Xpeng’s factory in Zhaoqing. The leading electric vehicle manufacturer has invested heavily in automation. Image credit: South China Morning Post

For the last twenty years, the West’s mentality toward China has been simple (and initially, accurate): Cheap labor. Endless factories. Container ships stacked with low-cost goods heading toward Los Angeles and Rotterdam.

The reality began shifting 15 years ago and this perspective is completely obsolete in 2026.

China’s economy is not collapsing, and it is not standing still. It is changing shape, and the transformation is happening in the most important place an industrial economy can change, the factory floor.

If you take nothing else away from this post, note that in 2024-25, China was responsible for 54% of the world’s industrial robotics installation, a rate about 10x as high as the US.

The Robot Surge

In the past decade, China has quietly become the largest adopter and implementer of industrial robots on Earth. According to the International Federation of Robotics, roughly half of all new industrial robots installed each year now go into Chinese factories. In several recent years, China installed more robots than the United States, Japan, Germany, and South Korea combined.

This statistic tends to surprise people in the United States because it clashes with a deeply ingrained image of China as a place defined primarily by cheap labor. That image was always incomplete, but today it is actively misleading.

The story of Chinese manufacturing is no longer one of vast armies of low-wage workers assembling products for export. Increasingly, it is a story about automated production lines, machine vision systems, and robotic factory cells running 24 hours a day.

Far from abandoning manufacturing, China is simply upgrading.

The End of the Cheap Labor Era

For roughly 40 years, China’s growth followed a powerful but relatively simple formula.

Hundreds of millions of workers moved from rural areas into coastal production cities like Shenzhen, finding jobs in factories for all kinds of goods- initially textiles and domestic products, then electronics. Exports surged, infrastructure expanded, and the country became the hub for global manufacturing.

However, the conditions that made this possible have changed:

1. China’s working-age population has begun to shrink.

2. Wages in the country’s coastal manufacturing hubs have risen dramatically compared with the early 2000s (anecdotally quadrupling in the Guangzhou-Dongguan-Shenzhen area in the past decade alone).

3. At the same time, the massive property sector that once drove domestic growth has become a drag on the broader economy.

These pressures are forcing a transition.

Rather than relying on endless labor supply and real estate expansion, China is trying to push its economy toward higher productivity, more advanced manufacturing, and greater technological self-reliance. Automation is central to that effort. Robots allow factories to maintain output even as labor becomes scarcer and more expensive. More importantly, automation raises the ceiling on what those factories can produce.

The country that once dominated global manufacturing through scale is now attempting to dominate it through technology.

A brief aside: It’s important for westerners to realize that there are basically two versions of China. Urban professionals in China make around ¥150,000 annually, while workers in rural China make closer to ¥50,000 annually (USD $1 = CNY ¥6.90 in March 2026). This is reflected in per-capita GDP varying wildly by province, such as USD $7k in rural Gansu versus $35k in Beijing and $25k in Jiangsu province (north of Shanghai). These differences are more significant than than what we see in California ($102k) versus Alabama ($39k) in the US.

Not Collapse, But Maturation

Many Western analysts, including commentators like Peter Zeihan, have argued that China faces an inevitable economic collapse driven by demographics, debt, and political rigidity.

There is truth embedded in those concerns. China’s demographic outlook is difficult, its debt levels are high, and the unwinding of the property sector will likely weigh on growth for years, but decline and collapse are not the same thing- and necessary reinvention builds a robust future.

A China growing at four percent annually while rapidly expanding automation, energy production, and advanced manufacturing capacity is not a collapsing economy, it’s a maturing one. Expansion slows, but industrial sophistication becomes a deadly economic tool on its own.

In fact, many of the industries driving China’s next phase are precisely the ones expected to define the global industrial economy over the coming decades. Electric vehicles, battery manufacturing, solar energy systems, industrial robotics, and advanced manufacturing tools are all areas where China has built enormous scale.

China is no longer competing on cost, they’re competing at the technological frontier of manufacturing.

The U.S.–China Paradox

This transformation is unfolding against the backdrop of one of the strangest geopolitical relationships in history. To state the obvious, the US and China have both an unprecedented economic partnership and somewhat unprecedented strategic rivalry. Can you imagine if the Soviet-US relationship was just as distrustful but also totally economically linked?

American companies remain deeply embedded in Chinese manufacturing ecosystems, relying on dense supplier networks that took decades to build. Yet Washington is increasingly attempting to reduce dependence on China through tariffs, export controls, and industrial policy.

The semiconductor sector sits at the center of this struggle. Companies like Taiwan Semiconductor Manufacturing Company and Nvidia represent critical nodes in the global computing infrastructure that both countries rely upon.

The United States is trying to restrict China’s access to the most advanced chips and manufacturing equipment. China, in turn, is investing heavily in domestic capabilities while simultaneously pushing forward in other parts of the industrial stack.

Automation sits right in the middle of this contest. The more automated China’s factories become, the less vulnerable the country is to labor shortages or rising wages. At the same time, a highly automated manufacturing ecosystem is extremely difficult to replicate elsewhere, because it depends on deep supplier networks, specialized equipment, and years of accumulated industrial knowledge.

The Question That Matters

Much of the public debate about China’s future revolves around GDP rankings. Commentators argue about whether China will surpass the United States as the world’s largest economy (outright, not PPP), or whether its growth rate will fall to three percent, but these debates miss the deeper shift now underway.

The real question for the next decade is not which country grows slightly faster in any given year. It is which country controls the automated industrial base of the twenty-first century.

The nation that dominates robotics, batteries, power electronics, and advanced manufacturing will shape the physical infrastructure of the global economy. These systems determine where products are built, how quickly supply chains adapt, and which countries maintain industrial leverage in times of crisis.

China appears to understand this very clearly.

Whether its strategy ultimately succeeds is uncertain. Demographics, debt, and geopolitics will continue to complicate the path forward.

But one thing is already visible on factory floors across the country: China’s next great leap forward will not be powered by cheap labor, it will be powered by robots.

Subscribe now

MYTH

Americans won’t work in heavy industry.

The more accurate myth is not that Americans don’t want to work in factories, it’s that Americans won’t tolerate productive discomfort. This belief is usually delivered with a shrug, as if it were some immutable cultural flaw, proof that the country has simply “moved on.” It’s convenient, comforting (for some), and mostly wrong.

Americans will tolerate difficult, repetitive, physically demanding work when the trade is honest. They always have. They did it in mines, mills, foundries, auto plants, shipyards, and machine shops for generations. What they will not tolerate is work that feels extractive, stagnant, and designed to disappear. This distinction matters. Americans, like all humans, weigh their options and take the best path that optimizes for long and short-term income, comfort, agency, and flexibility.

Stating that Americans don’t want to work in factories without acknowledging that Americans will always choose more comfortable office work if the two pay the same is just silly. The collapse of American factory employment was not a mass cultural refusal to work with one’s hands, it was a rational response to jobs that increasingly offered low pay, low dignity, little progression, and no future— and which were increasingly outsourced.

The deeper truth is that Americans were never really given a choice. Manufacturing didn’t hollow out because workers opted out en masse, it hollowed out because capital opted out first. Production was moved offshore at scale essentially in pursuit of margin protection. Entire ecosystems of skills, suppliers, and institutional knowledge were treated as interchangeable inputs, not strategic assets. When plants closed, workers didn’t “choose” service jobs because they preferred them, they took what was left. Companies optimized for shareholder value, and that was that.

None of this is uniquely American. The same pattern played out simultaneously across Europe, and we are watching it unfold in real time in China. Young Chinese workers, when given alternatives, are leaving factory floors just as American workers did a generation earlier. Manufacturing is being pushed further down the cost curve into Southeast Asia, while China races to automate and move up the value chain. Even the much-discussed 996 culture is less a cultural preference than a transitional phase, tolerated because options are constrained, not because it is desirable.

This is the uncomfortable implication of the myth. People everywhere respond similarly to incentives, dignity, and opportunity. Americans were not uniquely lazy, Chinese workers are not uniquely industrious, and factory work is not uniquely unattractive. What differs is whether a society treats production as a dead end to be minimized, or as a strategic function worth investing in.

MYTHOS

Industry is citizenship.

Long before it was an economic category, industry was a form of belonging. In the American imagination, to make things was not merely to earn a wage, it was to participate in the project. Detroit didn’t just assemble cars, it assembled mobility, scale, and modern life. Pittsburgh steel wasn’t just material, it was the skeleton of a rising power. Milwaukee’s machine shops, Superiorland’s iron and copper, Texas oil, California’s peculiar fusion of invention and production, these were not disconnected episodes. They were regional expressions of a shared belief that building tangible systems mattered, and that doing so anchored a nation in reality.

