This is the first of an occasional series of articles on the semiconductor industry. Many column inches have covered industry consolidation and in this first article, I aim to explain how the industry reached this point. Later articles will cover subjects including China, joint ventures, emerging players like Brazil and Vietnam, monopolies, M&A, national security/national development, customer concentration, verticalisation/disintermediation, ecosystem venturing, etc. The timing of these will be erratic out of practical necessity and the order of themes…in no particular order.
“If you want to turn a big pile of money into a small pile of money, start an airline – or a semiconductor company” – Anon
Semiconductors are examples of practical magic. They have enabled the transformation of existing industries and the creation of new ones, disrupting lives for both better and worse. The semiconductor industry has enabled the creation of an incredible amount of value, but has a demonstrably poor track record when it comes to capturing much of that value.
In 2011 McKinsey noted that the return on invested capital (ROIC) for many semiconductor companies lagged behind their weighted average cost of capital (WACC) – the rate of return a company must earn for its stock and debt holders. The consulting firm estimated that the semiconductor industry (excluding Intel) destroyed ~$47bn of shareholder value between 1996 and 2009. This was the result of many losers and a few big winners. But what of Intel? Intel alone created ~$57bn in value during the same period. This is a testament to the profitability of the little-challenged Wintel PC monopoly (to put this era into some kind of context, Apple’s iPhone was launched in 2007 and the iPad was not launched until 2010).
In 2009, the Berkeley economists Clair Brown and Greg Linden published their study of the semiconductor industry. In it, they analysed the Return on Assets (ROA) of a range of semiconductor companies for the period 1984-2005. The average ROA for a five-year US Treasury bond during this period was 6.5%. Intel gave a ROA of 13.1% in the same period, while integrated device manufacturers (IDMs) – excluding Intel – and fabless companies alike yielded an average ROA of 2.7%.
Some form of competitive advantage, whether monopoly, IP, secret sauce, etc. has helped to generate normal and supernormal profits for a small number of semiconductor companies. What has stopped the other players? Why would someone be better off investing in treasury bonds than in many semiconductor firms? Why does it appear as if some semiconductor firms are steadily competing their way out of business?
It’s complicated and rooted in recent history. Let’s examine this rich picture through a variety of lenses.
During the 1980s and 1990s, the semiconductor industry grew at a five-year CAGR of 11-23%, driven by ‘killer applications’ such as PCs and later, mobile phones.
The industry grew larger and matured, with single-digit growth now the norm. In part this is due to sheer scale – the semiconductor industry of 1990 was worth ~$50bn and in 2015 was worth ~$335bn. But also there has been slower growth or even decline in the former killer applications and there is a lack of new, market-driving applications – the buzz behind the IoT bandwagon and machine intelligence (which is a technology, not an application) notwithstanding.
Additionally, there has been another driver at work. During the early 2000s, there was a shift away from the corporate PC market as the growth engine of the chip business, to the fragmented and price-sensitive consumer market. The consumer market is typically low-price, low-margin and time-sensitive.
This consumerisation of end market demand had implications for cyclicality, chip lifecycles, average selling prices (ASPs) and their erosion, R&D focus, R&D costs and customer relationships. That’s a profound change.
With increasing standardization (required for interoperability) and the consumerisation of demand, value migrated downstream, away from chip design and manufacture towards brand, software, services and content.
This trend accelerated with the rise of mobile, its principal software platforms – from Apple and Google – and their network effects. It was helped along by vertical integration, causing the disintermediation of traditional chip suppliers, and sometimes including the backwards integration of semiconductor firms (like Apple’s acquisition of PA Semi).
Today, the consumer captures a lot of end use value – the functions in my iPhone are a testament to this. Nearly all the economic value goes to Apple, from the application processor, through to retail, apps and services. Foxconn earns low single-digit operating profit margins and the rest of the value system – vendors of IP, semiconductors and display panels, etc. – fights over the rest.
Thanks to a few very strong IT and consumer electronics brands and the high use of a limited number of original design manufacturers (ODMs) and electronics manufacturing services (EMS) providers, there is high concentration of demand – and hence buyer power – over semiconductor firms. According to Gartner, in 2014, the top 10 OEMs accounted for 37%, with Apple and Samsung accounting for 17%.
Where there is high buyer power, ASPs are dictated to the sellers, and thanks to a relatively high degree of standardization, switching costs are often relatively low. And in the case of Apple and Samsung, they have their own chip design capability, disintermediating semiconductor firms from some of the most profitable design slots.
Mega-customers demand lots of support, are not loyal in the conventional sense, and present a clear and present risk to the semiconductor supplier, in that the bulk of its revenues are hostage to the mega-customer. The supplier could spread the risk by winning other customers, but would need to put a lot of time and resources behind wooing and servicing a large number of smaller customers in order to create anything approaching a meaningful counter-balance to the mega-customer.
The fact that customer concentration appears to be stuck at high rates has changed one notable thing. There was a time when a product development programme could be a ‘bet the company’ moment (I am thinking of Boeing’s iconic 747). But now winning/losing a single key customer can be a ‘bet the company’ moment.
