What CT scans of a BYD reveal about Chinese manufacturing

What happened

Lumafield, the industrial CT-scan company that has turned X-raying consumer products into a recurring internet event, pointed its scanners at BYD. The latest Scan of the Month strips a handful of parts from BYD vehicles down to their internal geometry — castings, busbars, connectors, sensor housings — and shows what's actually inside the components Western automakers benchmark against on a spreadsheet but rarely see in cross-section.

The scans are not a teardown in the Munro & Associates sense. There's no bill-of-materials, no labor-hour estimate, no "we could build this for $4,200" headline. What Lumafield publishes is the geometry — and in automotive manufacturing, geometry is the strategy. A single gigacast replacing 70 stamped-and-welded parts isn't a cost story; it's a tooling, supply-chain, and design-validation story that ripples back through every CAD seat in the building.

The scans land at a politically loaded moment. BYD outsold Tesla in global BEV units in Q4 2023 and has spent 2024–2025 expanding into Europe, Latin America, and Southeast Asia while US tariffs keep it nominally walled off. The question every Western OEM engineering org is quietly asking — *how are they actually building these* — gets a partial, visual answer.

Why it matters

The interesting finding from the scans isn't any single clever part. It's the consistency of one design principle: fewer pieces, more function per piece. Connector blocks integrate strain relief and sealing into a single molded body. Sensor housings combine the mount, the EMI shield, and the cable gland. Castings absorb structural roles that would normally require separate brackets, reinforcements, and fasteners.

This is the same gospel Sandy Munro has been preaching about Tesla for half a decade, and it's the same lesson the aerospace industry learned with additive manufacturing: the part you don't have to make is the part that never fails, never gets miscounted, and never adds a supplier to your BOM. BYD has internalized it across a vehicle program, not just a hero component.

The second-order observation is vertical integration made visible. BYD makes its own batteries (Blade), its own motors, its own power electronics, and — through subsidiaries — its own semiconductors. When you control the silicon and the casting and the wiring harness, you can co-design across boundaries that a Tier-1-dependent OEM cannot cross without a contract negotiation. The scans show parts that look like they were designed by one team, because they were.

The community reaction on Hacker News (336 points and climbing) split predictably. One camp reads the scans as confirmation that Chinese automotive engineering has caught up and, in EV-specific subdomains, pulled ahead. The other camp points out, fairly, that CT scans don't measure durability, NVH, crash performance, or 10-year corrosion behavior — the things that actually separate a good car from a cheap one. Both are right; the scans are evidence of manufacturing maturity, not a verdict on the finished product.

What the scans do settle is the lazy framing that BYD's cost advantage is purely labor and subsidies. The geometries on display require expensive tooling, sophisticated simulation, and a willingness to amortize that investment across volumes Western OEMs aren't currently hitting. It's capex courage, not cheap workers.

What this means for your stack

If you're a software engineer reading a developer publication, the obvious question is: why is this in my feed. The answer is that the principle Lumafield's scans illustrate — integration as a competitive moat — is the same principle that's eating tech right now.

The AI labs that ship fastest are the ones that own their inference stack end-to-end: model, kernel, scheduler, hardware contract. The cloud providers winning margin are the ones designing their own silicon. The product teams shipping reliable LLM features are the ones who stopped chaining seven SaaS APIs together and built the integration layer themselves. BYD's gigacast is the industrial-economy version of the same bet your platform team is making when it kills a microservice.

The practical translation: when you're staring at a system diagram with 40 boxes and wondering why latency is bad and on-call is worse, the right question is rarely "which box should I optimize." It's "which three boxes can become one." Part-count reduction is a software discipline too — it just gets called *deletion*, *consolidation*, or *the boring rewrite that paid for itself*. The teams that do it well share BYD's posture: they treat the architecture itself as the product, not the individual modules.

The second takeaway is about benchmarking. Western automakers have been buying BYDs, tearing them down, and publishing cost analyses for two years. They mostly missed the geometry story because cost teardowns count parts and price them; they don't ask why the part is shaped that way. The scans are a reminder that benchmarking competitors by their outputs misses the upstream decisions that produced those outputs. If your competitor ships features twice as fast, the answer is almost never in the features. It's in the build system, the deploy pipeline, or the org chart.

Looking ahead

Lumafield will keep doing this — Apple products, Chinese drones, counterfeit chargers, and now cars — because the format works: visible engineering is shareable engineering. The deeper trend is that industrial CT, once a $500k QA tool buried in aerospace inspection labs, is becoming a public-facing analysis medium. Expect more of these scans, expect Western OEMs to commission their own (quietly), and expect at least one US automaker to announce a "giga-something" press release within twelve months that's directly traceable to what these images make undeniable. The geometry is out. The strategy follows.