This is what often gets lost when manufacturing is reduced to a spreadsheet problem. Industry shapes rhythms, identities, and expectations. It dictates how towns grow, how skills are passed down, how people understand their place in the world. Culture follows production more often than we like to admit. Music, language, humor, even politics bend around what a society spends its days doing. When you remove the act of making from daily life, you don’t just change the economy, you thin out the culture.

It’s worth comparing the culture that arises under such circumstances. Anecdotally, the difference between old-school smelter workers in Cudahy, Wisconsin or Saginaw, Michigan and modern electronics or textile factory workers in Saigon or Guangdong is much less significant than the difference between either of them and modern service/tech employees essentially anywhere in the US. Making physical products changes us.

Anyone who has spent time in Asia’s new industrial heartlands can feel this immediately. There is a gravity there, a sense of forward motion and shared effort that is difficult to fake. It is not romantic, and it is not gentle, but it is real. Cities are oriented around output. Skills compound quickly. Young people see pathways, not just jobs. What is striking is not that this energy exists there, but how recently it existed in the West, and how quickly it moved.

America has not lost this entirely. Pockets of it remain, in advanced manufacturing, in defense, in semiconductors, in specialized machine work and key verticals. But the mythos has weakened. Production is no longer treated as a civic act, something that confers dignity and responsibility, but as a temporary necessity to be optimized away. That shift did not just hollow out factories, it hollowed out a shared sense of contribution.

The danger is not that America forgot how to make things. It’s that it began to forget why making things mattered at all.

The Path Forward is Physical

Offshoring and the pivot toward a data and service economy have not been uniformly bad for Americans as individual workers. In many cases, compensation rose. Work became cleaner, safer, more flexible. The average white-collar professional gained comfort, mobility, and optionality that would have been unimaginable in the mid-20th century. It would be dishonest to deny that.

But what worked for some individuals did not work cleanly for the nation.

The strategic cost of offshoring was not simply lost jobs, it was lost leverage. As Chris Miller has argued, manufacturing matters less because it employs large numbers of people, and more because it anchors power. Production determines who controls supply chains, who sets standards, who learns fastest, and who can respond when systems are stressed. When manufacturing leaves, those capabilities leave with it, quietly at first, then all at once. It’s remarkable how fast this happens— the average working life is about 40-50 years. A shift can happen in half that time.

This is why the consequences only become obvious during crises. Pandemics, wars, trade shocks, and technological inflection points expose the fragility of an economy that consumes far more than it produces. A country that no longer builds at scale must negotiate from dependence rather than strength, no matter how sophisticated its financial markets or software ecosystems may be.

The answer is not autarky, and it is not nostalgia. It is a deliberate rebalancing toward physical capability. That means deeper vertical integration in strategically sensitive areas, especially upstream processing of materials and components where external choke points are most dangerous. It means aggressive automation, not as a substitute for labor, but as a force multiplier that collapses cost advantages and accelerates learning. And it means rebuilding credibility around production as a career, with clear advancement paths, technical prestige, and tangible upside for the people who make complex systems work.

It’s worth asking whether the US can survive if we’re bifurcated into venture capitalists and baristas. An exaggeration, but rooted in uncomfortable truth. An economy without a broad, skilled, productive middle is not just unequal, it is unstable, and frankly upstream of a lot of other challenges we see today and that could worsen in the future.

The path forward is not purely digital. It is physical, embodied, and grounded in the act of making. Without that, prosperity becomes thinner, power becomes borrowed, and culture itself begins to drift.

Addendum

The US seems to be jointly and apolitically waking up to the reality that we need to make things. The growth of hard tech in the US cannot be overstated- firms like NVIDIA, Tesla, SpaceX, Apple, Anduril, and countless others have led the way.

This all points to less of a manufacturing resurgence, certainly not one led by tariffs, but more of an acknowledgement of strategic shortcomings and subsequent realignment. I expect a lot of ebb and flow and discussion around manufacturing— but it’s increasingly back in the conversation, for now.

A few bright spots and complications in this next phase of the journey:

• Intel’s Silicon Heartland initiative involves the construction of a new semiconductor fab in New Albany, Ohio, targeting 2030 for completion (after experiencing delays). There are reportedly challenges in identifying customers for these chips, likely in part due to a premium on manufacturing costs. Similar challenges have been reported with TSMC’s new Arizona fab (10% cost increase vs Taiwan).

• Andreesen Horowitz’s American Dynamism practice is funding civic-minded founders with visions to improve our national defense and infrastructure.

• Semiconductor manufacturing growth in the US has created the need for 300,000 trained employees by 2030.

High-bandwidth memory (HBM) chip, a big focal point of 2025. Image credit: Wevolver.com

Electronics is no longer a single industry. It’s a stack of interdependent bottlenecks, shaped as much by geopolitics and industrial policy as by Moore’s Law, and actively scaling with the size and shape of the global economy. In 2025, that reality became unavoidable. The global electronics market continued to grow, but the story of the year wasn’t volume, it was constraint. Power, memory, packaging, export controls, and capital allocation quietly determined who mattered and who didn’t.

Recap of 2025

By most estimates, the global semiconductor market finished 2025 somewhere just north of $700 billion, depending on who you ask and how they count. That number matters less than where the profits pooled. AI infrastructure pulled capital and talent upstream, while large swaths of consumer electronics remained soft. The result was an industry that looked healthy in aggregate but deeply uneven underneath.

If the $700B figure surprises you, you’re not alone. Semiconductors are volumetrically a smaller industry than most expect, and their impact pertains more to being a bottleneck for other industries. Their competitive landscape, similarly, is shaped less by purely money and more by a blend of coordination, specialization, and deep expertise over decades- which is why it’s been so hard for new entrants to succeed, let alone gain dominance.

Events that Shaped the Year

If there was one reminder in 2025 that “boring chips” are no longer boring, it was Nexperia. What should have been a routine story about analog and discrete components instead became a lesson in how geopolitics now reaches deep into ownership structures, packaging lines, and export permissions. The takeaway wasn’t about one company, it was that mid-stack semiconductors are now strategic assets by default, whether executives like it or not, and more bluntly, that China is willing to use this class of chip as leverage in economic negotiation. This will force trading partners to reassess exposure and contingency planning.

2025 also clarified that memory, particularly HBM (high bandwidth memory), is no longer a supporting actor. It is the plot. As AI training and inference workloads scaled, memory bandwidth and yield became the pacing constraint, not compute. HBM became a critical input that feeds data into NVIDIA (and other) GPUs, which provide parallel-path computing that drives AI development. The industry spent the year reorganizing itself around this reality, with capital, roadmap priority, and customer relationships flowing toward whoever could ship reliably at scale.

Finally, 2025 closed with a subtle but important shift in tone. For the first time, credible forecasts began treating a trillion-dollar semiconductor market as plausible rather than aspirational. That number matters less as a target than as a mindset, it changes how governments plan subsidies, how suppliers justify capacity, and how executives think about risk.

It’s worth noting that the 2025 semiconductor market likely looked very different to different companies depending on their specific market and product categories. Most consumer electronic manufacturers likely dealt with symptoms of component excess due to oversupply of trailing-edge nodes, commodity logic chips, and generally consumer-oriented chips, and relatively lower demand. By stark contrast, AI-related memory chips, components requiring advanced packaging, and power chips all remained quite constrained due to growth around AI.

Winners and Losers

The winners of 2025 were unsurprising in hindsight. Companies exposed to AI infrastructure, advanced memory, substrates, and packaging captured outsized value. The losers weren’t necessarily poorly run, they were simply tied to slower replacement cycles or exposed to regulatory friction. In this environment, neutrality is no longer safe.

Winners

SK Hynix - The big surprise winner of 2025 was South Korea’s SK Hynix, who committed to the high-bandwidth memory trend early, gambling that AI needs would outpace consumer memory demand. They now own over 50% of the HBM market.

2025 market cap growth (USD): from $90B to $300B (+$210B)

Micron - The story here is much the same. Micron made the pivot a bit later, but surpassed Samsung Electronics in 2025 to become the second largest HBM player (21% market share).

2025 market cap growth (USD): from $94B to $320B (+$226B)

NVIDIA - If it feels obvious to include them here… that’s why they made the list. The greatest success story of the post-COVID era, NVIDIA has essentially reshaped the industry. Their parallel-compute GPUs have extreme memory needs- most of this is driven by demand from AI LLM developers. In this sense, SK Hynix, Micron, TSMC, and ASML are all critical partners, but are increasingly shaped by NVIDIA’s requirements.

