Over the past few years US policy makers have developed a high degree of anxiety about America’s need to have onshore leading edge fab capabilities. The press has been filled with stories about all the ways the US has fallen behind, how we can no longer produce chips at the leading edge. This anxiety is rooted in both geopolitical concerns around the rivalry with China but also around concerns that somehow the US lost this ability to produce chips, the way we seem to have lost the ability to manufacture many things. We would argue that the problem is very much an economic problem rather than a technological one.
Put simply, Intel can produce leading edge semis today, they just cannot do so economically. They have the know-how, and the equipment. Their shortcomings are entirely based on the fact that they can not yet make 18A, 3nm chips profitably.
Consider for a moment everyone’s favorite fear-mongering scenario – what if the US lost access to TSMC’s Taiwan fabs – war, blockade, alien invasion – whatever reason. And associated with that, imagine the US faced some military conflict and the lack of leading edge chips became an acute national security problem. That would obviously cause immense disruption to the economy, but how long would it take for Intel to get its process working? It is reasonable to assume that the government could throw enough money at Intel to get its process up and running very quickly. At first Intel yields would be terrible, and so the government would effectively be covering those losses – paying fixed prices for wafers with only 5% or 10% good die. In semis, volumes solves a lot of problems, and in a time of great need, they could afford all the bad wafers it would take to get sufficient learning under the belt to improve those yields.
Obviously, that is not a thing anyone hopes for, but we offer it here because it speaks to an important but largely ignored reality today. The traditional narrative on TSMC’s rise to success is that Intel missed the boat on mobile, TSMC became the foundry of choice for phones, and that drove their volumes. With that volume they were able to generate those learnings faster than others, and with time win process leadership. All that is true, but it misses one critical factor. During that period, TSMC enjoyed massive subsidies. The best known of those were direct subsidies from the Taiwan government which allowed them to import wafer fabrication equipment in the early days, but a far bigger subsidy was indirect – the undervalued New Taiwan Dollar.
The NT$ is ostensibly a free floating currency, but as the graph above shows, it appears to have been fixed to the US dollar for over 20 years. Economist Brad Setser has written extensively about the mechanisms Taiwan uses to achieve this (TL;DR – Commercial banks, then life insurance companies), essentially managing the currency to keep Taiwan competitive.
During this period, the US dollar has inflated considerably, which effectively means the NT$ has fallen in value as well. The best way to demonstrate this is the Economist’s Big Mac Index. This is a handy way to demonstrate the degree to which a currency is under/over-valued relative to the US dollar. As this graph shows, the NT$ has steadily eroded in value against the US dollar for the past twenty years, exactly the time when TSMC was rising. From what we understand, this decline really began with the Asian Financial Crisis in 1998 and has only grown over time.
This provides a massive, indirect subsidy to TSMC. They can pay their employees wages that are great in Taiwan but are effectively much less than what their competitors in the US would have to pay. TSMC’s revenues are priced in US dollars, but its workforce is paid in NT$, and they are the company’s true asset. And this discount has been compounding for decades. We tend to think of devalued currencies as providing a way for exporters to undercut foreign competitors by offering lower prices. Instead, TSMC used the effects of that currency suppression to invest in their own talent pool. We have read research that shows the NT$ is effectively underpriced by around 30% (using a more comprehensive method than the Big Mac Index which shows a >50% level). So we think it is no coincidence that TSMC has said its US plant would be 20%-30% more expensive than their Taiwan fab’s wafer costs.
To be clear, we are not diminishing TSMC’s technical talent. They have immense capabilities, and unrivaled human capital. But we think it is important to understand how they achieved those. TSMC management’s true power move was investing their currency advantage into their own talent pool, and not squandering it on far-flung acquisitions with tenuous ties to the core business.
This all begs the question as to what anyone can do to compete with that. The US government is fully aware of the state of Taiwan’s currency and has steadily declined to take any action against it. Competing in foundry requires outside funding pools. For Intel, that has lead to further calls for direct government support. We would obviously prefer a more commercial solution, in the form of investment from prospective customers, but there is a rising chorus begging encouraging the US government to “save” Intel by subsidizing some group of equity holders. The less obvious question is how can Samsung respond. The South Korean Won’s status is actually not far off from the NT$, for similar reasons, but Samsung has chosen to spend that windfall elsewhere. In the short run, Samsung Foundry will likely need support from the broader Samsung Chaebol. It is unclear to us to what degree the chaebol is interested in propping up Foundry. The memory business is important for the whole group. Does Foundry put that at risk? Does the cash burn from Foundry get so large that eventually the chaebol decides to fall back and focus on memory? What is clear is that both Intel and Samsung will need considerable outside support to stay in the game.
Photo by Google Gemini
Leave a Reply