For decades, the basic philosophy of hardware manufacturing was relatively simple: keep inventories low, forecast demand, negotiate aggressively with suppliers and make as many components as possible as cheaply as possible. The model known as just-in-time manufacturing helped companies avoid tying up capital in warehouses full of parts while allowing manufacturers to respond quickly to changing demand.
That model has not disappeared, but it is no longer sufficient for the most strategically important parts of the technology industry. The pandemic exposed the fragility of concentrated supply chains, geopolitical tensions have made cross-border dependencies more complicated, and the artificial-intelligence boom has created demand for components whose production cannot simply be increased overnight. Semiconductor supply chains now face constraints not only in leading-edge wafers, but also in high-bandwidth memory, advanced packaging, substrates, power and other upstream inputs. Gartner describes this as a broader restructuring of semiconductor supply chains as AI investment changes where the industry's bottlenecks sit.
The result is a quiet but important change in strategy. Technology companies are increasingly trying to buy resilience before they need it.
They are reserving manufacturing capacity years ahead, entering long-term agreements with critical suppliers, helping finance new factories and redesigning products so that a shortage at one supplier does not bring an entire system to a halt. In other words, the supply chain is no longer simply something a company manages after designing a product. It is becoming part of the product's architecture.
From Buying Components to Buying Capacity
The clearest example can be found in AI infrastructure.
A company building a conventional electronics product can often respond to a component shortage by finding another supplier. That becomes much harder when the missing component depends on a highly specialized manufacturing process with years of lead time.
Advanced AI accelerators illustrate the problem. A modern accelerator is not simply a processor manufactured on a leading-edge wafer. It can depend on high-bandwidth memory, advanced packaging, substrates, interposers, testing capacity and increasingly sophisticated optical and networking components. A shortage anywhere in that chain can prevent the finished system from shipping.
Research from Epoch AI estimated that the four largest AI chip designers consumed around 90% of global CoWoS advanced-packaging capacity and HBM supply in 2025, despite using only a relatively small share of advanced logic-die production. That illustrates why adding more wafer capacity alone does not necessarily solve the problem.
The bottleneck can simply move further down the chain.
That is changing how major technology companies think about procurement. Instead of asking a supplier, "How many units can you deliver next quarter?", companies increasingly want to know, "How much capacity can we guarantee over the next several years?"
Nvidia's latest supply-chain investments provide a particularly clear example. In October 2026, the company announced agreements involving advanced packaging, substrates, testing and optical-engine manufacturing. It said its arrangements with several substrate suppliers would support a 2.5-fold increase in capacity over three years, while Amkor plans to increase certain advanced-packaging capacity by more than 300% by the end of 2028. Nvidia is also providing a $1.1 billion capacity-reservation advance to testing company KYEC for new US facilities.
That is not ordinary component purchasing.
It is closer to industrial capacity planning.
The Factory Is Becoming Part of the Product
This represents a deeper change in hardware strategy. In the old model, engineers designed the product and procurement teams later figured out how to source its components. Today, the availability of those components can influence the engineering decision itself.
If a particular chip is available from only one supplier, for example, selecting it may create a supply-chain dependency that survives for the entire life of the product. A theoretically superior component can therefore be strategically inferior if the company cannot obtain enough of it.
Hardware engineers increasingly have to ask questions that once belonged almost entirely to procurement: Can another supplier make this component? Can the firmware accommodate a substitute? Can the PCB accept a different part? Does the package require a particular substrate? What happens if the supplier's factory is closed for six months?
This is where resilience becomes an engineering discipline.
The goal is not necessarily to make every component dual-sourced. That would be expensive and, in many cases, impossible. Instead, companies identify the components whose failure would have disproportionate consequences and design alternatives into the system before production begins.
A $1 component that can stop a $30,000 server from shipping may deserve more strategic attention than its price suggests.
The AI Chip Is Only as Strong as Its Weakest Link
The semiconductor industry offers an increasingly obvious demonstration of this principle.