2025 market cap growth (USD): from $3.4T to $4.6T (+$1.2T)

TSMC - The Taiwanese giant is another obvious choice as they’ve become a byword for the pure-play foundry model and frankly the semiconductor industry itself. They have worked around strategic vulnerabilities and constructed a large-scale fab in Arizona, remaining an industry cornerstone for the foreseeable future.

2025 market cap growth (USD): from $1.2T to $1.7T (+$500B)

ASML - The most upstream firm on the list and another slam-dunk annual winner. ASML got a lot of press in 2025. As light was shone on the industry and its many bottlenecks, none was more evident than that created by ASML, whose EUV machines will continue to define this era of chips.

2025 market cap growth (USD): from ~$350B to $500B (+~$150B)

Note: I’m using market cap here not as a single metric to capture total performance, but as a proxy for where investors believe durable leverage lives as of Jan 2026.

Losers

In an industry that is experiencing hyper-growth, it’s hard to call anyone a loser. “Losing” in semiconductors in 2025 looks more like flat growth, lack of execution on a consistent vision, waning brand power, or an inability to capitalize on trends. For this reason, I won’t name many names- we’re not picking stocks here.

Intel is perhaps the best example. They’re notably in a challenging position, offering foundry services to compete with TSMC while also competing with their customers by building their own internally-designed chips. This has created very public market challenges. In the past 5-10 years, they’ve lost the lead in process and look to recapture this technical capability while also reorganizing strategically. If they were the Detroit Lions, we’d call it a rebuilding year. A longer horizon will be fairer to them.

GlobalFoundries is a good parallel example. A pure-play foundry and once closer in ambition to TSMC, they’ve intentionally focused on trailing-edge and specialty nodes and not compete in the extremely capital-intense leading-edge. As demand softened and excess inventory crept up in consumer-tier chips, companies like GlobalFoundries were impacted in an outsized way.

Subscribe now

Expectations for 2026

I’m not especially good at short-term prediction- it’s just not in my wheelhouse, and I find that most forecasts end up being fairly incorrect. Instead of forecasts, let’s look at broad constraints and trends. For 2026, three forces look likely to shape outcomes regardless of how individual products perform.

First, high-bandwidth memory will continue to behave like a strategic commodity. Supply will improve, but qualification, yield, and packaging capacity will keep it scarce enough to confer leverage. This will ripple outward into substrates, foundry allocation, and even data center design. Packaging may in particular end up being a particular vulnerability if new OSATs need to be opened. While these used to be lower-margin and cost-sensitive, they’re now complex differentiators. Geopolitics may become a bigger factor in site choice.

Second, portfolio triage will accelerate. Micron’s shift away from low-margin consumer lines toward data center and AI workloads isn’t an anomaly, it’s a preview. In 2026, more companies will quietly abandon legacy businesses that no longer justify capital or engineering attention. As we saw this year (Nexperia), these consumer lines can be massively critical if generally less newsworthy. I predict that the industry will continue to bifurcate into two classes of firms- leaders will play in the capital-intensive space that serves AI, government-backed projects, and aerospace/defense, and the gap between these and dedicated consumer-oriented manufacturers will solidify.

Third, industrial policy stops being a backdrop and becomes the weather. Tariffs, subsidies, localization rules, and security reviews won’t fade, they’ll harden. Electronics supply chains in 2026 will be designed assuming friction, not efficiency. There’s a lot that can be said around this, but in general, dual-sourcing becomes the bare minimum for survival. Companies that don’t have the SCM resources to deal with this will outsource their strategic vulnerabilities to firms that offer constant analysis, safety stock, and even development and kitting.

That’s a Wrap

This piece is less about predicting winners and losers or even specific trends. It’s inevitable that we take our current position each year and basically project it into the future with no knowledge of disruptions.

I’d like this to become an annual tradition that forms a baseline against which all my other thoughts will be weighed. What’s become clear is that semiconductors are the engine, but they’re hardly the only fascinating topic. In 2026, I’ll wade into frontier tech, hard tech, and the confluence of a lot of factors that shape these industries.

By the end of this year, we’ll have discussed rockets, rare earth minerals, and wearable tech, and my next State of the Industry will be all the richer for it.

Buckle up people, this is a long one. I’d say about once a day on LinkedIn I see a new iteration on a diagram designed to maximize likes, something mapping out “the semiconductor industry” and comparing companies like TSMC, Nvidia, and ASML by market cap. It’s not inherently bad, but it does a poor job of defining the flow of the industry, and it’s unclear who the diagram is directed at and what value it really brings.

It’s my hope that that’s not the case for this post. The Hidden Layer is all about pulling back the curtain on the inner workings of the world’s most critical industry for consumers and people who are not in the know.

This aims to be a touchpoint that I can reference in the future, a definitive piece that explores the depth and complexity of the world of semiconductors, microchips, and electronics, layer-by-layer, and explains WHY they are critical and what possible bottlenecks exist heading into 2026.

The purpose of this is to give readers a baseline to digest the news through- when you see an event (political, weather, or otherwise), you’ll have a framework for what it means and how it affects car prices, or delays on the new iPhone.

The classic example is an earthquake hitting Taiwan. Yes, it’s a huge concentration of semiconductor manufacturing, especially front-end fab sites. Yes, it could swiftly bring the global economy to a halt, or at least create some sleepless nights for supply chain operators (as it did in April 2024).

But as we’ve seen with China’s restrictions on rare earth mineral exports, or recent friction between the Netherlands and China over Nexperia, a maker of comparatively less cutting-edge chips (which are essential in many products), it takes a lot less than an earthquake to disrupt supply chains, and it could happen at literally any point in the process.

The stack begins with raw materials and moves its way up to finished goods. Let’s dig in.

Layer 0: Raw Materials

0.5A: Material Refining

0.5B: Semiconductor Tools

0.5C: Chip Design

1: Island of Chips (Front-End)

2: Packaging and OSAT (Back-End)

3: Copper Nervous System (PCB)

4: Megafactories (EMS)

5: Middlemen of Stability (Distributors)

6: Architects of the Final Product (OEM)

Wrapping Things Up

Subscribe now

Metro Mining barge with bauxite in Queensland, Australia. Image credit: Mining.com.au

LAYER 0: Dirt with Lawyers - Raw Materials & Processing

1. What this layer does:
This is where the electronics world begins, long before a wafer or a circuit board or even a transistor comes into view. The raw inputs of modern technology start as ordinary minerals pulled out of the ground and then pushed through layers of chemical processing until they become something usable. Examples:

• Quartz becomes metallurgical silicon, which becomes polysilicon.

• Bauxite becomes aluminum.

• Rare earth ores become magnets and phosphors.

• Lithium brines become battery-grade lithium salts.

None of this looks like electronics yet, but without this layer nothing further up the stack can exist.

2. What it touches:
Upstream is geology and mining. Downstream is everything else. Semiconductor fabs (layer 1) need hyper-pure polysilicon- arguably the most important but as we’ll see, these are all points of failure. Packaging houses (layer 2) need gold wire, copper, and molding compounds. PCB (layer 3) factories need copper foil, fiberglass, resins, and specialty chemicals. Even the cleanrooms depend on fluorinated gases, solvents, etchants, and ultrapure inputs coming from this layer. The entire electronics industry is downstream of a surprisingly small group of materials suppliers.

3. Who the major players are:
In polysilicon you have Hemlock Semiconductor in Michigan, Wacker Chemie in Germany, OCI in South Korea, and a collection of Chinese giants that dominate solar-grade output. In rare earth refining, China controls most of the midstream conversion steps, from ore to separated oxides to magnet precursors. For specialty chemicals you see names like Shin-Etsu, JSR, Merck, Linde, and Entegris. It’s an upstream world, but it is full of quietly important companies that determine what is possible further down the line.

4. Why this layer is a bottleneck in 2025:
The bottleneck is not the minerals themselves, it is the processing. Mining is distributed. Refining is not. Most critical materials pass through a handful of countries, and often one country. Polysilicon refining is energy-intensive and expensive, so capacity sits where electricity is cheap and environmental rules allow it. Rare earth separation is dominated by China because they built the refining ecosystem early and at scale. The geopolitical climate has turned this into a pressure point. Export controls on gallium, germanium, and graphite in 2023–2024 showed how easily upstream constraints can ripple into entire industries. Lead times for adding new refining capacity run into years, sometimes decades. The world keeps discovering new uses for these materials, but the ability to refine them has not kept pace.

