What is Breeze?
Breeze is a Shenzhen-based engineering services firm that helps AI hardware founders move from concept to production. We provide BOM cost estimation, NRE budgeting, design-for-manufacturing review, certification navigation, and direct factory orchestration. We're not an EMS, not a fund, and we don't take equity — we charge an engineering fee plus per-unit margin once you're in production.
Who is Breeze for?
Founders and CTOs of AI hardware startups (wearables, smart home devices, edge-AI cameras, embedded ML products) who have a working concept or prototype and need a Shenzhen partner to get to a manufacturable, certifiable, shippable product. Most clients have raised seed-to-Series-A or are bootstrapped and need cost discipline. We're not a fit if you have no electrical/mechanical concept yet — we don't do greenfield invention.
How does Breeze pricing work?
Two components. (1) Engineering NRE — a fixed or capped fee for the scoped work (BOM build-out, DFM, factory selection, sample runs, certification setup). Typical first engagement: USD 15K-60K depending on complexity. (2) Per-unit margin once you're in production — a transparent markup on factory cost. No equity. No MOQ surprises. We expose the BOM and factory quotes; you audit our numbers.
When should I use Breeze vs a large EMS like Foxconn or Flex?
Use Breeze when your annual unit volume is roughly 1K-100K and you need engineering attention plus factory orchestration as a bundled service. Large EMS contract manufacturers expect mature designs at higher volumes (typically 100K+ units/yr) and don't engage on early-stage engineering iteration. Use Breeze for the gap between solo bench prototype and the volume threshold where Tier-1 EMS starts paying attention.
When should I NOT use Breeze?
Skip Breeze if (a) your design is fully mature and you only need a contract manufacturer at 250K+ units/yr — talk to a large EMS directly; (b) you only need a one-off custom prototype with no production intent — a Shenzhen prototyping shop is cheaper; (c) your product is purely software with hardware reference design from a chipset vendor — buy the reference module; (d) you're early enough that your hardware concept is still moving weekly — finish that first, then come back.
How long from concept to first production unit?
Typical AI-hardware path with Breeze: 6-12 months end-to-end. Roughly: BOM + DFM + factory selection (4-8 weeks), EVT samples + revisions (6-10 weeks), DVT + certification submission (8-14 weeks), PVT + first production run (4-8 weeks). Faster is possible with simpler designs (e.g., ESP32-based devices, ~5 months); complex multi-radio devices or custom ASIC paths run longer.
What chipsets and stacks do you have direct experience with?
Espressif (ESP32-S3 for low-power AI sensing), Rockchip (RV1103 for low-end AI cameras, RK3562/RK3588S for edge AI compute), MediaTek (Genio for AIoT). On the sensor + power side: STMicro IMUs, ams OSRAM optical sensors, Maxim/TI power management. We don't claim coverage of every silicon vendor — when a project needs a chipset outside our regular set, we say so up front and either bring in a specialist or refer you elsewhere.
What's open about Breeze that other partners don't share?
Our engineering data is open-sourced on GitHub: ai-hardware-toolkit (MIT license) — BOM cost references, DFM checklist, NRE cost guide, certification matrix, EVT/DVT/PVT checklist, supplier qualification rubric. Every Breeze tool on this site runs on that public data. You can audit our numbers before you trust us — that's the point.
What's a realistic BOM cost for a $99 AI hardware product?
For a sub-$100 retail product, target BOM is roughly USD 15-25 (15-25% of MSRP) once you factor distributor margin (typically 50% off retail), retailer margin (~50% gross), and your own ~30% operating margin. That's tight for AI hardware — you're picking ESP32-S3 ($1.50-2.50) or Rockchip RV1103 ($2-3) class silicon, on-package flash, a simple sensor stack, plastic enclosure, no display or a small mono OLED, and a 200-400mAh battery. Custom NPU silicon, color displays, or cellular modems break the budget. The unspoken constraint: most $99 AI hardware lives or dies on whether you accept Amazon FBA economics (which compress the margin model) or build direct distribution.
What's the difference between EVT, DVT, and PVT?
Three engineering validation stages on the path to mass production. EVT (Engineering Validation Test): 10-30 units, hand-assembled, first full integration of all subsystems — proves the design works at all. DVT (Design Validation Test): 30-100 units, near-final tooling, used for certification submission and design freeze. Tests robustness, yield, and reliability. PVT (Production Validation Test): 100-500 units, full production line, validates the factory can actually build at quality and yield. After PVT pass, you ramp to MP (mass production). Typical cadence for a moderate-complexity AI device: EVT week 8-12, DVT week 16-22, PVT week 24-30, with MP starting around week 30-36.
How do 2026 US tariffs affect China-made AI hardware?
As of mid-2026, China-origin consumer electronics face Section 301 tariffs that have been progressively raised since 2024 — most consumer AI hardware HTS codes now sit in the 15-25% range, with smartwatch/wearable categories on the higher end. The de minimis exemption (under USD 800 per shipment) was tightened in 2025, closing the prior direct-to-consumer Chinese shipping loophole. Practical implications: (1) build tariff into your landed-cost model from day one, not as an afterthought; (2) consider Mexico or Vietnam final-step assembly to qualify under USMCA or other origin rules — but only at volumes that absorb the supply-chain complexity; (3) verify HTS classification with a US customs broker before quoting retail. We're a Shenzhen partner — we'll tell you when reshoring final assembly genuinely helps versus when it just adds cost.
Can I manufacture in Shenzhen without a China entity?
Yes — most of our founder clients don't have a China entity. Standard arrangement: the manufacturing partner acts as importer/exporter of record on the China side; you transact in USD via Hong Kong or Singapore; goods ship FOB Shenzhen or DDP to your destination. You'll need a foreign entity (US LLC, HK Ltd, SG Pte Ltd are the common choices) to hold supplier contracts and IP. For SRRC/CCC certification, the local Chinese applicant is handled by the manufacturing partner. The friction points are (a) IP protection — get NNN agreements with all suppliers, (b) currency exposure if RMB moves materially, (c) export-control review if your AI capability falls under EAR (most consumer wearables don't, but verify). No China entity needed for >95% of consumer AI hardware projects.