Why Surface Art is built for NPI and high-mix, low-volume assembly.

NPI is a different job from volume production

In volume production the design is frozen and the process is under formal change control. In new product introduction (NPI) almost nothing is frozen. The data package is still evolving, parts change between builds, and every revision is effectively a new first article.

The same failure modes show up again and again. Fixes made by hand on prototypes never become documented process. Parts that are easy to find for 10 boards aren't easy to find for 1,000. One US contract manufacturer describes customers moving from under 100 units to 500–1,000 and running straight into shortages: "Finding 50 ICs right now can be difficult. Finding 1,000 can be a 56-week lead time."

Researchers have noticed how poorly served this step is. A Microsoft Research and Carnegie Mellon study interviewed 25 people who had taken prototypes into low-volume production and found the jump from one working device to many identical ones is still hard. A 2025 follow-up is titled, plainly, "Making Hardware Devices at Scale is Still Hard."

What a good NPI partner does

• Reviews design for manufacturing, assembly and test (DFM, DFA, DFT) before parts are placed, and again when the design changes.
• Reviews the BOM for availability at production quantities, not just for this order.
• Tells you what it learned on the first build, in writing.
• Has engineers who help with engineering changes and alternate parts.
• Carries stencils, programs and tooling from NPI into production, so the second build doesn't start over.

In high-mix, low-volume work, changeover is the cost

High-mix, low-volume (HMLV) has no formal IPC definition. In practice it means many part numbers, small and variable lots, and frequent engineering changes. One equipment maker notes that at contract manufacturers, batch sizes of 1 to 250 "are not uncommon." It's the normal pattern in industrial, medical, instrumentation, aerospace and defense electronics.

At small lot sizes, setup time drives cost far more than placement speed does. An I-Connect007 worked example: a 250-board batch at 45 seconds per board runs about three hours. Add a four-hour changeover and overall equipment effectiveness falls below 50%, and "as the batch size goes down, the changeover time dominates."

A line tuned for long runs is optimized for the wrong variable when lots are small. HMLV rewards fast, verified setups, programs built quickly from CAD data, and engineers who stay with the job from build to build.

Why the largest manufacturers deprioritize small programs

Electronics manufacturing services (EMS) is a concentrated industry. According to Manufacturing Market Insider, the top 50 EMS providers earned $477 billion in 2024. The top 10 took 87.6% of it, and it took $521 million in revenue just to make the list. Those businesses are built around anchor programs.

Practitioner advice reflects that. Altium suggests your program should be about 10% of a plant's revenue, and that "if you're less than 5%, it's unlikely you'll get prioritized." That's a rule of thumb, not data, but it matches older analyst findings. Technology Forecasters (c. 2006) found that industrial, medical and aerospace work made up about 12% of revenue at the largest EMS providers but almost half at smaller ones.

None of this means a large manufacturer can't build a good prototype. It's about fit. A high-mix program worth a few million dollars a year is a significant account for a shop Surface Art's size, and a rounding error to a plant built around anchor programs.

The expensive part is what you find late

Analyst figures from Lifecycle Insights, widely cited in trade press, put a typical PCB project at about three board re-spins, at roughly $28,000 to $46,000 each. Estimates vary by source, and the original report isn't public. Schedule usually costs more than the invoice. A widely cited 1983 McKinsey model estimated that shipping six months late could cost a product about a third of its lifetime profit, against a few percent for overrunning the development budget. It's a model, not a survey, but the direction is hard to argue with.

Rework has its own cost. IPC's rework standard (IPC-7711/7721) doesn't cap the number of rework operations, despite what you may have heard. The familiar "three reflows" comes from JEDEC J-STD-020. Moisture-sensitive parts are qualified with three reflow passes because double-sided boards see two and may be reworked once. On a double-sided board, a single BGA rework can use up the last of that margin. Peer-reviewed studies show solder joints change with repeated reflows, and that reworking one BGA can partially reflow its neighbors. Careful, profiled rework keeps the damage small, but every rework spends margin.

That's the case for catching problems on screen instead of on the line, and for treating first-pass yield as a design goal rather than a happy accident.

AI can help design the board. Someone still has to build it.

AI is moving into PCB design. Tools from established design-software vendors and from startups now generate component placement and routing, and AI copilots help draft and check designs. The category is still uneven, with "autonomous" covering everything from a smarter autorouter to a complete layout. But the direction is clear: getting from an idea to a layout is getting faster.

Faster design means more design iterations, and every one of them ends up as a physical board. That puts more weight on the build, not less:

• Software follows the rules it's given. A layout can pass every design rule and still carry a footprint that doesn't match the part's datasheet, a thermal pad that won't reflow evenly, or a part you can't buy at production quantities. Those are exactly what a DFM review and BOM scrub catch before the first part is placed.
• More spins mean more first articles. Each revision is a new first build, so your NPI partner's speed and discipline decide how quickly design gains turn into working hardware.
• Physics doesn't change. Solder joints, reflow limits and inspection don't get easier because a board was designed faster. A board going into a drone or a satellite still has to survive the field.