5. Example disruption:
When China tightened export controls on gallium and germanium in 2023 and then expanded restrictions to include materials like graphite in 2024, it was a reminder that upstream pressure can move silently and suddenly. These are not glamorous elements, but they feed into RF semiconductors, photovoltaics, power electronics, and batteries. Overnight, companies from Europe to Japan were forced to assess their exposure, track inventories, and search for alternate refining sources. Nothing physically broke, no factory collapsed, and yet the entire downstream system felt the shock.

That’s the nature of Layer 0- it’s not as flashy as a natural disaster, but it’s the ground on which all the rest is built.

LAYER 0.5: The Hidden Architects — Materials Refiners, Toolmakers, and Chip Designers

Layer 0.5 is the semi-hidden region where upstream leverage quietly concentrates. These are companies that never touch your phone or laptop directly, yet everything you own depends on them. They sit between nature and the fab, and each one creates constraints that ripple through the whole system.

LAYER 0.5A: Materials Refining - Hemlock, Wacker, OCI

1. What this sublayer does:
We discussed this a bit before, but this is more of a focus on the processing. This is where minerals stop being rocks and start becoming usable inputs for semiconductor manufacturing. Refiners take quartz, metallurgical-grade silicon, or complex chemical precursors and push them through high-temperature, high-purity, high-energy conversion steps until they reach “electronics grade.” This is where the purity jumps from parts per million to parts per trillion.

2. What it touches:
Upstream: mining and chemical feedstocks.
Downstream: wafer manufacturers, solar ingot makers, and specialty chemical companies.
They are the hinge between geology and semiconductor physics.

3. Who the players are:
Hemlock Semiconductor (US), Wacker Chemie (Germany), OCI (South Korea), plus Chinese solar giants who dominate lower-purity output of polysilicon. A handful of companies globally can make true semiconductor-grade polysilicon, which makes this layer surprisingly concentrated.

If we think about Hemlock Semiconductor, in my backyard of Michigan, that’s America’s primary source of high-purity polysilicon. It’s one location, effectively in the middle of nowhere (not too far from Dow Chemical’s HQ). If that goes down, it’s a national security event for the US.

4. Why it is a bottleneck in 2025:
Refining is capital-intensive, energy-intensive, and slow to scale. Environmental rules and electricity prices determine where capacity sits. China’s dominance in many refining steps, plus export controls on materials like gallium and germanium, revealed how upstream chemical chokepoints can instantly affect multiple downstream sectors.

5. Example disruption:
China’s 2023–2024 gallium and germanium controls were a reminder that nothing here needs to “break” physically for the world to feel a shock. A policy change at the refining layer can squeeze RF chips, photovoltaics, and power devices overnight.

EUV (extreme ultraviolet) lithography machines, such as those made by ASML, are used to etch patterns onto silicon wafers. Image credit: Azo Nano/Shutterstock

LAYER 0.5B: Semiconductor Tools - ASML, Tokyo Electron, Lam Research, Applied Materials

1. What this sublayer does:
These firms build the machines that make chips possible. Lithography, deposition, etch, metrology- everything in a fab runs through tools built by only a few companies. Famously, ASML’s EUV scanners are the most complex machines humans have ever manufactured.

2. What it touches:
Upstream: precision optics, lasers, specialty metals, and mechatronic subsystems.
Downstream: every wafer made at advanced nodes.
Without this sublayer, Layer 1 cannot even exist.

3. Who the players are:
ASML (EUV lithography), Tokyo Electron (deposition, etch), Lam Research (etch), Applied Materials (deposition, CMP), KLA (metrology). Each has near-monopoly control over an essential step.

4. Why it is a bottleneck in 2025:
Lead times for EUV tools stretch past a year. There is no alternative supplier. Export controls shape who is allowed to buy certain tools. A supply hiccup in a single optical subsystem in Veldhoven (Netherlands) can delay an entire fab’s ramp. This layer is one of the deepest strategic chokepoints in the global economy, and although ASML has become something of a household name in past years, it’s still one of the least known bottlenecks.

5. Example disruption:
The 2022–2023 semiconductor equipment export restrictions to China showed how geopolitics can essentially redraw the map of who may manufacture at what node. ASML’s technology has become a geopolitical artifact, not just an industrial tool.

At this point you may also see an emerging trend- export restrictions (usually by the US and EU) on goods to China- usually equipment -are countered by export restrictions on goods from China (usually raw materials). The goal of the US in this case is to restrict its own dependency on China for chips and the ability of China to gain more leverage. The goal of China is to counter that, maintain its moat, and to retain its position as the world’s manufacturer.

LAYER 0.5C: Chip Design - Nvidia, AMD, Apple Silicon, Qualcomm

1. What this sublayer does:
Chip designers are the architects of computation. They do not manufacture chips themselves. They create the blueprints- GPUs, CPUs, modems, accelerators — that fabs turn into physical silicon. Modern design is an ecosystem of RTL engineering, verification, EDA tooling, IP licensing, and architectural breakthroughs.

This layer (within this framework) pertains only to fabless designers, who will typically rely on TSMC or Samsung for manufacturing (as we’ll see next layer). There are many players who both design and manufacture chips, such as Infineon, Texas Instruments, STMicroelectronics, and others. These firms often have a specialty in one category, such as analog, automotive, or power electronics, with some overlap between.

2. What it touches:
Upstream: EDA tools (Cadence, Synopsys), IP cores (Arm), memory interface standards, packaging roadmaps.
Downstream: fabs, OSATs, board-level integrators, entire industries built around their silicon.

3. Who the players are:
Nvidia in AI accelerators. Apple in mobile SoCs. AMD in CPUs and GPUs. Qualcomm in wireless. Even though they outsource manufacturing, these companies effectively determine the performance frontier for consumer and enterprise electronics.

4. Why it is a bottleneck in 2025:
The bottleneck here is architectural leadership. Demand for compute has exploded past the capacity of fabs to scale linearly. The result is that companies like Nvidia exert outsized influence on hardware design, software ecosystems, supply allocation, and manufacturing priorities. A single architecture decision- like memory bandwidth or packaging type -cascades through every downstream layer from HBM suppliers to EMS firms.

5. Example disruption:
Nvidia’s 2023–2024 GPU allocation crisis showed how a design-layer bottleneck can look identical to a manufacturing shortage. Demand outpaced packaging capacity, and the entire AI supply chain scrambled- not because fabs failed, but because the architectural center of gravity shifted faster than the physical world could adjust.

It’s worth noting that there are potential alternatives to traditional computing and digital ICs. I’ve covered Normal Computing’s development of the first thermodynamic chip, and there are other frontiers being explored, such as ultra-low power chips and reversible computing. These have the potential to upend the paradigm, and their impact would be most felt at this layer. I’ll look to dig into these in a future post.

LAYER 1: The Island of Chips - Front-End Semiconductor Manufacturing

1. What this layer does:
This is where sand finally becomes computation. Front-end fabs take hyper-pure polysilicon, slice it into wafers, and then use an almost theatrical sequence of depositions, etches, ion implants, and photolithography steps to carve billions of transistors into silicon. A modern fab is closer to a physics experiment than a factory. Everything here happens at atomic scales, in cleanrooms that scrub the air a thousand times cleaner than a hospital, inside tools that cost as much as skyscrapers. This is the layer that turns raw materials into the digital brain of every device we own.

2. What it touches:
Upstream, they rely on the refined materials from Layer 0- polysilicon, specialty gases, photoresists, copper, and exotic chemicals produced by a handful of firms. They also rely on equipment monopolies like ASML’s EUV tools, Tokyo Electron deposition systems, and Lam’s etch platforms. Downstream, every single electronics layer depends on what comes out of a fab. If Layer 0 is geology with lawyers, Layer 1 is physics with billion-dollar price tags. Nothing downstream can substitute or work around a shortage of chips at a given node.

3. Who the major players are:
TSMC is the gravitational center of advanced logic. Samsung sits beside them at the cutting edge, with Intel rebuilding its footing. For mature nodes you see companies like GlobalFoundries (USA), UMC (Taiwan), and SMIC (China). Each has its own process technology and capabilities, but only a few- TSMC and Samsung in particular- can build the high-performance processors that drive AI accelerators, smartphones, and data centers. The geographic clustering is stark. So much of the world’s leading-edge logic sits on one island and one peninsula.

4. Why this layer is a bottleneck in 2025:
The fragility here comes from concentration. Advanced logic is overwhelmingly produced in Taiwan and South Korea. EUV lithography- the tool you need to make cutting-edge chips- is produced by a single Dutch company (ASML, as discussed). The cleanroom ecosystem is so specialized that recreating these fabs elsewhere is a decade-long challenge. Even small disruptions matter. A brief power sag can ruin wafers worth tens of millions of dollars. More broadly, demand for high-performance computing has ripped past the industry’s pace of capacity additions. Even with CHIPS Act incentives, the world cannot simply “build another TSMC” on demand. The barrier is not money, it is know-how, supply chains, and time.