For years, much of the attention around advanced computing focused on transistor density and leading-edge manufacturing nodes. The race was about who could produce the smallest, fastest and most power-efficient logic chips.
AI has complicated that picture.
High-bandwidth memory has become critical because accelerators require enormous amounts of data to be moved between memory and compute. Advanced packaging has become essential because increasingly complex systems combine multiple dies and memory stacks in extremely dense configurations. Substrates and interconnects have become strategically important because the finished package cannot function without them.
The result is a supply chain in which adding capacity in one part may accomplish little if another part remains constrained.
PwC's 2026 semiconductor analysis describes HBM as a critical component of modern AI infrastructure and notes that new HBM capacity can still be constrained by packaging infrastructure, interposers and other specialized processes.
That changes the definition of manufacturing capacity.
A company does not really have capacity to produce a million AI accelerators if it has enough wafers for a million chips but only enough advanced packaging or HBM for half a million finished units.
Effective capacity is determined by the bottleneck.
Resilience Starts Before the Factory
That is why the supply-chain strategy increasingly begins during product design.
A resilient hardware company may qualify alternative power-management chips, memory devices, controllers and other supporting components before the first production run. It may design boards that can accommodate more than one supplier or build software abstractions that make switching between components less painful.
The objective is not redundancy for its own sake. It is optionality.
If one supplier experiences a production problem, the company wants the ability to move a meaningful portion of its production elsewhere without redesigning the entire product.
This can make products slightly more expensive to engineer. Alternative components have to be tested. Firmware may have to support different devices. Manufacturing lines may need additional validation. Engineers may have to sacrifice a small amount of theoretical performance to gain sourcing flexibility.
But the calculation has changed.
The question is no longer simply, "Which component is cheapest?"
It is increasingly, "What is the cost of being unable to buy it?"
China Plus One Is Becoming a Hardware Strategy
Geography is another part of the same transformation.
For years, the concentration of electronics manufacturing in China and semiconductor production in East Asia was justified largely by efficiency. Dense supplier ecosystems, skilled manufacturing workforces, mature logistics networks and enormous production volumes created powerful economic advantages.
Those advantages have not disappeared. But companies are increasingly uncomfortable with having critical production concentrated in a single country or region.
The result is the rise of strategies commonly described as "China Plus One" or broader geographic diversification. Companies are expanding or qualifying manufacturing capacity in countries including Vietnam, India and Mexico while governments in the United States, Europe, Japan and elsewhere are subsidizing domestic and allied semiconductor production.
The objective is not necessarily to abandon existing Asian supply chains. That would often be economically unrealistic. Instead, companies want another option if tariffs, export controls, conflict, natural disasters or political decisions disrupt the primary route.
The semiconductor industry is already seeing substantial investment in geographically distributed manufacturing. The Semiconductor Industry Association says announced investments are expected to significantly increase US fabrication capacity over the coming decade, including new capabilities in advanced logic, memory and advanced packaging.
The important point is that these factories are being valued not only for their economic output but also for the strategic redundancy they create.
The Supply Chain Does Not End With the Supplier
There is another problem that traditional procurement models often struggle to capture: the supplier's supplier.
A company may have a perfectly reliable contract manufacturer while remaining vulnerable because that manufacturer depends on one source of a particular substrate, chemical, mineral or specialized machine.
Modern semiconductor production is an enormous web of dependencies. A shortage of a relatively obscure material several tiers upstream can eventually prevent a sophisticated chip from being manufactured.
Recent research from McKinsey and the Semiconductor Industry Association highlights precisely this problem, finding that important vulnerabilities in the US semiconductor ecosystem can exist several layers upstream from the companies that directly manufacture chips. Critical minerals, engineered materials and equipment subcomponents can become hidden points of failure.
This is why supply-chain visibility is becoming a technology problem of its own.
Large manufacturers increasingly need systems capable of mapping suppliers beyond Tier 1 and identifying where the same raw material, factory or geographic region appears repeatedly throughout their network.
The goal is to discover the dependency before it becomes a shortage.
AI Is Turning Supply Chains Into Competitive Weapons
The most important change may be that supply chains are no longer merely defensive.