The steady increase in integration delivered by Moore’s Law made it virtually impossible for any single industry player to be an expert in everything. The technological advances made thanks to Moore’s Law altered industry structure, by encouraging disaggregation and specialisation, creating packaging and assembly houses, foundries, design services companies, IP vendors, etc.
With competitive strategies principally based on integration, semiconductor firms jumped to more advanced process nodes, and the costs of design and fabrication rose in line with complexity. These costs multiplied with the emergence of highly complex system-on-chip (SoC) devices requiring lots of verification and software. With firms taping out more designs to satisfy multiple consumer end market segments, it became clear that some of these segments were not offering appropriate levels of return on investment (ROI).
Both de jure and de facto standardisation efforts have greatly helped to develop markets, with competitors coming together to collaborate on baking a pie, before competing to get the biggest slice. There is a virtuous (or vicious, depending on one’s POV) circle of globalisation and interconnectedness – a result of standardisation. As interconnectedness has increased, so has the level of standardisation in products. In some cases, this has constrained the space for differentiation. And at the worst extreme, has created a herd mentality, leading to a slew of ‘me too’ efforts.
It’s worth saying that something is not necessarily innovative if everyone else is doing the same thing. Many industry ‘innovations’ have involved integration – a direct benefit of Moore’s Law. There are times when one can defend a particularly difficult form of integration involving a novel circuit and special process technology, but it is very hard to defend when the modules are digital and fabricated on a standard digital process offered by a 3rd party foundry to all comers. Innovation should really be aimed at getting ahead of competitors. Anything that merely keeps up with competitors is just playing the game, not winning it.
True innovation takes time, often comes with great risk and sometimes flourishes under constraints. That often conflicts with corporate cultures that are shaped – in terms of leadership, risk appetite, accountability and incentives, etc. – by powerful customers, standardisation and short cycle times.
Some engineering teams have been driven to reinvent the wheel – with clear implications for the cost base and cycle time – leading to the development of ‘products that people don’t want, late’ (what I call the ‘high-cost, slow-follower’ strategy), which impacts economic returns, and the retention of talented engineers (who want to do something interesting).
Differentiation is concerned with making a product different to other, similar products, usually for greater economic return. A differentiation strategy goes awry if it is focused on features instead of benefits (the latter is what customers pay for) and is not sustainable for long if it relies purely on integration. Thanks to standardisation and use of 3rd party IP (which lowered some entry barriers to semiconductor newcomers), differentiation has become quite limited in several application areas. This may be a result of industry mindset – the ‘product’ is actually only one part of a value proposition.
In the 1990s, semiconductor companies rarely provided software. Much of what was provided was device drivers and middleware, often not of production quality. At that time, it wasn’t seen as a core activity. It was often given away for free as semiconductor companies didn’t want to be on the hook for supporting it. This created an ingrained attitude in the minds of some customers that software from semiconductor companies was free.
As complexity increased, leading the industry into the SoC era, software was increasingly (and unwittingly) becoming a part of a semiconductor company’s product offering. It was also becoming a big variable in development programmes – obtaining support from a software platform vendor, persuading ecosystem members to port, racing to deal with incompatibility and performance issues, etc.
This led to software requiring more focus and resources within semiconductor companies, increasing their cost base, as well as challenging management mindsets. But with customers continuing to regard software as a freebie, this helped to push margins downwards.
In terms of business strategy, a general distinction can be drawn between those who believe that the major source of profitability is industry attractiveness, while others believe it is competitive advantage through superior resources and capabilities. In practice, a successful strategy combines both elements, supported by organisational sensemaking, agility (for when circumstances change…and your response must also change), and a degree of luck.
In the industry, we have often espoused generic strategies such as cost leadership, differentiation or focus (cost or differentiation within a focus market) strategies. At times, such strategies have led to unintended consequences, such as a race to the bottom (when competing on price), or believing that what is important to the supplier is equally critical to the customer (on a specific point of differentiation), or simply under-delivering by virtue of hedging your bets, and diffusing finite resources across several areas (failing to invest and focus heavily on a ‘focus’ area).
To the list of strategies we can add variations, such as the low-cost fast-follower strategy, as pursued by Taiwanese – and later – Chinese semiconductor companies. We can also think of perspectives, for instance the ‘Blue Ocean’ strategy of uncontested market space versus the notion that if a company has no competitors, it may not have a market.
Whether any or all these strategies are appropriate or inappropriate, and successful or not, depends a lot on context (internal and external) and successful execution.
Interrelationships have made competitive strategies somewhat tricky to formulate. Arguably, there is an interrelationship between standardisation and differentiation, and there is another between licensing 3rd party IP to reduce time to market and differentiation.
Factors such as standardisation, accelerated design cycles and the use of 3rd party IP may have hampered risk appetite and the development of core competences & IP that could form a sustainable competitive advantage. Likewise, holding a monopoly position in one market may hinder efforts to become competitive in a different market – essential when the monopoly market becomes unattractive.
