the breakeven that determines strategy. HLH Rapid's analysis finds that 3D printing and injection moulding cross over at roughly 500 units.⁸ Below that threshold, a team can print parts at approximately $7 each and iterate freely. Above it, moulded parts fall to under $1 each — but only after the mould is paid for. At 10,000 units, injection moulding costs approximately $0.99 per part versus $7.00 for 3D printing, a 7:1 ratio.⁸

The MVP framework treats all of this as a single question: is the product viable? The cost structure demands a sequence of narrower questions, each matched to the tooling tier it requires.

The Replacement Is Not a Product but a Sequence of Falsifiable Assumptions

The concept that fits hardware is not new, though it carries several names. Rik Higham called it the Riskiest Assumption Test in 2013. Giff Constable and others refined it.⁹ The principle is the same: identify the single assumption whose failure would kill the venture, and test that assumption — and only that assumption — at the lowest tooling cost that produces a valid signal.

For hardware, this means staging tests by the cost of the tooling each test requires.

Stage 1: cardboard and code ($0–$500). Test desirability. A foam model, a rendered video, a landing page collecting email addresses. No electronics, no firmware. The assumption under test: does anyone want this object in their life? Juicero, which raised $120 million before shipping, skipped this stage entirely — building a device with custom force sensors, a machined aluminium body, and a proprietary packet-scanning system before validating that consumers would pay $400 for pressed juice packs they could squeeze by hand.¹⁰

Stage 2: dev boards and 3D prints ($500–$10,000). Test feasibility. Off-the-shelf microcontrollers, 3D-printed enclosures, hand-soldered PCBs. The assumption under test: can the core technical claim be demonstrated? James Dyson spent five years and 5,127 prototypes at roughly this tier, testing cyclonic separation in hand-built rigs before committing to production tooling.¹¹ His cost per iteration was low. His confidence at the end was high.

Stage 3: soft tooling and small-batch manufacturing ($10,000–$50,000). Test manufacturability. Aluminium moulds (Class 105, $1,500–$8,000 each), small PCB production runs, pre-certified wireless modules that cut FCC costs by 60–80%.⁴ ⁷ The assumption under test: can this design be manufactured at a cost that supports the target retail price? Bolt recommends a minimum 2.5x ratio between bill of materials and retail price.⁵ This is the stage to find out whether that ratio holds.

Stage 4: hard tooling and certification ($50,000–$500,000+). Commit to a specific design. Steel moulds, FCC/CE certification, UL safety testing. Production timelines at this stage run 180 days on paper but stretch to 270 or more in practice, per Bolt's data.⁵ Early production scrap rates average 5%, declining to 0.5% over subsequent runs.⁵ The assumption under test is no longer about the product. It is about the operation: can this team manage a supply chain, hit a yield target, and fulfil orders?

Each stage gates the next. No team should cut steel until aluminium has validated the geometry. No team should enter certification until the bill of materials is stable. The framework is not faster than MVP. It is more honest about what each dollar buys.

The Falsification: Tooling Costs Are Falling, and the Stages May Compress

The strongest objection to staged assumption testing is that the cost tiers it rests on are not stable. Desktop 3D printers that cost $3,500 in 2014 now cost under $1,000 at higher resolution. PCB prototyping has collapsed from weeks and hundreds of dollars to days and single digits.⁶ Chinese contract manufacturers like JLCPCB offer assembled PCBs at price points that make a ten-board test run trivial.

If mould costs follow the same trajectory — through advances in 3D-printed injection moulds, CNC-machined soft tools, or additive metal tooling — the gap between stage 2 and stage 4 narrows. At some point, the cost of testing manufacturability may drop low enough that a team can afford to iterate on production tooling the way it now iterates on a PCB. The staged framework assumes a five-order-of-magnitude cost spread. If that spread compresses to three, the stages collapse with it, and something closer to the original MVP loop becomes viable for physical products.

That has not happened yet. Jaycon's 2026 data shows Class 101 steel moulds still starting at $60,000.⁴ Semiconductor NRE remains stratospheric — Silicon Analysts estimates $30–50 million at 28nm and $200–400 million at 5nm, with mask sets alone costing $10–20 million at 3nm.¹² For any hardware category involving custom silicon, the staged approach is not optional. It is the only approach that does not require burning nine figures before a customer touches the product.

What Changes for the Person Holding the CAD File

A founder staring at a bill of materials and a bank balance does not need a methodology. That founder needs a decision rule: which assumption do I test next, and what is the cheapest tooling that produces a valid answer? The answer is never "build the minimum viable product." The answer is always "build the minimum testable version of the one claim that, if wrong, makes everything else irrelevant." The difference is not semantic. For a fifteen-part consumer device, it is the difference between a $375,000 bet and a $7,500 one.