Bert Liu

A capability deck tells you a factory can make your part. It does not tell you they made yours right.

When you are scaling a high-value order from the other side of the world, the riskiest stretch usually is not choosing the supplier. It is the gap between the PO and the delivery — when the part is actually being built, decisions are being made, and no one from the customer’s side is in the room.

That is the part I care about most, so I spend time on the floor.

It starts before any machine runs: the sheet itself. Right gauge, right finish, right material — no quiet substitution that nobody flags until it shows up as a field failure.

Then the part lives on a screen for a while. Sitting with the engineer over the DFM, you catch the spec problems while they are still pixels — the feature that fights the tooling, the tolerance that will not hold, the drawing decision that turns into production pain later.

On the floor, the NCT turret does the punching, but only after the right tools are loaded and checked against the setup sheet. Wrong tool, wrong sequence, wrong assumption, and you have lost a shift — which means you have lost lead time the customer was counting on to the day.

Then bending. Then welding — hardware onto the formed panel, where heat and fixturing decide whether it stays flat or pulls out of tolerance.

Then the first article, measured against the drawing before the rest are allowed to follow.

But the real question is whether the thousandth part is as good as the first — and whether someone catches the drift before it ships.

When the stakes are high and the customer is an ocean away, the value is not a quality report after the fact. It is someone in the room who knows what correct looks like at each step, and catches the problem while there is still time to fix it.

I can do that because I have been hands-on with the process and learned where the details and bottlenecks really show up. You cannot understand that only by watching from a distance.

That is the work.

One thing I enjoy about Computex is getting to ask dumb questions.

Looking at this Delta micro fan, my first thought was: if AI servers are moving toward liquid cooling, why aren’t phones doing the same?

The R&D engineer’s answer was surprisingly simple.

Phones don’t have the space.

This fan sits near the camera area because that’s one of the few places with usable volume left once cameras, batteries, processors, antennas, and everything else are fighting for space.

But the more interesting detail was what happens after the fan moves air.

The airflow path is hidden inside the housing.

Small outlet pockets. Woven mesh inserts. Molded plastic features. Bonded or heat-staked edges. Dust control. Vibration control.

At that scale, cooling is not just a thermal problem.

It is a packaging and manufacturing problem.

The fan blades are incredibly thin, pushing the limits of injection molding and precision manufacturing while still needing to survive years of vibration, dust exposure, drops, and daily use.

One pattern I noticed throughout Computex: the most interesting engineering is not always the newest technology.

Sometimes it is pushing an existing technology right to its limit inside a space where every millimeter is already taken.

Three days into Computex, one pattern kept showing up: AI cooling is attracting expertise from unexpected places.

Walking through booths focused on CDUs, cold plates, manifolds, pumps, immersion cooling, and two-phase systems, many of the strongest players were leveraging capabilities they had already built elsewhere.

Delta is applying decades of experience in motors, power electronics, and thermal management to cooling infrastructure.

I also met founders who previously worked at larger companies and are now building businesses around specific cooling bottlenecks.

What stood out was not the cooling technology itself. It was where the underlying capability comes from.

A cooling demo is relatively easy. But a system that runs 24/7 in a customer environment for years without leaks, contamination issues, service headaches, or reliability failures is a completely different challenge.

That is less of a thermal problem and more of a manufacturing, quality, and supplier capability problem.

When evaluating suppliers, that is what I find myself paying attention to:

• How do they validate reliability?
• How do they leak-test?
• Which components are actually made in-house?
• What kind of talent is required?
• What parts of the system are outsourced?
• What happens when something fails in the field?

Most teams can compare spec sheets. What is harder is determining whether a supplier can consistently execute once real production and customer requirements begin.

I walked into the showroom and the first thing I noticed was the scale.

Landing gear components. Engine structural parts. Heavy, precise, expensive to get wrong.

I visited an aerospace CNC shop in Taiwan recently. 500+ people across multiple facilities. They started on smaller parts and built toward large workpiece capability over time.

The production floor could not be filmed. Confidential programs. Confidential customers. Confidential geometries. AS9100 certified. NADCAP accredited.

That confidentiality is part of what you’re buying when you qualify a supplier at this level.

Cutting the part correctly is the baseline. Proving how it was cut, inspected, measured, controlled, and repeated is what aerospace customers are paying for.

Some parts can ship with documentation packages longer than their machining time: process logs, inspection records, material traceability.

Not bureaucracy. The documentation is the product.

Materials like Inconel do not behave like standard alloys. Tooling can cost several times more, and tool wear becomes a production risk variable, not just a maintenance item.

One bad decision costs more than the scrap.