5. Example disruption:
When the Hualien earthquake hit Taiwan in 2024, it jolted the world’s most important semiconductor cluster. TSMC paused operations, evacuated cleanrooms, and inspected tool stability. Some wafer cycles were interrupted mid-process, which effectively destroys them. Even a temporary outage forced customers across automotive, compute, and industrial sectors to run recalculations on supply assumptions. Nothing catastrophic happened, but the episode was a reminder of how thin the margins are. If you shake the island where the world keeps its most advanced fabs, you feel it everywhere else. It takes only hours of downtime to ripple through quarters of production planning downstream.

A microchip after packaging and OSAT.

LAYER 2: The Bottleneck Next Door - Advanced Packaging and OSAT

1. What this layer does:
Once a wafer leaves the fab, it is still useless. The dies have no protection, no electrical contacts, and no way to be handled, tested, or assembled. Advanced packaging is the stage where individual chips are cut from the wafer, bonded to substrates, wired, encapsulated, tested, sorted, and prepared for real-world use. This includes everything from basic wire bonding to cutting-edge 2.5D and 3D integration, chiplets, interposers, and high-bandwidth memory (HBM) stacks. If the front-end is physics, packaging is precision carpentry and electrical engineering fused together. It is the bridge between silicon and the rest of the world.

Traditionally, packaging and OSAT has been a lower-cost activity that was outsourced to countries that could support volume at a lower cost. However, this has changed somewhat in recent years as packaging has become more complex to suit needs for mobile devices and data centers.

2. What it touches:
Upstream, they receive bare dies from fabs, often still fragile and untested. They rely on substrates, molding compounds, underfill materials, copper, gold wire, and tooling that has its own supply constraints. Downstream, they feed into PCB assembly, module integrators, EMS firms, and ultimately OEMs. Packaging is the connective tissue. If a fab makes the brain, packaging provides the skull, nerves, and skin that allow it to function.

3. Who the major players are:
The giants here are ASE Group, SPIL (Siliconware), Amkor, JCET, and Powertech. On the more advanced end, TSMC and Samsung have moved aggressively into high-end packaging to support chiplet architectures and AI accelerators. This is also where you see major substrate manufacturers- Ibiden, Shinko, Unimicron — and where cutting-edge packaging increasingly lives inside the same geographic footprint as the fabs they serve.

4. Why this layer is a bottleneck in 2025:
Historically, packaging was treated as low-margin, low-glamour work and was offshored accordingly. But that strategic assumption collapsed the moment AI demand exploded. AI chips require massive bandwidth and thermals that only advanced packaging can deliver. Suddenly, OSAT capacity became the new scarcity, especially for substrate-based 2.5D and HBM-intensive designs. The world realized that it could build more wafers, but without somewhere to package them, the bottleneck simply moved downstream. Add geographic concentration- much of the world’s advanced packaging sits in Taiwan- and you get a layer that looks mundane on paper but is now one of the most strategically constrained parts of the electronics stack.

Packaging and OSAT is a particularly unique vulnerability for national defense for countries not in Asia (where it is heavily concentrated, more than any other piece of the equation- in part due to the cost-sensitive nature of packaging). The US in particular only has about 5% of global OSAT volume within its borders.

5. Example disruption:
When the Hualien earthquake shook Taiwan in 2024, the immediate headlines focused on fab downtime. But packaging houses were hit too. Facilities paused operations for inspections, tool stability checks, and safety protocols. Even temporary disruptions here create subtle but far-reaching ripples: chip testing delayed, substrate deliveries pushed out, downstream module builds paused. These delays do not always show up in the news, but they show up in lead times. This is the quiet vulnerability of Layer 2. It exists in the seam between silicon and the supply chain, and when that seam shifts, the entire product schedule shifts with it.

Subscribe now

LAYER 3: The Copper Nervous System - PCB and PCBA Manufacturing

1. What this layer does:
If chips are the brains, printed circuit boards are the nervous system. A PCB is a stack of copper and fiberglass that routes power and signals between components. PCBA- printed circuit board assembly -is the step where components are placed, soldered, inspected, and transformed into functional modules. This is where electronics truly take shape. Every device, from a laptop to an EV to a dishwasher, passes through this layer. It is ordinary compared to fabs and packaging, but it carries the full weight of the industry. No board, no product.

There is intentionally some overlap between this step and the next, where EMS firms will typically own PCBA (but not PCB) manufacturing, and often higher-level assembly as well.

2. What it touches:
Upstream, this layer depends on copper foil, resins, laminates, solder paste, stencils, and a stream of components from distributors and semiconductor houses. It also relies on precise machinery: pick-and-place robots, reflow ovens, x-ray inspection, SPI and AOI systems. Downstream, PCBs and PCBAs flow into EMS factories, automotive modules, telecom base stations, consumer electronics assembly lines, and thousands of industrial applications. Nearly everything that gets built touches this layer, even if only for a few seconds on a pick-and-place line.

3. Who the major players are:
On the fabrication side, you see companies like Unimicron, Compeq, Tripod, and Ibiden. In Southeast Asia, Vietnamese and Thai facilities have become essential for medium- and high-volume builds. On the PCBA side, the ecosystem ranges from massive EMS firms handling assembly internally to regional specialists like Jabil (which does both), Pegatron, Sanmina, and a constellation of mid-sized assemblers that keep industrial and automotive sectors running. It is a sprawling, distributed network, but with geographic clusters that matter a lot.

4. Why this layer is a bottleneck in 2025:
PCB fabrication is heavily concentrated in China, with Taiwan, South Korea, Thailand, and Vietnam handling much of the rest. Environmental rules, cost structure, and decades of accumulated know-how mean there is very little capacity in the US or Europe. PCBA is equally clustered in Asia, especially for high-volume consumer devices. These clusters make the system efficient but fragile. A flood, a power outage, or a lockdown in just one region can delay millions of units downstream. And unlike chips or packaging, PCBs are low dollar-value items with long lead times and limited fungibility. You cannot easily replace a specific board with a similar one. The design is the design. If it is delayed, everything above it waits.

5. Example disruption:
During Vietnam’s COVID shutdowns in 2021, factories producing everything from PCBs to assembled boards were forced into extended closures. This hit global electronics in a way that surprised many observers. Suddenly, laptop makers, automotive suppliers, and even consumer appliance manufacturers found themselves short of basic PCB assemblies. These were not cutting-edge semiconductors. They were everyday boards. Yet the absence of a $3 PCB halted $500 devices. It revealed a structural truth: the industry has built extraordinary resilience around chips and components, but far less around the humble board. It’s the layer you only think about when it stops.

Smartphone PCBA. The PCB (previous layer) is visible underneath the components. Image credit: Best Technology Co., Ltd. (bestpcbs.com)

LAYER 4: The Megafactories - EMS and Final Assembly

1. What this layer does:
Electronics Manufacturing Services (EMS) firms are the places where all the upstream complexity finally comes together. Chips, substrates, PCBAs, displays, batteries, connectors, plastics, antennas- everything lands on their production lines and becomes a finished product. These factories run at staggering scale. Phones, laptops, routers, wearables, automotive modules, medical devices, and industrial controllers all move through EMS lines in some form. If Layer 3 provides the nervous system, Layer 4 builds the body. This is where supply chains stop being abstract and become physical objects.

I’ve created this with a bit of overlap- typically “EMS” will refer to much of the previous step, PCBA assembly, and sometimes add batteries or other electronics, or even build the finished good.

2. What it touches:
Upstream, they depend on every layer you have discussed so far: semiconductors, packaging houses, PCB fabricators, distributors, mechanical suppliers, and an ocean of components. They also rely on logistics networks to keep thousands of part numbers flowing on a just-in-time cadence. Downstream, they feed into OEMs and brands who add the final software, testing, customer packaging, and global distribution. An EMS firm is the central junction where engineering, procurement, forecasting, and operations collide.

3. Who the major players are:
Foxconn (Hon Hai) towers over the landscape, assembling everything from iPhones to servers. Pegatron, Wistron, Luxshare, BYD Electronics, Jabil, Flex, Celestica, and Sanmina form the rest of the global backbone. Some OEMs like Samsung or Dell do part of their own assembly, but for the most part, EMS firms are the industrial muscle behind modern electronics. They run campuses the size of small cities, with hundreds of thousands of workers, robot cells, SMT lines, injection molding, machining centers, subassembly plants, and in-house logistics.