Sadly, many semiconductor companies were unable to create a persuasive case to be paid more – or be paid at all, in the case of software. This was in part due to a lack of differentiation between chipmakers, which allowed their (powerful) customers to play them off against one another to reduce prices.
Some of the mindsets that had prevailed – market share over profit, a race for scale – may have blinded some semiconductor companies to profitable niches, instead launching them into head-to-head confrontations over mega-markets with a small number of strong players, leading to prices being competed down. In some cases, such responses became bad, persistent habits.
But it didn’t necessarily have to play out this way.
Over time, a company’s top ten customers would change in composition; the customers may have changed strategy, been acquired, had lost out to new entrants or were simply mired in maturing markets. But the change in composition was inevitable. Many suppliers were unable to avoid being a mere observer of such changes.
Mediatek presents an interesting counterpoint. Many fabless start-ups (and upstarts) have tried to conduct business with tier one OEMs. They are frequently rebuffed, for both sensible and nonsensical reasons. Mediatek pursued another strategy; it developed ‘total solutions’ that not only lowered the technical barriers to many Chinese electronic system makers, it enabled totally new players to enter. The emergence of the shanzhai (bandit) mobile phones is testament to that strategy. Now of course, Mediatek is at some risk of being supplanted in the Chinese supply chain by the Chinese government’s revised semiconductor policy, as well as its ‘Made in China 2025’ manufacturing policy further downstream – and Taiwanese government restrictions on FDI from China.
Scale and consolidation
Increased costs, lower prices and diminishing returns have driven consolidation amongst semiconductor companies. Scale, so the argument goes, will help the combined entity weather industry changes. If only it were that simple; for there is good scale, and there is bad scale.
‘Good’ economies of scale are the cost advantages achieved by increased output of a product, with fixed costs spread over more units of output. That assumes that the product is the same or has minimal variety.
‘Bad’ economies of scale often come with increased scope(and I don’t mean economies of scope, which can also be a benefit through product diversification, bundling, etc.). For instance, acquiring a company with a similar/overlapping product line is an example of bad scale if you cannot offload that product line and bring its customers over to the acquirer’s similar product line, because you are still funding two R&D activities. Bad economies of scale, if not rectified, increase costs and complexity.
Consolidation merely buys time (especially if peer companies are also at it). Without changes elsewhere in semiconductor companies and the industry at large, any benefits of consolidation will be competed away, benefitting powerful customers. It also goes without saying that synergies achieved at one level of the industry (for example a chip maker) may have knock-on effects for those upstream (e.g. IP and EDA vendors), who may suffer revenue reduction due to a reduced number of design starts and engineers.
In theory, consolidation should help reduce market power, but only if accompanied by a change in behavior – that is much harder to achieve than finding synergies.
The late Peter Drucker talked about the ‘theory of the business’, the assumptions on which a business has been built and with which it is operated. Many semiconductor firms foundered because their assumptions were no longer valid; the industry and the end markets had changed greatly.
The preceding text may have made for grim reading. But as noted at the beginning of the article, some semiconductor companies have managed to create value. And this is not by any means the end of the semiconductor industry. It’s merely the end of the beginning. The industry has had an adventurous adolescence and must now act as a mature adult (I will refrain from calling this phase ‘Semiconductor 2.0’ or the ‘post-chip era’). The industry has spent more time and energy battling complexity than it has on capturing more of the value it has created. Shareholders would argue that the industry had its priorities wrong.
There is of course, no one silver bullet, no semiconductor fairy godmother, and no miracle cures. The industry has matured and things have simply become harder. So the industry playbook must change.
A semiconductor company must decide what areas it plays in and equally, what areas it does not play in. It now requires a lot more detailed work and faith to choose the right target market and segment, and devise a compelling, hard-to-imitate value proposition, backed up by internal resources and patience, and committed 3rd party alliances (it is a crowded industry, after all). It must focus equally on strategy and execution – in all areas of its business. It requires an increase in risk appetite, a change in mindset, an ability to make sense of the environment, and agility – and that is a long change journey to undertake.
Above all, we need to think differently about our industry. How do we create our own luck, instead of being dictated to by a handful of powerful customers? How do we renew our industry gene pool, when private equity has largely retreated from funding semiconductor startups due to the risk/reward imbalance?Share this post via:
Sino Semicap Sanction Screws Snugged- SVB- Aftermath more important than event
Sino Semicap Sanction Screws Snugged- SVB- Aftermath more important than event
IEDM 2023 – 2D Materials – Intel and TSMC
Scaling the RISC-V Verification Stack
SPIE Advanced Lithography Conference 2023 – AMAT Sculpta® Announcement
The Billion Dollar Question – Single or Double Digit Semiconductor Decline
AAA Hypes Self-Driving Car Fears
IoT in Distress at MWC 2023
Scaling is Failing with Moore’s Law and Dennard
AAA Hypes Self-Driving Car Fears