Intel, Nvidia, Et Al., And American Semiconductor Hegemony
One thread in the fabric of alarmist political/economic commentary on American Decline has to do with the state of the semiconductor industry. Supposedly the U.S. is losing/could lose/has already lost leadership in this “critical” field of technology, the understructure of the digital economy, a loss which puts at risk practically everything we might care about – life (“Semiconductors are essential to modern life”), liberty (“American leadership in semiconductors also is vital to the technological superiority of the U.S. military”), and presumably happiness as well (Facebook, Youtube, Instagram and Netflix NFLX all run on silicon)…
Politicians, journalists and think tank analysts have been pushing this dire story year after year, decade after decade, administration after administration.
- The Reagan Era: “The health and vitality of the U.S. semiconductor industry is essential to America’s future competitiveness. We cannot allow it to be jeopardized.” – Ronald Reagan (1987)
- The Clinton Era: “The decline of American semiconductor producers… the days of American dominance of science-based industries might be numbered… the first domino in a cascading fall of downstream electronic-systems industries.” – An industry assessment (1999)
- The Obama Era: “Semiconductor innovation is slowing… A concerted push by China to reshape the market in its favor, using industrial policies backed by over one hundred billion dollars in government-directed funds, threatens the competitiveness of U.S. industry.” – The President’s Council of Advisors on Science & Technology (2016)
- The Present Day: The U.S. needs to invest ambitiously in semiconductor research….[or] risk losing its innovation edge and the global competition for technology leadership. – Report by the Semiconductor Industry Association (2019)
The threat is (of course) China. According to The Boston Consulting Group, the game is probably already over:
- “A continuation of the status quo or a complete decoupling of US and Chinese tech industries could have profound negative repercussions for the US semiconductor industry, well beyond the expected impact of the “Made in China 2025” policy.
- “Once the US loses its global leadership position, the industry’s virtuous innovation cycle reverses direction, throwing US companies into a downward spiral.”
Reversals, cascades, spirals, “the annihilation of the American industry,” – signs of imperial atrophy…?
American Semiconductor Hegemony (more or less)
The story is all wrong. It is based on a crude misunderstanding of the structure of the industry. In fact, the U.S. is in firm control of the high ground in the semiconductor business, which dominates the rest of the industry (with one quasi-exception).
Chip-making comprises several technological processes: design; precision manufacturing of the integrated circuits; and “packaging” (which includes assembly of the integrated circuits into durable “chips”). The original business model of the industry combined all three. Such companies are known as the Integrated Device Manufacturers (IDMs). A few firms still employ this model, notably Intel INTC , the overall industry leader (by revenue).
Beginning in the 1980s, a new trend developed, based on the disarticulation of these functions. Some companies, known as foundries, focused strictly on the manufacturing business. Others specialized in design, and contracted out to the foundries to manufacture chips based on their designs. These become known as “Fabless” semiconductor companies. Other companies focused on packaging and are known as Outsourced Assembly and Test (OSAT) segment.
(Firms in the OSAT segment are comparatively small; the business is crowded, competitive, low margin, and very cyclical, with weak pricing power compared to IDMs. A fourth segment, the suppliers of the capital equipment used by foundries and IDMs in the fabrication process, is highly profitable and economically powerful. It stands apart from the rest of the industry, and is a story for another column.)
The Fabless vs. Foundry contrast
The difference between the Fabless and Foundry models is based on the economics of IC fabrication, which are staggering. A new IC fab plant can cost up to $20 Bn – about twice the cost of a Nimitz class aircraft carrier, three times the cost of a nuclear power plant – and an IC fab may have a useful lifetime of just a few years before it becomes technologically obsolete. The implications of this are significant, and mostly negative from a strategic standpoint. It is a scale-based business (as all heavy-capex industries are) and it is dominated today by one firm: Taiwan Semiconductor Manufacturing Company (TSMC).
The Fabless model is “asset-lite.” They do not need to buy a new aircraft carrier every year, so to speak. Instead, they can invest in design talent and functional innovation.
IDMs like Intel still do both. But the capital expenditure demands weigh heavily. Over time, a number of IDMs have surrendered to the economic logic and have split themselves into two pieces – a Fabless design firm and a foundry – to go their separate ways. AMD was an IDM until 2009, when they spun off their foundry business into a new company, Global Foundries. Even Intel has flirted with the idea of a split.