4. Why this layer is a bottleneck in 2025:
Scale is both their strength and their fragility. High-volume assembly is concentrated in a few countries- China above all, with Vietnam, Thailand, Malaysia, Mexico, and India growing but not yet equivalent. Labor availability, local policy shifts, energy costs, and public health disruptions all hit this layer hard. OEMs often have less visibility into EMS capacity constraints than they assume. A line cannot be conjured out of thin air during a surge. Certifications, tooling, line design, worker training, and quality systems take months to establish. When allocation hits here, it is not components being rationed, it is physical space and human bandwidth.

5. Example disruption:
In late 2022, Foxconn’s Zhengzhou facility- the world’s largest iPhone assembly site -faced extended lockdowns and labor unrest. Output dropped sharply. Apple issued rare public warnings about delays. What looked like a local labor-health event became a global electronics story within days. Phones, accessories, and downstream retail inventories were affected. This was the EMS bottleneck in full view: a single megafactory, specializing in a single family of products, is a point of enormous leverage. When it stumbles, the world notices immediately. Unlike chip shortages, which are invisible to consumers, EMS failures show up on store shelves.

To me, the EMS is the sweet spot of electronics that’s just 1 tier below OEMs, easy for consumers to understand (the PCBA is large and visible and tangible), and it’s also a piece that explains how so much of the modern world comes together. PCBAs are not a commodity, but they lean more that way than other parts of the ecosystem. EMS firms hold a large part of the market, and are the backbone on which modern electronics are built.

LAYER 5: The Middlemen of Stability - Component Distribution

1. What this layer does:
Component distributors are the quiet infrastructure behind the electronics economy. They sit between manufacturers and the thousands of OEMs, EMS firms, and engineering teams who need a constant flow of parts. On paper, they buy and resell components, but in reality they perform a dozen hidden functions: credit extension, inventory buffering, lifecycle management, demand smoothing, obsolescence forecasting, kitting, and design-in support. They are a mix of bank, logistics firm, and consultant. Without them, the system would be volatile to the point of failure.

2. What it touches:
Upstream, distributors work with semiconductor manufacturers, passive component makers, interconnect suppliers, and the entire long tail of electronics suppliers. Downstream, they feed every type of customer: consumer electronics, EVs, industrial controls, aerospace, medical devices, telecom, and all the EMS and PCBA plants that sit one layer below OEMs. They synchronize information and material in a way no OEM can replicate individually. In shortages, they become the rationing authority. In gluts, they become the shock absorber.

3. Who the major players are:
Globally, the key names are Arrow, Avnet, and Future. There are also smaller firms that specialize in high-mix, low-volume (helpful for prototyping), such as Digi-Key, Mouser, and Farnell, as well as some that focus specifically on Asia. Regionally, there are dozens of smaller specialists servicing automotive, aerospace, or industrial niches. These firms collectively manage staggering catalogs- millions of SKUs with constantly shifting lead times, lifecycle stages, and demand curves. They are the ecosystem that translates chaos into availability, or at least predictability.

If everything up to this point represents engineering, design, and manufacturing excellence, distributors represent the market side, which is unbelievably complex in its own right.

4. Why this layer is a bottleneck in 2025:
Demand is volatile, product lifecycles are shortening, and semiconductor supply remains uneven across nodes. When the bullwhip effect hits- late visibility, panic ordering, double-booking, cancellations -distributors are the ones who absorb the shock. But they are not infinitely elastic. Once inventory tightens, allocation rules kick in and customers discover how little visibility they truly have into the upstream system. The complexity of modern BOMs, the pace of product updates, and the reality of geopolitical constraints mean that even large OEMs cannot forecast accurately without distributor partnership. The bottleneck is not only physical components, it is information precision.

5. Example disruption:
During the 2020–2022 automotive chip crisis, demand forecasting errors converged with pandemic shutdowns and soaring semiconductor lead times. Automakers cancelled orders early and then tried to reorder at massive volume six months later- right when consumer electronics demand was surging. Distributors found themselves caught between fabs with no spare capacity and OEMs asking for allocations they could not fulfill. The result was cascading shortages across everything from microcontrollers to power MOSFETs to simple analog parts. A single missing $1 automotive-grade IC could halt a $40,000 vehicle. This was the bullwhip made visible, and it revealed how much stability depends on distributors being able to read the market before the market reads itself.

Smartphones, tablets, and laptops are prime examples of products shipped by OEMs that are heavily PCBA and component-dependent. These are classic examples, but there are infinite others.

LAYER 6: The Architects of the Final Product - OEMs and System Integrators

1. What this layer does:
OEMs transform assembled electronics into finished, coherent products. They decide what gets built, what performance matters, what technologies are worth integrating, and how thousands of components turn into a device people recognize and trust. They orchestrate design, firmware, mechanical engineering, regulatory compliance, quality assurance, supplier selection, and forecasting. A phone, an inverter, a medical device, a tractor, a router -all of them begin as decisions made at this layer about architecture, cost, and time to market. This is where technology is shaped into something that can be sold.

2. What it touches:
Upstream, OEMs depend on every preceding layer: distributors for parts, EMS firms for assembly capacity, PCB and substrate houses for boards, OSATs for packaged chips, fabs for silicon, materials refiners for inputs, and sometimes even raw material markets when risks rise. Downstream, they touch retail channels, regulatory regimes, after-sales support, and entire end-use industries. The OEM is where technical choices become business consequences. A thermal limit on a chip, a substrate shortage, a PCB footprint change, or a late firmware fix can alter the entire production calendar.

3. Who the major players are:
Apple, Tesla, Dell, Cisco, Siemens, Bosch, Sony, and thousands of industrial and automotive OEMs that never trend on social media but build critical infrastructure. In many ways, the most influential OEMs today are those who blend hardware with software ecosystems- Apple with its end-to-end silicon strategy, Tesla with over-the-air updates and in-house chips, Cisco with integrated networking stacks. These firms do not simply buy parts, they shape the direction of entire supply chains through their architectural decisions.

When you think of OEMs, it can be helpful to think beyond the classic examples (i.e. Apple and Samsung)- every consumer electronic, appliance, vehicle, or piece of industrial equipment is produced by an OEM.

4. Why this layer is a bottleneck in 2025:
The fragility here comes from complexity and decision speed. OEMs forecast demand imperfectly, often with long feedback loops. They redesign products faster than suppliers can adjust. They push for cost reductions while demand for performance rises. They often do not have deep visibility beyond tier 1 suppliers (although this is changing with new tools), which means disruptions at layers 2, 3, or 5 arrive as surprises even when upstream signals existed for months. And because OEMs place the largest and most visible orders, their forecasting errors can propagate downstream as panic, bullwhip, or overcorrection. This layer is not fragile because of shortage alone, it is fragile because the entire stack is shaped by decisions made here under uncertainty.

5. Example disruption:
When the automotive chip crisis unfolded, OEMs realized that their long-favored “just-in-time” procurement strategies offered little resilience. A single microcontroller shortage shut down assembly lines across the US, Europe, and Asia. Factories producing complex products were halted not by labor or metal, but by the absence of a chip small enough to fit on a thumbnail. In another sphere, Apple’s struggle to manage its assembly footprint during the 2022–2023 period demonstrated how even the most sophisticated OEM can be exposed when a megafactory stumbles. Redesigns, alternative sourcing, and emergency engineering changes all flowed from disruptions they did not fully see coming. The vulnerability here is not technical, it is systemic: the OEM sits at the intersection of everything, and when that intersection clogs, there is no easy detour.

Wrapping Things Up

You’ve made it to the end, so let’s keep things short and sweet. When people think of “electronics”, they probably picture something like a circuit board assembly, and to most consumers, the actual manufacturing processes and materials are completely unknown. That’s understandable, because of the level of complexity is so high that it may as well be meaningless to most people.

But in a world that is increasingly defined by electronics, both economically and in how we live our daily lives, I think it’s essential for consumers to at least have a primer or a framework to filter the news through.

Hopefully you took that away from this lengthy post. I’m sure I’ll refer back to it as we continue to explore industry bottlenecks and tech advances, and these 7 layers (plus 3) will become part of an ongoing series exploring this industry.

Subscribe now

Christmas trees being baled and loaded at Omni Farms in West Jefferson, NC. Image credit: Chris Keane/REUTERS/Alamy/Atlas Obscura.

I’m in Shenzhen today (a few weeks before Christmas, at time of writing), in the Nantou Ancient City district. This place is as lively as I’d expect for China’s third largest city any time of the year. But today, with bursts of red and green, faux-snow-covered Christmas trees, and Santa and Rudolph ornaments for sale in every third shop, it feels even busier and more jovial.

