Congratulations on your design win. Now you only have two years to wait for the money.

There is a wonderfully confusing moment in semiconductor sales when a startup wins what may be the biggest commercial deal in its short history and, financially speaking, almost nothing happens. The customer has selected your technology. Your chip is going into the next platform. Engineers have spent months, sometimes years, getting to this point. Champagne seems appropriate. And it probably is. Just don't accidentally book the design-win value as next year's revenue (your CFO will object). A design win is a very serious commercial commitment, but it is not yet a volume order. In semiconductors, those two events can be separated by years. That sounds like industry trivia. It isn't. For a young deeptech company, understanding what exactly you have sold at each stage before volume production can be the difference between financing your growth and slowly going bankrupt while sitting on an apparently fantastic pipeline.

Socionext explains this unusually clearly in its investor materials. After it wins a custom SoC design, development starts, the customer pays non-recurring engineering costs in phases, the chip goes through customer evaluation and only then does mass production begin. Socionext says the journey from design win through development and evaluation to product revenue typically takes more than two years. More interestingly, NRE revenue was already 17% of its consolidated sales in FY2024. In other words, development is not merely an unfortunate valley the company has to finance until “real” revenue arrives. Development itself is something customers pay for.

That distinction matters far beyond semiconductors. Deeptech founders are often taught to think about commercialization as a slightly crude sequence: R&D, pilot, product, revenue. The problem is that for technologies requiring customer-specific engineering, qualification, certification, integration or tooling, there may be an enormous amount of economically valuable work between “the customer wants this” and “we can ship 100,000 units”. If you treat all of that as pre-revenue R&D, you are effectively volunteering to finance part of your customer's product development. Very generous. Less obviously good business.

The semiconductor industry has a useful answer to this: separate the economics of creating the product from the economics of producing it. NRE, or non-recurring engineering, is the money paid for work that needs to happen once: adapting a design, developing customer-specific functionality, verification, tooling, test development, integration and other work required to get to production. The eventual chip price pays for the thing that gets produced repeatedly. That sounds obvious when written down. Yet early deeptech companies routinely roll both into an imagined future unit price and then spend twelve or eighteen months doing bespoke engineering for free because a large corporate has promised attractive volumes afterwards.

And attractive future volumes can be very attractive indeed. NVIDIA reported an automotive design-win pipeline of $14 billion through FY2029 back in 2023. Infineon Technologies bought Marvell's Automotive Ethernet business last year partly on the strength of a roughly $4 billion design-win pipeline through 2030. The business itself was expected to generate only $225 million to $250 million of revenue in calendar 2025. Those aren't contradictory numbers. They are a rather nice illustration of how semiconductor economics works: a design win can represent years of potential future production revenue.

There is another useful warning buried in Socionext's numbers. The company tracks something it calls its “design win balance”, essentially the remaining expected value of active design wins. For some earlier cohorts, projects that were subsequently cancelled represented around 20% of the original design-win amount. Other projects grew enough in price or volume to compensate, but the point remains: even after being designed in, future revenue is not entirely yours. Products get cancelled. Volumes change. Launches slip. Customers redesign things.

This creates a commercial mistake I think deeptech companies should be much more careful about: pricing today's certain engineering work against tomorrow's uncertain volume.

Imagine an industrial customer wants your new photonic sensor, advanced material, power semiconductor or neuromorphic processor incorporated into its next product. Getting there requires twelve months of work from your team. The customer tells you that, once launched, the program could be worth €5 million a year. Wonderful. It might be. But the €5 million is conditional on qualification, their product actually launching, expected volumes materialising, your technology remaining in the design and probably half a dozen things neither side can control. Your engineers' salaries next Tuesday are rather less conditional.

So ask a different question: what is the customer buying from us now?

Sometimes the answer really is an evaluation. Fine. Make the evaluation small and explicitly limited. Sometimes they are buying a pilot. Price the pilot. Sometimes they want substantial customization because your technology does not yet fit their system. That starts looking like NRE. Sometimes both sides are genuinely creating something new and sharing development risk, in which case a joint development agreement may make more sense, with the inevitable and extremely important discussion about who owns the resulting IP. And sometimes the technology is ready and what the customer really wants is production access, in which case you can start talking about units, licenses, royalties or whatever the eventual business model happens to be.

The labels are less important than separating the commitments. A useful commercial architecture might therefore have three layers: development money now, milestones while risk is being retired, production economics later. The first makes sure you are not providing unlimited engineering for free. The second keeps both sides committed as the program progresses. The third preserves the upside that made the opportunity interesting in the first place. Socionext's model is a particularly clean real-world example because its NRE payments arrive in phases during development before the eventual product revenue.

This also changes how I would look at a deeptech pipeline. €20 million of “potential customer value” is a lovely number for a pitch deck. It tells you surprisingly little about whether you have a business. I would rather know: how much of that pipeline is merely technical interest? How much is under evaluation? How much has a defined development scope? How much has funded NRE? How much has reached a genuine design win? How much is qualified? And how much has an actual production forecast attached to it? Those stages are not cosmetic CRM labels. Each represents a different amount of commercial risk having disappeared.

There is a broader lesson here for deeptech beyond chips. We tend to inherit software language for businesses that behave nothing like software. ARR, trials, conversion funnels and seat pricing are useful concepts if your marginal deployment looks roughly like the previous one. They become less useful when selling involves twelve months of engineering, customer qualification, physical integration, certification and a production ramp. Semiconductor companies have spent decades learning how to make money across that awkward middle. Advanced materials, photonics, robotics, industrial biotech and other deeptech companies would do well to steal some of the machinery.

So yes, celebrate the design win.

Then send the NRE invoice.

If you're working out what your first corporate customer should actually be paying for, rather than simply what your eventual product should cost, that is exactly the kind of commercial problem we like at SLSCRW.