The confusion of these business models under the non-discriminating category “semiconductor companies” is surprisingly common. For example, the Wall Street Journal recently listed the “Top Semiconductor Companies by Sales” in a chart I have redrawn here (adding a few other prominent semiconductor companies not included in the WSJ chart):
This is a heterogeneous mishmash. Intel and Samsung are IDMs. TSMC is a pure foundry. Qualcomm QCOM , Nvidia NVDA , AMD are Fabless IC firms. Texas Instruments TXN is an IDM of a special sort, focusing on what are called analog and mixed signal chips (not the digital processors and memory chips that most of the industry produces). Analog chip manufacturing has different and possibly more favorable economics. In short, these companies are very different, with very different cost structures, and are valued very differently by investors. Throwing them together under one label is like grouping Ford and General Motors GM with Hertz and Uber UBER as “automotive companies.”
The first flag that the “Decline thesis” is invalid is the fact that 10 of the 14 firms listed here are American. One is Taiwanese, two are Korean. Only Hi-Silicon – a Fabless IC designer – is based in mainland China. HiSilicon is a fully owned subsidiary of Huawei, which puts them in the middle of the risky geopolitical chess game I have discussed elsewhere. Huawei takes about 90% of HiSilicon’s output – “output” which HiSilicon does not and cannot manufacture, but outsources to Taiwan’s TSMC (and perhaps others now).
All in all, the American position, even by this crude tabulation, is strong.
But if we divide the firms into their true categories, the picture clarifies. Compare them on the basis of the capital investment required — property, plant and equipment, “capex” — and the distinction between the foundries and IDMs on one hand, and the Fabless and mixed signal firms on the other is striking. Nvidia – which has a market value $60 billion higher than Intel – creates this value with 25 times less capital investment.
Then consider how many dollars of revenue a company generates from each dollar of capital investment.
Which would you rather own? The company that can take your investment dollar and generate 75 cents of sales, or the company that can take that same dollar and generate $8 of sales?
The market sees and respects this difference.
The Fabless model creates 2 to 4 times more shareholder value per dollar of earnings-per-share than the asset-heavy foundries and IDMs.
Now notice the national clustering. The high-valued firms (high P/E), and the highly efficient business models (high dollars of revenue per dollar of investment) are all U.S. companies.
Summary: Show Me The Money
- First, with the exception of Intel (an IDM which is truly a special case and perhaps even that rara avis, an American “national champion”) the U.S. semiconductor industry has essentially conceded the capex-heavy foundry model to one company, TSMC. (TSMC in Taiwan is the quasi-exception to America’s semiconductor hegemony. Dependence on TSMC may create some risk potentially, but to date it has been a successful symbiosis between the Fabless players and their main foundry.)
- The few important non-U.S. companies in the foundry and IDM segments – TSMC, Samsung – are closely coupled to the U.S. market and supply-chains, and are reliably “friendly.”
- America dominates the Fabless sector, and is in no jeopardy of losing that position
- The Fabless segment is where the market sees the real value in semiconductors, and the market is not wrong; the foundry business is a commodity-type business, favoring consolidation and scale; heavy investments in rapidly depreciating plant and equipment do not inspire investors; the Fabless business model is where the value-creating innovation we associate with the digital economy is concentrated
- China’s only Fabless entry is Hi-Silicon, which is a captive subsidiary of its one main customer; they have yet to prove they can compete in the global market
- China’s entry in the foundry sector — SMIC, another of Huawei’s step-children – has struggled, and not likely to gain much ground against TSMC in Taiwan any time soon
In short, the U.S. controls the semiconductor world, and controls the direction of technological innovation. China is starting at the bottom, and it will be tough hill to climb. In the 21st-century economy, value accrues to companies that control the “intangible” assets like design, brand, human capital (talent), loyal customers, and intellectual property. Traditional 19th-century style “assets” like factories, machinery, inventories, accounts receivable, and even excess accumulations of cash begin to seem more like, well, liabilities. U.S. companies – in many segments of the economy, not just semiconductors — have figured this out. Except for Huawei – troubled Huawei – China is mostly still stuck in the industrial age.