Before Capacity Becomes Capacity

A new fab is only the visible end of a longer chain.

A simplified comparison instrument looks like this:

This is not a claim that every manufacturer follows this exact order.

The point is the opposite.

Different manufacturers appear to harden these layers in different sequences.

That sequence matters because it determines where uncertainty sits inside the system.

A company can commit a technology while keeping manufacturing scale open.

It can commit a building shell while delaying equipment configuration.

It can build against multi-year customer roadmaps before volume is contractually firm.

Or it can use staged public funding and technical milestones to create capability before mature commercial demand is visible.

The distinction is not “committed” versus “uncommitted.”

It is what has already become difficult to reverse.

Samsung provides the most explicit physical-layer staging mechanism in this comparison.

At Samsung Foundry Forum 2022, the company described a “Shell-First” strategy.

The logic was explicit: build cleanrooms first, irrespective of immediate market conditions, then install and configure manufacturing equipment later in line with future demand.

The strategy continued to be publicly described in 2023 as Samsung expanded manufacturing lines in Pyeongtaek and Taylor.

This is a materially different object from “building capacity.”

A cleanroom shell is already a significant physical and capital commitment.

But a shell without a fully installed process toolset preserves more configuration flexibility than a fully configured production line.

In other words:

the building can harden before the exact manufacturing configuration does.

That creates an intermediate state between an empty shell and a fully configured production line.

Samsung’s broader foundry architecture reinforces this layered model.

Its public strategy combines process technology, process design kits, the SAFE ecosystem, electronic design automation, intellectual property, design services, advanced packaging and increasingly integrated offerings across foundry, memory and packaging.

In Q1 2026, Samsung said its foundry business planned to reach full utilization of advanced-node production lines in Q2 while pursuing larger 2nm customers. Its Q2 results later reported strong U.S. customer orders and continued expansion of design wins, including 2nm HPC engagements. Samsung also disclosed very large facilities and R&D investment plans, while explicitly stating that long-range investment plans remain subject to future market and business conditions.

The resulting public architecture is therefore different from TSMC’s and Intel’s.

Samsung has historically disclosed a mechanism that allows one physical layer — the cleanroom shell — to be committed ahead of final equipment configuration.

That does not mean every Samsung expansion in 2026 uses Shell-First in the same way.

The strongest explicit descriptions are from 2022–2023.

So the evidence supports this statement:

Samsung publicly articulated a Shell-First strategy and continued to describe it in 2023, separating cleanroom commitment from later equipment commitment.

It does not support the stronger claim that every current Samsung foundry project follows that architecture.

Rapidus is not an equal peer in this comparison.

It is a boundary case.

TSMC, Samsung and Intel are operating advanced manufacturing systems with established customer bases and active production.

Rapidus is still building a new 2nm foundry capability ahead of planned high-volume manufacturing in 2027.

That lifecycle difference is too important to hide.

But precisely because the system is still being formed, it reveals another possible commitment sequence.

Rapidus installed a volume-production-ready EUV system at IIM-1 in December 2024.

Its pilot line began operating in April 2025.

By July 2025, the company had demonstrated electrical characteristics from 2nm GAA transistor prototypes.

Later in 2025, Rapidus was selected by the Japanese government as an official business operator for measures intended to support stable advanced-semiconductor production.

In 2026, the company secured large public-private funding rounds, received annual NEDO approval for its plan and budget, and continued to build process, packaging and customer-development infrastructure.

In October 2026, Rapidus announced CORE, an ecosystem connecting research institutions, EDA, IP, design and manufacturing, with the explicit goal of allowing customers to begin advanced development early.

The sequence is therefore visibly different from the mature foundries.

Public capital, physical pilot capability and technical milestones harden before mature high-volume commercial demand is established in the public disclosures reviewed.

The most defensible interpretation is not that government funding substitutes for market demand.

It is that Rapidus carries uncertainty through a staged institution-backed capability-formation process.

Annual plans, budgets, public-private capital and technical milestones become visible gates before the system reaches mature foundry economics.

The comparison does not support a ranking of which system is “more flexible.”

It supports something more useful.

Different systems appear to hold uncertainty in different places.

These are not four versions of the same decision.

They are different ways of deciding what must become irreversible now and what can remain an option until later.

Across all three core manufacturers, several important quantities remain outside the evidence ceiling.

Public disclosures do not reliably establish:

• exact binding customer volume commitments;
• customer-specific cancellation economics;
• internal hurdle rates;
• the exact point at which sunk cost becomes economically irreversible;
• a common denominator for leading-edge physical capacity;
• all options to repurpose tools, shells or process infrastructure;
• the complete timing of customer-specific design decisions.

This prevents several tempting calculations.

There is no defensible common “percentage of future capacity committed.”

There is no reliable synthetic optionality score.

There is no clean normalized cost-per-unit-of-advanced-capacity comparison.

Those numbers might look precise.

The public evidence does not support them.

A fab may take years to become visible in output, but the future starts hardening much earlier.

TSMC, Samsung and Intel do not expose the same public sequence for turning uncertain demand into manufacturing commitment. TSMC emphasizes continuous demand-coupled capacity planning. Samsung has explicitly separated cleanroom commitment from later equipment configuration. Intel has committed to node development while leaving manufacturing scale dependent on committed demand and acceptable return.

The common lesson is not that one architecture is more flexible than another.

It is that “capacity” hides a sequence of decisions.

By the time a fab appears as a completed physical asset, several earlier layers may already be difficult to reverse.

The more useful question is therefore:

This research uses publicly available primary sources from the companies and relevant formal disclosures.

Evidence is separated into:

The research does not compare process-node quality, wafer economics, transistor performance, market share or investment attractiveness.

Node labels such as “2nm” and “14A” are not treated as technologically equivalent units.

No synthetic capacity-utilization, optionality or technology-ranking metric is constructed.

Rapidus is not treated as an equal operating peer to TSMC, Samsung Foundry or Intel Foundry. It is used as a boundary case for capability formation before mature high-volume production.

TSMC

Samsung Electronics / Samsung Foundry

Intel

Rapidus