Skip to main content

Fabless vs Foundry vs IDM

Updated Jun 24, 2026 at 8:22pm

Research Draft High 1,291 words

The semiconductor industry splits into three structural business models defined by who designs the chip and who physically manufactures it. A fabless company designs chips and sells them but owns no factory (Nvidia, AMD, Qualcomm, Broadcom, Apple's silicon group). A foundry (also "pure-play foundry") owns the multi-billion-dollar fabs and manufactures chips to other companies' designs but sells no branded product of its own (TSMC, GlobalFoundries, and the contract-manufacturing arms of Samsung and Intel). An IDM — Integrated Device Manufacturer — does both: designs and manufactures its own chips in its own fabs (historically Intel, Samsung, Texas Instruments, Micron, Analog Devices). The central tension is capital intensity vs. control: fabs are among the most expensive assets in any industry, and the choice of where to sit in this stack drives a company's margins, cyclicality, balance sheet, and competitive moat. This is the most important structural distinction an investor in the sector must understand before reading any semiconductor company's financials.

The three models and how they differ

The split traces to a real technological breakthrough. In the late 1970s, Carver Mead and Lynn Conway codified standardized chip "design rules" that decoupled the act of designing from the act of fabricating — a "Gutenberg moment" that made it possible for a designer to hand off a layout to any factory that followed the rules (SemiWiki). That decoupling was inert until Morris Chang founded TSMC in 1987 as the world's first dedicated pure-play foundry, explicitly to remove the fab as a barrier to entry. TSMC was originally a joint venture in which the Taiwanese government (via the Executive Yuan Development Fund) held the largest stake — roughly 48% — alongside Philips at about 28%, with the remainder held by private Taiwanese investors (Construction Physics; Wikipedia: TSMC). Once a neutral factory existed, the fabless model exploded — by 2000 roughly 60% of TSMC's revenue came from fabless firms like Qualcomm, Broadcom, and Nvidia (SemiWiki).

The economics of each model are sharply different:

  • Fabless — asset-light. R&D and design are the cost center; manufacturing is a purchased input (a cost of goods). Balance sheets carry little PP&E. Gross margins are typically high — commonly cited in the 55–70% range, with Nvidia operating above 60% (TMT IB Guide; PatentPC). The model lowers the barrier to entry and lets the company scale capacity up or down by adjusting foundry orders.
  • Foundry — extremely capital-intensive. A single leading-edge fab costs on the order of $10–20 billion+, and major foundries spend $20–40 billion annually on capex (PatentPC; TechInsights). The payoff is operating leverage when fabs run full: TSMC reported full-year 2025 gross margin of 59.9% and a Q4 2025 margin of 62.3% (TSMC 6-K, SEC).
  • IDM — owns both ends. Captures the full value chain margin when its own products sell well, but carries the foundry's fixed-cost burden with only one internal customer to fill the fab.

How it's used in practice

Investors use the model classification as a first-pass lens on what financial statement risks a company carries. A fabless name's risk is design competitiveness and foundry access/pricing — its capex line is small, free cash flow converts quickly, and a demand downturn cuts orders without stranding owned plant. A foundry's risk is utilization: fixed costs are enormous, so the swing factor is whether the fabs run full, which makes the stock a leveraged play on aggregate chip demand. An IDM combines both risk profiles and is the most operationally fragile in a downturn, because weakness in its design/product arm directly starves its own fabs of volume.

The structural argument for the fabless–foundry split over the IDM is diversification and scale-sharing: a foundry pools demand across many fabless clients, so it can keep utilization high and amortize the immense cost of a new process node across that whole customer base — whereas an IDM must justify the same node investment off internal volume alone (Global Opportunities Trust). This is a core reason the IDM model has lost relative ground; one analysis cites IDM market growth of only ~1–4% CAGR between 2007 and 2024 (Global Opportunities Trust).

The models are also converging at the edges, and the boundaries are not clean:

  • Intel's "IDM 2.0" keeps internal manufacturing for core products, uses third-party foundries (including TSMC) for some components, and opened Intel Foundry to external customers with its own P&L — effectively bolting a foundry onto an IDM (Yahoo/Intel; AInvest).
  • Samsung is a hybrid IDM that runs a large external foundry alongside its own memory and logic products — creating the conflict-of-interest problem that a fabless customer may be handing designs to a competitor.
  • "Fab-lite" IDMs (e.g., parts of the analog/legacy world) own some trailing-edge fabs but outsource leading-edge work.

Adoption, debate & evidence

The fabless–foundry model is the dominant structure of modern leading-edge logic, and the empirical concentration is extreme. TrendForce data puts TSMC at ~69.9% of the entire foundry market in 2025 (up from 64.4% in 2024), with reports of ~72% share in 2H 2025 (Taipei Times; Dataconomy). Its grip is commonly described as even tighter at the leading edge (3nm/5nm/7nm), where trade press routinely puts TSMC's share well above 80–90% — though that bleeding-edge figure is a frequently repeated estimate rather than a measured, single-source statistic, so treat it as directional. That dominance is the central debate: the same separation that lowered barriers for designers has produced a single near-monopoly chokepoint in Taiwan, concentrating geopolitical and single-point-of-failure risk for the entire fabless ecosystem.

The honest counterpoint to "fabless won": the IDM is not obsolete — it remains the right model for memory (Micron, SK Hynix, Samsung) and much of analog/power (TI, ADI, Infineon), where owning trailing-edge fabs and tight design-process co-optimization is an advantage, not a liability. Texas Instruments has actively invested in owning 300mm analog fabs as a competitive moat. So the correct framing is not "fabless beat IDM" but "fabless–foundry won at the bleeding edge of digital logic; IDM persists where process and product are tightly coupled." Beware analyses that treat the three labels as fixed company identities — Intel and Samsung straddle categories, and the model a company operates matters more than the label it carries.

Strengths & limitations

The classification is most useful as a risk-and-margin decoder: it tells you immediately whether a company's swing factor is design wins (fabless), fab utilization and capex timing (foundry), or both (IDM). Its main limitation is that the categories are increasingly porous — applying a pure "fabless = asset-light, high-margin" heuristic to a hybrid like Intel or Samsung will mislead. The #1 misuse is comparing gross margins across models as if they were like-for-like: a fabless 60% gross margin and a foundry 60% gross margin describe completely different cost structures and capex obligations, so they cannot be ranked head-to-head without adjusting for capital intensity and where R&D sits.

Sources

Dispute flagged: overall-foundry market-share percentages are TrendForce figures (FY2025 TSMC 69.9%, up from 64.4% in 2024) reported via secondary press and vary slightly by quarter and definition — directionally robust but treat exact decimals as approximate. The "80–90%+ at advanced nodes" figure is a widely repeated trade-press estimate, not a measured single-source statistic. Margin "55–70% fabless" is a commonly cited range, not a measured industry average. TSMC's FY2025 (59.9%) and Q4 2025 (62.3%) gross margins are verified against its SEC 6-K filing.