They are becoming competitive advantages.
If two companies have equally capable AI chips but one has guaranteed access to HBM and advanced packaging while the other is waiting in line, the first company can bring systems to market faster.
If two server manufacturers have similar products but one can produce in multiple regions while the other depends on a single factory, the first may be better positioned when trade restrictions or disruptions occur.
And if one company has already reserved scarce manufacturing capacity while its competitors are trying to obtain it after demand has surged, the advantage can become enormous.
TrendForce reported in April that AI competition was already tightening capacity across advanced logic and 2.5D/3D packaging and that Nvidia had moved early to secure volumes across wafers, CoWoS packaging, materials, substrates, PCBs, HBM and SSDs.
This is why the modern hardware race increasingly looks like a race for capacity allocation.
The winner is not necessarily the company that can design the best chip. It may be the company that can guarantee enough of everything else to actually manufacture and deploy it.
Resilience Comes With a Price
None of this is free.
Dual-sourcing increases qualification costs. Maintaining factories in several countries can be more expensive than concentrating production in one location. Reserving capacity means committing capital before demand is certain. Domestic production can carry higher labor, construction and operating costs.
There is also a danger in overcorrecting.
If every company builds enormous inventories, reserves excessive capacity and duplicates every supplier, the industry could replace one inefficiency with another. Semiconductor manufacturing has always been cyclical. The investments made to solve today's shortage can eventually produce too much capacity if demand slows.
The memory industry illustrates the danger. New facilities take years to build and qualify, meaning companies must make capacity decisions based on forecasts of future demand. If demand continues rising, the investment may look brilliant. If growth suddenly slows, the same capacity can contribute to another industry downturn. CSIS has warned that the large investments being made to relieve today's AI-driven memory shortage could eventually recreate the conditions for another oversupply cycle.
Resilience therefore cannot mean simply "build more."
It means building the right options.
The New Question for Hardware Executives
The traditional hardware supply chain was optimized around efficiency. How cheaply can we manufacture this product? How little inventory can we carry? How quickly can suppliers respond?
The new model asks a different set of questions.
How quickly can we switch suppliers? Which component could stop the entire product? How many months of capacity are effectively guaranteed? Which critical materials come from a single region? Can our board accept an alternative component? Where does our Tier-3 dependency actually begin? What happens if our primary manufacturing country becomes inaccessible?
Those questions turn supply-chain strategy into something much broader than procurement.
It becomes a combination of engineering, finance, geopolitics and corporate strategy.
That is particularly visible in AI infrastructure, where the scale of investment makes even seemingly minor supply constraints economically significant. Deloitte estimates that the global AI data-center buildout will require enormous spending over the next several years, while the underlying semiconductor system depends on a tightly interconnected chain spanning logic, memory, packaging, equipment and materials.
The companies building that infrastructure therefore have an incentive to secure the chain before the next shortage arrives.
Hardware's New Competitive Advantage
For much of the technology industry's history, supply chains were treated as something happening in the background. Customers saw the laptop, smartphone, graphics card or server. They rarely thought about the hundreds of companies and thousands of components required to produce it.
That is changing.
In a world of geopolitical fragmentation, semiconductor shortages and AI-driven demand, the supply chain is becoming part of the product's competitive architecture.
The strongest hardware companies may increasingly be those that can tolerate a broken link without stopping the entire machine. They will design for alternative components, reserve scarce capacity, diversify manufacturing locations and understand dependencies far beyond their immediate suppliers.
That does not mean the era of cheap, efficient manufacturing is ending. It means efficiency is no longer the only objective.
The new hardware strategy is about buying something that cannot easily be added after a crisis begins: time.
Time to switch suppliers. Time to move production. Time to qualify another component. Time to secure memory. Time to find another packaging line.
In the old supply chain, companies bought components.
In the new one, the smartest companies are increasingly buying options, capacity and resilience.
And as AI turns computing infrastructure into one of the world's most strategically important industries, that resilience may prove to be as valuable as the chips themselves.