During a difficult period, cost pressure nearly broke the business. A long-term anchor customer continued supporting the company and gave them room to survive, reinvest, and keep building capability.

Standing in a showroom full of aerospace components, that context lands differently as those parts did not get there by accident. They got there through difficult programs, years of documentation discipline, and customers who trust them enough to stay when things get hard.

The original founder was there during the visit. There was a straightforwardness about him. He knew the company needed different leadership to push through the next phase, so he made that call himself. Handed it to his partner and kept building.

A polished CNC website tells you almost nothing. The management team in the room tells you more than the website ever will.

Aerospace supplier capability is not just machines and certifications. It is what gets built around them over time. And the people who had the discipline to do it honestly.

A coating spec doesn’t guarantee coating performance.

“200°C for 20 minutes” is a target, not a confirmation.

I was reviewing a finishing process with one of our coating partners in Taiwan last week. They handle work across consumer electronics and motorcycle components — Garmin, Ducati, that range.

Honestly there was too much to absorb in one visit. I’m still going back through the footage now.

But one thing stood out immediately: how many steps are involved before and after the coating is actually done.

It starts with pretreatment. Not the spray booth.

Aluminum uses trivalent chromium conversion. Steel uses phosphate. Immersion time, bath temperature, degreasing sequence, hanging method, fixture design. All controlled and posted on the wall where operators work. Not filed away somewhere.

Inconsistent prep or cure conditions can produce a coating that looks fine on delivery and fails in the field. Peeling, corrosion, degraded performance. By the time it shows up, the part is already with the customer.

Cure verification is not the same as cure specification. Oven air temperature and part temperature are not the same number. They verify actual conditions using profiling equipment — multiple sensors placed across different sections of the part as it moves through the line.

Then there’s what happens after the booth. White discs, paper inserts, surface separators at the packing table. The shop started in IKEA lighting assembly before specializing in coatings. That background shows here. A correctly made part that arrives damaged is still a failure.

One thing worth noting on materials: Taiwan has a mature coating supplier base. Most applications are covered locally. Global systems like AkzoNobel or PPG are accessible when customers specify them. The coating brand is one part of it. Everything around it still has to work.

Most of this doesn’t appear in a capability presentation.

If you’re evaluating a finishing supplier and want a second set of eyes on what to look for, feel free to reach out.

Manufacturing isn’t a single process. It’s a system.

A part can move through die casting, surface prep, X-ray inspection, CMM measurement, painting, masking, packaging, and logistics before it is ready.

The real question for hardware brands is not whether to outsource everything or build everything.

It is what you should control directly, and what you should scale through trusted partners.

Still improving the filming and audio, but this is the start.

I’m a big fan of engineering content on YouTube, and what Hacksmith built is impressive.

This isn’t an insider case study. It’s just how I think about manufacturing scaling decisions.

A hit product can validate demand. But it does not automatically justify taking on a much heavier operating model.

That’s where things can shift from product momentum into operational debt: machinery, hires, management, training, QC systems, maintenance, internal systems, and monthly operating burn.

To me, there are usually three paths: full in-house, full outsource, or the middle path — validate trusted suppliers, forecast demand, and keep only critical layers in-house.

That middle path is often the smartest one for businesses whose original engine is still brand, product, and audience.

The goal isn’t to outsource everything or build everything. It’s to know what to protect, what to outsource, and when to scale each layer.

If you’re thinking through similar manufacturing scaling decisions, feel free to reach out.

Reshoring doesn’t mean bringing everything back.

What parts of manufacturing are actually critical, and what parts are not?

Some things may need to stay close to home:

• final assembly
• sensitive testing
• software / electronics integration
• critical dimensions
• anything that protects real IP

Other things may still be better handled by the right specialists:

• castings
• standard CNC
• paint / surface finish
• injection molding
• sheet metal fab
• non-critical subcomponents

So I don’t think the answer is full outsourcing or full reshoring.

I think the real competitive advantage is knowing what to protect, what to outsource, and how to split the risk without losing the product.

Precision looks very different depending on how a part is made.

On one end: thin-fin copper heat sinks (stamped at high volume) > tight fin spacing (pitch control) > tool wear over long runs > consistency at micro scale.

In some cases, final adjustments are still done by hand - using precision tools to align and assemble components into a larger structure.

On the other: large aluminum die-cast housings > material flow > thermal behavior > structural integrity at scale (porosity, leak paths).

Both require precision - but in completely different ways. And what is often overlooked is this: precision does not come from a single process.

It comes from how processes stack - as a system. Casting for shape at scale, then machining where tolerances actually matter.

That is where real manufacturing capability shows up. Not all precision is created equal.