The Expensive American dream

TSMC reports Q1-2025 result

Amid the Tariff Tantrums (deliberate or irrational—pick your preferred option; the outcome is the same), TSMC sneaked in the Q1-25 result and gave insight into the potential impact of the new world trade order.

While the Trump administration (a longer word for Trump) has been backpaddling lately, the damage has already been done, and what should have been a bull in a China shop turned into a dragon in a Pottery Barn, as the Chinese did not blink.

TSMC clearly stated that they have not seen any change in customer behaviour due to the increased tariffs. When asked specifically about the Nvidia downgraded-for-China H20, they responded that although they did not comment on specific customers, this was baked into the forecast (clearly a comment about a specific customer).

TSMC also clearly stated that they were not involved in any tariff negotiations (I am confident the Taiwanese authorities consulted them) and that they have no opinion about them—they focus on respecting them as a private company.

All in all, this is good news. However, every investor call is now seen as a political statement that needs to curry favour with specific individuals and can no longer be viewed as a neutral statement about a company's state.

Fortunately, that has never been my weakness. While I listen to what is said, I will analyse the data and then decide what it means.

While the result was a couple of per cent lower than guidance, it was inside the expectations window. A predicted result, however, is not the same as a good result.

The result included damage from an earthquake and was also in the seasonally declining smartphone shipments quarter. Although the quarter was down, the annual growth rates were strong, with revenue growing close to 39% while operating profits rose by 59%.

Also, the Q2 guidance was strong, suggesting 11.5% growth in Q2-25.

As seen below, the smartphone revenue decline was larger than last year's impact in the same quarters. This is a hint about the likely performance of Apple that will reveal their hand on May 1st.

This was also confirmed by the decline in 3nm revenue, which decreased by over 20% compared to last quarter. 5nm also declined a fraction, but this did not prevent the complete dominance of high-performance computing from a revenue perspective.

The HPC share is now 59% while the Smartphone share has declined to 28%.

The dominance of High performance is also seen in TSMC’s revenue by country, where the US AI companies increasingly dominate. This is a key reason behind TSMC’s geographically dispersed strategy and overseas expansion plans.

While the Chips Act and Trump Tariff Tantrums certainly have impacted the size and timing of investments, TSMC has been driven by the US market. Getting some leading-edge capacity outside the Chinese army's strike range also plays a role.

The US-based revenue is now at 77% and will likely increase over the foreseeable future.

The problem with the US expansion is that while it might be necessary, it is not cheap compared to expanding in Taiwan, which would be the optimum financial place for the investment.

Gross Margins

The gross margin declined by 20 basis points to 58.8% due to several factors detailed in the call. Firstly, cost improvement efforts offset a 60-basis-point earthquake impact and dilution from the overseas operations.

An additional 80 basis points from the ramp of the Arizona operation in Q2-25 negatively impact the outlook. The combined operations of Kumamoto and Arizona fabs will expand to 2-3% by 2025.

With the expanded 100B$ commitment to Arizona, the margin dilution will continue to be 2-3% in the early stages and grow to 3-4% in the later stages.

TSMC still believes that it is possible to maintain a 53% long-term gross margin despite the increasing overseas operations and will work with customers and suppliers to make that happen.

Analysing TSMC's gross margin against the long-term goal suggests the impact of overseas operations could be significantly higher. The margin for this quarter would have been 59.4% without the earthquake, or 640 basis points higher than the long-term Gross profit target.

This is significantly more than the long-term 350 basis points indicated from the overseas operations. This could be management's result padding, giving TSMC room to move in, or due to increased uncertainty. While TSMC can expand operations in Arizona, it is not a given that it is easy for its supply chain to follow.

While the overseas expansion might be needed for political reasons, it is not a sound economic decision. Fortunately for TSMC, the Taiwan operation is solid and can fund a US political PR campaign.

CapEx and plan

TSMC reiterated its Capex plan of 40 B$ for 2025, and as the chart below shows, this represents a continuation of the elevated level of the last two quarters.

TSMC responded to the US Chips Act by declining CapEx investments as the Taiwanese chip giant prepared for its expensive US campaign. The CapEx was so low that it flirted with maintenance investment levels for a while, decreasing PPE. In other words, the value of the TSMC manufacturing plants decreased in this period, although the capacity increased predominantly due to optimising the most advanced nodes.

The additional 100 B commitment towards the Arizona plants will only minimally impact the 2025 budget; this is a longer-term venture and can change with the US investment environment. TSMC understands that the 100 B US investment proclamation is a lot more critical to the PR-hungry administration than the actual outcome under another administration (if the US democracy can withstand the pressure).

The plans are tangible and involve four additional wafer fabs, the two fabs in progress, and a major R&D centre. This will bring the total investment in Arizona to 165B$. The first 4nm fab has been in production since Q4-24 with comparable yields to Taiwan (but not costs), and the second is a 3nm fab. The two additional fabs will be 2nm and 1.6nm, with the third additional fab to be constructed in line with customer expectations and will involve even more advanced technology.

TSMC's investment also includes two new advanced packaging plants and a major R&D centre. Thus, the expected gross margin decline is plausible as the US footprint grows.

Once completed, the US footprint is expected to be 30% of the 2nm and more advanced capacity.

The R&D centre will focus on the needs of the local manufacturing cluster to improve its technology and independent operation. Local R&D is a clear focus of the US administration, and TSMC plans to employ around 1000 employees, which should be compared to TSMC's current level of R&D employees, which is 10K people.

TSMC was not willing to disclose a more specific timeline for the expansion other than the fact that it depended on the labour shortage in Arizona (another example of why it is not an optimum location) and the timing of permits.

To offset the overseas investment margin drag, TSMC needs an outsized domestic investment with optimum financial performance. While no details were offered, this undoubtedly mirrors the overseas investment to dilute the high costs necessary to satisfy predominantly US government and customer needs.

The CapEx budget allocates 70% to advanced technologies and 15% to speciality technologies and advanced packaging. This aligns with the TSMC 2.0 foundry strategy that includes the OSAT market in the foundry market.

This will have a profound impact on the current OSAT companies.

TSMC’s CapEx will be at the same scale as the combined CapEx of all the OSAT companies and their investments are not entirely directed towards advanced packaging but packaging in general. This is massive and TSMC is determined to become the leader in advanced packaging as it understands it is vital for advanced computing short term and all advanced products longer term.

The specialty investment is often mistaken for being mature as the feature sizes are the same as the more mature technologies, however TSMC clearly stated that the spciality investments were into areas with no and limited competition. Asked about if it involved moving equipment from mature node manufacturing in Taiwan to Europe and Japan did not garner an answer.

Intel and other rumours

TSMC took time to point out that none of their US plans involved joint ventures, technology licensing, technology transfer or sharing effectively burying the endless Intel rumours. If people only knew a fraction of anticompetitive laws in US and Europe, they would know that such an alliance would never be approved by the authorities.

I have earlier written about the Herfindahl-Hirschman index that governs the anticompetitive laws of Europe and the US and the fact that any TSMC merger is impossible, while JV’s are an option (The immense changes in the Semiconductor Industry)

Capacity

From a capacity perspective TSMC commented that while they were not able to supply all the AI chips needed last quarter, the demand vs capacity was more balanced this quarter. This demand is predominantly from the 3 and 5nm technologies while there is ample capacity at 7nm and above.