The more of the design that's automated, the more it matters that the build is something you can rely on: reviewed before parts are placed, inspected on every build, and made by a manufacturer that publishes its delivery and return rates. We publish ours.

Why build NPI in San Jose

Close to your engineers

Most of the benefit of co-location in NPI is human: walking the line together, looking at the first board in person, answering a DFM question the same afternoon. San Jose is 15 to 16 hours behind Shenzhen, so every question sent offshore tends to cost a day. The effect goes beyond convenience. A Management Science study of optoelectronics firms found that moving production to East Asia changed which designs were economical to build at all. For a Bay Area hardware team, "co-located" means a contract manufacturer a short drive away.

No customs entry on every prototype

The $800 de minimis exemption on low-value imports has been suspended for every country since August 29, 2025, and US Customs made the suspension indefinite in June 2026. Every offshore prototype shipment now needs a customs entry and pays full duty. US tariffs have also changed legal basis several times since 2025. Boards built locally never cross a border, which keeps a program's landed cost stable from first prototype to production. Components are still largely imported, so local assembly reduces tariff exposure but doesn't eliminate it.

Control over where your design goes

An NPI build means handing over a near-complete design: Gerbers or ODB++, a BOM with part numbers, assembly drawings, test fixtures, often firmware. The US Trade Representative's 2026 Special 301 Report kept China on its Priority Watch List and called its trade-secret enforcement "weak." It also found that China and Hong Kong accounted for over 87% of the value of counterfeit goods seized by US Customs in fiscal 2025, and pressed India and Vietnam on trade-secret gaps too. The argument isn't that offshore partners will steal your design. It's about knowing, and controlling, where your data goes.

For defense work, it's the law

Under the International Traffic in Arms Regulations (ITAR), anyone who manufactures a defense article in the US must register with the State Department's Directorate of Defense Trade Controls (DDTC), even if they never export. Assembly and test drawings count as technical data, and sending them to a foreign manufacturer is an export that needs authorization. China is under a policy of denial. Separately, a federal law (10 U.S.C. § 4873) bars the Department of Defense from buying covered PCBs from entities tied to China, Russia, Iran or North Korea starting January 1, 2027. The implementing rule is still being written.

Certifications, and what they actually mean

The honest trade-off

Local manufacturing doesn't win on unit cost at high volume, and we won't pretend it does. Kearney's 2026 Reshoring Index models a laptop built in Vietnam at 67 and in China or Mexico at 75, against 100 for full US production. US assembly using imported parts came out higher still, at 114. Kearney's index also stayed firmly negative for 2025: electronics imports grew much faster than US output, and circuit boards still largely come from Asia.

Domestic order books are strong: North American EMS book-to-bill reached 1.68 in September 2026, so capacity is tightening. But the case for building in San Jose isn't that it's cheapest. It's that NPI and high-mix, low-volume work are dominated by engineering attention, changeovers, late-found mistakes and schedule, and those are exactly the costs a nearby, engineering-led shop reduces. Build prototypes, NPI and HMLV close to your engineers, and make the volume decision on its merits once the design is stable. A clean, documented NPI build makes that move easier wherever production ends up.

How Surface Art runs an NPI build

1. Clarify before quoting

Before we quote, we send a short list of clarifying questions and state our assumptions about quantities, scope and requirements. We quote each quantity break you need and note your expected annual volume.

2. Review before a single part is placed

Every package gets DFM (footprint, package and thermal review), DFT (test-point and coverage analysis) and a BOM scrub (sourcing, lifecycle and cost), and we review the datasheet for every part on your BOM. Send the BOM in Excel with manufacturer part numbers; our NPI checklist covers the rest.

3. Build and inspect

Automated optical inspection on every build, X-ray for hidden joints such as BGAs, flying probe for fixtureless electrical test, and functional and in-circuit test fixtures for production programs. Leaded (Sn63/Pb37) or lead-free (SAC 305). Our inspection and test guide explains what each method catches.

4. Into production, when the volume fits

We design our line to hit a 99%+ first-pass yield and hold internal rework under 1%, and we work within the heat-cycle limits component makers publish. When a program's volumes fit our line, we carry it from prototype into production. When they don't, a clean, documented NPI build makes the move to a volume manufacturer easier.

Questions to ask any NPI partner

• Do you review DFM, DFT and the BOM before placing parts, and again when the design changes?
• Do you check part availability at production quantities, not just for this order?
• What do you send me after the first build?
• Who is my engineering contact, and can I walk the line?
• How do your changeovers work for small lots?
• Which inspection and test methods will this board get, and why?
• What exactly do your certifications cover? Can I see the certificates?
• If this program is controlled under ITAR, how do you restrict access to my data?

Ask us any of them about your design.