I’m surprised by this. Mainland China doesn’t strike me as a place to celebrate Christmas, and there are essentially zero foreigners here today. Traveling with me, my Hong Kong friend seems to read my mind and interjects that Chinese people look for any reason to celebrate and shop- Christmas just fits the bill for December. I’m not entirely convinced, as shops are going above and beyond with the decor, but that’s beside the point.

She muses about the evolution of the holiday in China, and as she reflects on her Christmas traditions over the course of her life in Asia, my mind naturally starts to map out the surprisingly complex supply chain of Christmas. It’s my first December living abroad, and I’ve uncovered two surprising Yuletide connections between the US and China, revolving around two tiny but high-impact towns.

Oregon’s Christmas Trees

Christmas trees are everywhere in Hong Kong right now- there’s a big one in front of the Mandarin Oriental Hotel in Central that I’ve passed a dozen times, assuming it and the many like it came from a forest up in Hunan or Guangdong. I couldn’t be more wrong. My friend informs me that every year, she would order her tree from a local retailer selling them directly from Oregon. That’s 10,600 km away.

This little fact shocked me, especially because of the obvious- trees are alive, and cross-Pacific freighter trips take 2 weeks. The answer is apparently refrigerated containers, which help preserve freshness substantially.

Sitting on the edge of Oregon’s Willamette Valley, the town of Estacada (pop. 5,000) has quietly become a global powerhouse and was designated the “Christmas Tree Capital of the World”. The town and region happen to have perfect growing conditions, long-standing family farms, and export infrastructure that allows thousands of trees to move through cold chain without losing freshness.

Most years, the global Christmas tree market breaks down roughly like this: the United States and Canada dominate production of natural firs, Europe relies heavily on Denmark and Germany, and Asia imports nearly all of its real trees from North America. Within the US, Oregon and North Carolina are the giants (my home state of Michigan coming in third, as of 2022), with Oregon leading by volume thanks to its enormous Douglas fir and Noble fir output. The trees often take seven to ten years to mature, and because the harvest cycles are predictable, farms can plan around global demand.

This creates a small but fascinating seasonal artery: Oregon farmland to Hong Kong port terminals, onward to flower markets and apartment lobbies. It’s a reminder that even something as nostalgic as the scent of a pine tree is occasionally a product of cross-Pacific logistics.

For those curious, this HK-based retailer is selling Oregon-sourced Noble firs for USD $280-612 depending on tree size.

Christmas decorations in Yiwu, China. Image credit: Denis Staunton/The Irish Times.

Yiwu, the City That Predicts Christmas Trends

Opposite the ocean from Oregon and a ways north of Hong Kong is another comparatively small city: Yiwu, in Zhejiang province, population a mere 1.86 million. At face value, a smaller manufacturing hub a few hours south of Shanghai is not cause for any special attention. But Yiwu is different.

A few weeks ago, another good friend explained to me that this otherwise un-noteworthy city has predictive powers. The actual practicality of the “Yiwu Index” is heavily debated, but many believe that because the city is home to manufacturers of a wide range of household goods and novelty items, quick surges in demand for certain specific goods might predict political and economic trends (i.e. a big uptick in yellow vests pointed to an increase in protests in France circa 2018).

Walking through markets in Yiwu, Christmas unfolds six months before anyone in the West thinks about it. If Americans suddenly want pastel ornaments, metallic garlands, or inflatable Santas with LEDs, this city knows first. Orders flow in long before retailers publish trend reports, and the assembly lines become an early indicator of where US holiday taste is drifting.

Factories here don’t just make Christmas goods. They act like cultural sensors. They absorb demand signals months ahead of time and convert them into millions of units. It’s a strange quiet power. Workers who have never been to the US can tell you what the “hot color palette” will be that year by watching what Western buyers request in July. Yiwu exported $244M of Christmas goods globally this year. Apparently, sustainability and modular decor is big in 2025, particularly items that can be used from Halloween through the Christmas season (possibly like adding Santa hats to those 12-foot skeletons?).

If Oregon exports the feeling of Christmas, Yiwu exports the look of it. Between them, a holiday aesthetic travels the world long before people put lights on their houses.

I like these little stories because they’re great examples of supply chains in our everyday life, and some of the strange complexity that delivers otherwise completely normal products to our homes. It’s not all about semiconductors and frontier tech- sometimes it’s nice to appreciate how a couple of out-of-the-way cities and towns an ocean apart both have unexpected impact on Christmas for people the world over. A forest in Oregon makes a Hong Kong living room smell like Christmas. A cluster of workshops in Yiwu decides which ornaments end up on a tree in Milwaukee.

As the year winds down, I’m wishing all Hidden Layer readers a Merry Christmas and an enjoyable time with your families. Cheers.

Normal Computing’s thermodynamic processor, which looks to reduce AI energy usage by a factor of 1000. Image credit: Normal Computing

New York-based Normal Computing just made a bold claim: they’ve built the world’s first thermodynamic processor.

Not a GPU, but a new kind of analog computing core that replaces Boolean logic, using heat and noise as signals, not problems to be fixed. This sounds a little out there, but if the physics holds and the architecture scales, it could represent the most meaningful shift in computation since the dawn of the microprocessor in 1971.

We can pump the brakes a little- this has just been designed and “taped out”, meaning the design has been finalized and sent to the fab. There’s a long road of testing ahead before commercialization.

What’s the Big Deal?

This isn’t about marginal efficiency gains. Normal Computing is claiming three orders of magnitude improvement in energy use for probabilistic workloads, especially in machine learning and AI inference.

Their chip is analog, probabilistic, and grounded in the core laws of thermodynamics. Basically, instead of forcing bits into binary states using precise voltages and timing, it utilizes the thermal noise (heat energy) that is created as a probability distribution, which it uses in computation. The result is increased speed and decreased energy consumption for certain specific (but critical) cases.

If we look at a very rough analogy, consider the use of statistical modeling in demand forecasting to determine production or inventory. Historically, planners may have settled on an average or small range, but ignored the broader distribution. Modern best practice is to not only consider the distribution, but to actively track error- which was previously considered noise.

Just as this noise can be used to determine safety stock levels (improving the quality of future forecasting), the thermodynamic processor aims to use thermal noise to model probability distributions, which are in turn used in computation, particularly for AI.

Background

Before diving into what’s new, it’s worth briefly revisiting the foundational concepts:

• Moore’s Law: An empirical observation first noticed by Gordon Moore, stating that the number of transistors on an IC doubles about every 18-24 months. This has essentially held true since the late 1960s, fueling most of the growth in computing power, but it’s coming up against hard economic and physical limits.

• Landauer Limit: The minimum amount of energy required to erase one bit of information; a thermodynamic floor for digital logic.

• Thermal noise: In traditional computing, it’s treated as interference. Here, it’s the raw material of computation.

• Boltzmann distribution (below): A statistical model that describes the likelihood of particles occupying various energy states. Normal Computing’s chip reportedly uses this principle in hardware form.

All of this ties into the broader slowdown of Moore’s Law and the growing physical and economic limits of transistor-based scaling. If computation can’t keep getting cheaper and faster the old way, new approaches are on the table.

How does this work?

The Boltzmann distribution is essential for understanding the key principle here. In reality, there’s a big difference between useful noise or randomness and “wrong” randomness. In our analogy, think of this as the difference between capturing true raw demand data and skewed, biased, or data that’s otherwise affected in some way.

In the processor, thermal noise is governed by physical factors- voltage, current, materials, and circuit parameters. This random noise may be different from the probability distribution that is needed for the computation. Mapping between these is the biggest technical challenge.

The Boltzmann distribution is a statistical law defining how physical systems engage with and explore energy states- essentially the distribution of probabilities of these different energy states that arises. If the computational problem follows the same structure that arises in the system, there are no issues. However, if the two are even slightly different, there will be bias in the samples. Therefore, this new class of processor will likely require some form of calibration to account for this.

Normal Computing controls this a couple of different ways: controllable parameters (such as temperature / noise amplitude), and hybrid digital-analog control loops (digital systems watching the analog core and making adjustments).

Implications for Electronics

Who uses this?
The first applications will likely be in edge AI and energy-constrained environments- this would include autonomous drones, mobile robotics, IoT inference, and probabilistic simulations.

What changes?
If this architecture delivers even half of what’s promised, it would shift multiple parts of the semiconductor and electronics stack:

• Chip design: New analog-core architectures, distinct from existing digital logic.

• Packaging: Thermodynamic chips may have distinct thermal management, shielding, or calibration needs.

• Fabrication: Foundries accustomed to digital logic may need to adapt to analog-optimized processes.

• Sourcing: Analog components have different suppliers, tolerances, and visibility challenges than digital logic.

Supply chain implications

• Are sourcing teams prepared to track the availability and compliance of analog thermodynamic cores?

• Are fabs equipped to handle the yield variability of these components?

• Could demand for conventional GPUs decline in narrow workloads?

Practical Takeaways

1. Track Normal Computing’s early results closely: If tape-out and test data support their claims, the ripple effects could be rapid, especially in AI-adjacent markets.

2. Begin mapping analog sourcing capabilities: Identify vendors, materials, and packaging partners who support analog thermodynamic workflows.

3. Flag legacy assumptions: Many EMS workflows, sourcing risk models, and supplier networks assume deterministic digital logic. Those assumptions may not hold.

The impact of this is likely going to be targeted and specific to a few key use cases, let alone industries, at first, but it’s hard to overstate just how enormous a change this could be for computing. The design has been solidified and Normal Computing has moved from 0 to 1, and while there may be scaling challenges ahead, this first step is the largest one.

Cai Mep-Thi Vai Port. Image credit: Vietnam Pictorial

It’s clear to any visitor that Vietnam is defined as much by its future as by its past. As soon as I touched down from Hong Kong to Saigon’s central Tan Son Nhat Airport, I hopped in a Grab and was taken through the land of smiling Tiger Beer ads, a continuous but pleasantly humming stream of motorbikes, and nods to the modern country’s founder.

Hong Kong is not a pessimistic place, but Ho Chi Minh City felt distinctly warmer and more youthful by comparison. I could wax poetic about the egg coffees and French-era architecture, but instead, I’m here to talk about the evolution of the city and country more broadly, and that became apparent only on the latter half of my stay, traveling 4-6 hours a day for factory visits. I’ve seen a fair cross section of Asia, and Vietnam firmly falls somewhere “in-between”, but to the surprise of no one, growing quickly.

In the industrial zones outside Ho Chi Minh City, half-built factories line the highways beside billboards for Samsung and Foxconn. There’s an unmistakable hum of progress here: trucks carrying reels of copper, workers on scooters wearing matching company vests, and the steady rhythm of construction cranes edging toward the skyline. Vietnam feels like a country in motion, but it’s also one learning to manage the growing pains that come with that speed.

As it unfolds, there are some serious challenges ahead, but Vietnam is already well into its play to become an electronics manufacturing powerhouse, and it hinges on one key factor.

Bigger Picture

Governed by the Communist Party of Vietnam since 1975, the country enjoys a stability and uniform directionality that few others regionally can boast (excepting China). It’s likely for that reason that they’ve seen such intentional investment in an export economy as companies and countries broadly explore a China+1 strategy. Vietnam forms a solid analog to China, with a young and well-positioned workforce, ports, and a willingness to bring business in.

As of time of writing, the country has a population of just under 102 million people (growing at 0.6% annually), with Ho Chi Minh and Hanoi each containing 20 million people in their metros. GDP sits at $476B USD (7.1% annual growth rate).

That growth is built on decades of gradual reform. In the 1990s and 2000s, Vietnam’s economy was powered by textiles and footwear, its labor advantage attracting companies like Nike and Adidas. In recent years, those same industrial parks have pivoted toward electronics. Where once stood garment lines, now stand cleanrooms assembling smartphones and circuit boards.

Foreign direct investment (FDI) topped $36 billion in 2023, with South Korea, Japan, and Singapore leading the charge. China, once viewed primarily as a rival, is now a complex partner, investing in logistics and energy projects even as companies relocate manufacturing out of it. For multinationals, Vietnam has become the defining hedge in a world that no longer tolerates single-source dependency.

Bottleneck

Despite its momentum, Vietnam’s next leap forward hinges on something deceptively simple: infrastructure.

While industrial parks have sprung up across the southern provinces, moving goods between them remains painfully slow. The 1,600 km stretch from Hanoi to Ho Chi Minh City lacks a unified high-speed corridor, and the country’s logistics costs remain around 16-17% of GDP, significantly higher than regional and global competitors.

There are several huge bright spots:

• North–South Expressway: This will connect Vietnam, from Lang Son on the Chinese border to Ca Mau on the Gulf of Thailand.

• Long Thanh International Airport: Expected to open in 2026 as Saigon’s new international airport (Tan Son Nhat will revert to domestic-only).

• Expanded deepwater ports at Cai Mep–Thi Vai: The major port near Saigon has been upgraded to Vietnam’s largest international transshipment port.

• Ho Chi Minh City Metro: Launched in December 2024 with 1 line, 650 km total rail planned.

These projects will be transformative, but they’re running against the clock. Each delay keeps lead times high and dulls the country’s competitive edge.

Vietnam’s infrastructure paradox is that its industrial growth outpaces the physical means to sustain it. Trucks queue for hours outside ports; container traffic competes with everyday commuters; power outages, while less frequent than a decade ago, still threaten production schedules. Even within Ho Chi Minh City, metro construction has stretched into its second decade, a reminder that the path from ambition to reality can be uneven.

The government has launched the “National Logistics Action Plan,” which achieved its target to cut logistics costs to 16% of GDP by 2025. New expressways are linking industrial corridors from Binh Duong to Dong Nai, and Long Thanh Airport, when completed, will handle up to 100 million passengers and five million tons of cargo per year. If Vietnam can align its infrastructure push with its manufacturing boom, it could create a feedback loop of investment and capability rarely seen outside East Asia’s great economic miracles.

Implications for Electronics

Vietnam’s electronics boom will depend on whether its supply chain can match its enthusiasm. Assembly operations are growing fast, but most critical components, such as semiconductors, sensors, and PCBs, still flow through China, Taiwan, or South Korea.

Until local suppliers scale up and logistics efficiency improves, Vietnam will serve as an assembly hub rather than a manufacturing ecosystem, essentially a new way station in an already complex system.

This distinction matters. Assembling smartphones or laptops is labor-intensive but low-margin. The true economic power lies in the upstream stages: chip fabrication, component production, and design. Today, those remain concentrated in economies with deep capital markets, skilled engineers, and research infrastructure. Vietnam’s advantage lies in execution, not invention, at least for now.

That said, the government’s partnerships with Intel, Samsung, and Amkor, plus tax incentives for component production, signal a push to move up the value chain. The country’s success may rest on whether its roads, ports, and policies can connect its newfound ambition with reliable execution.

To climb further, Vietnam will need more than factories- it will need engineers, data infrastructure, and trusted trade policy. The country has launched dozens of technical training programs and partnerships with Korean and Japanese firms to upskill its workforce. If successful, this could replicate the South Korean model of state-guided industrial learning. If not, Vietnam risks being trapped in the assembly tier as automation erodes its cost advantage.

Upshot

Vietnam stands at a familiar crossroads: a country on the verge of greatness but bound by the limits of its roads and runways. The optimism is pragmatic. Every kilometer of new highway or fiber-optic cable expands the country’s ability to participate in the digital economy. Every partnership inked with a multinational teaches local managers how to compete on a global stage. Progress here is incremental, not explosive, but it’s unmistakably forward.

I’m high on Vietnam. The country feels noticeably vibrant and full of optimism in a way that most are not, in 2025. It’s unusual that a pivot to higher-tech manufacturing would require solving problems that by comparison are relatively lower-tech or that have already been solved elsewhere, but I view that as encouraging. The incentives are aligned and initiatives are underway, and I think this is just the beginning.

The egg coffees can stay, but I’m looking forward to eliminating the 2-hour commutes.

I’m early in my career. Fairly early in the arc of understanding how chips move across oceans, how risk creeps into digital supply chains, and how strategy gets distorted between boardrooms and loading docks.

But I’m not passive about it.

The Hidden Layer is a space for exploring the complex, often overlooked systems shaping our world, and the systems behind those systems. I’m especially focused on semiconductors, supply chains, and the digital transformation happening across segments of the electronics industry.

What this is not: A collection of polished takes from someone who has it all figured out.

What this is: A lab notebook from someone who’s deep in the trenches, trying to understand one layer at a time.

Right now, I work in electronics sourcing on a global scale, and lead a semiconductor supply chain transformation project across Asia. I’m also working on an academic case study around component-level sourcing risk. Alongside that, I’m pursuing the MITx MicroMasters in Supply Chain Management, building skills and perspective as I go.

The goal of this newsletter isn’t to make claims or push hot takes. It’s to think in public, to explore big questions, and to bring clarity to topics that rarely get the attention they deserve.

If you’re someone who enjoys systems thinking, supply chain complexity, or the hidden architecture of how technology moves through the world, you’ll feel at home here.

This is just the beginning. I’ll be going deeper from here, one layer at a time.

See you in the stack,
Neil