Growth Cases for Semiconductors and Construction Amid Rising Rates… AI Infrastructure Emerges as a New Investment Theme

As U.S. Treasury yields rose, government bond yields in other regions, including Europe, also moved higher. U.S. Treasuries serve as a leading safe-haven asset and a benchmark for interest rates in global financial markets, so when U.S. rates rise, rates on relatively riskier assets also tend to move higher. The rise in European yields should not be viewed as an exceptional development, analysts said.

Rising interest rates mean falling prices for existing bonds. With rates having risen sharply, more institutional and individual investors see bond prices as lower and are looking for buying opportunities; many investors are also asking whether they should buy U.S. bonds. At the same time, inflation concerns and the possibility of policy rate hikes continue to put upward pressure on rates.

Institutional investors’ bond positions are also being unsettled. Some had pursued strategies based on expectations that short-term rates would remain low while long-term rates rose, but continued increases in 10-year yields have prompted a reassessment of those outlooks. With views divided between seeing high rates as a buying opportunity and preparing for further increases, markets are reacting sensitively even to minor news.

U.S. Treasury yields were presented as the starting point for gauging returns available from other investments. The point is to assess whether an investment can be expected to deliver returns higher than those available from U.S. Treasuries, a safe-haven asset. A yield of around 5%, cited as an example in the broadcast, illustrated how investors can establish a benchmark for comparison.

As Treasury yields rise, the returns demanded from risk assets also increase. Investors should assess whether the expected returns on other assets, such as equities, are sufficient when compared with the returns available from Treasuries without taking on additional risk. Rather than concluding that the market is in crisis simply because rates have risen, investors should also consider the relative appeal of different asset classes.

Neel Kashkari, president of the Federal Reserve Bank of Minneapolis, maintained his view that further policy rate increases are necessary, but said he had no strong view on whether rates should be raised in October and was open to the pace of increases. He also cited the possibility of one further rate hike each this year and in 2027.

Kashkari is generally known for his strongly hawkish stance, so markets focused less on his familiar hawkish comments than on his lack of conviction about an October hike. With Fed officials sending mixed messages, investors are examining speakers’ usual policy leanings and changes in their wording for clues about the rate path.

Noh Geun-chang, CEO of Semicon Research Lab, said calls to moderate the pace of AI investment are unlikely to translate immediately into sharp cuts in data center construction and semiconductor purchases. As AI systems are designed to perform security, monitoring and verification functions, hardware complexity and costs rise together, supporting more advanced semiconductor products and price premiums, he said.

One example cited was a server management non-memory chip, of which a single unit is used in a conventional server but several are installed in an AI server. Components and functions for security and verification are also increasing in memory, including the addition of error-correction capabilities. When customers request and adopt these functions, the longer production timelines and higher costs may be reflected in product prices.

He said remarks emphasizing the risks of AI may reflect concerns about safety as well as corporate business strategies. Noh interpreted that emphasizing AI’s powerful capabilities and risks around the time Anthropic pursues an initial public offering could benefit its valuation, while highlighting safety and regulatory compliance costs could raise barriers to entry for later competitors. He added, however, that the potential misuse of AI itself must be managed as a real risk.

Noh cited projections that U.S. data center power capacity will expand from about 40 GW currently to around 120 GW by 2030 as evidence of semiconductor demand. Data centers are difficult to secure when needed unless they are reserved in advance, he said, adding that some customers are seeking to secure supply even at higher memory prices. He also saw no clear signs of slowing demand in revenue indicators from Taiwanese semiconductor companies or in the results of foundry and back-end process companies.

Industry projections cited suggest the HBM market could grow to about $60 billion this year, around $120 billion next year and about $560 billion by 2030. Noh said it would take time to build new factories and ramp up mass production, so a significant increase in supply should not be expected until 2029–2030 or later. Higher prices for HBM4 and next-generation products, along with an improved product mix, could support memory makers’ revenue, he said.

SK hynix’s share of the HBM market could decline as Samsung Electronics gains share, but SK hynix’s revenue could still increase as the overall market expands, he forecast. A market outlook cited operating profit of 114 trillion won for Samsung Electronics and 78 trillion to 80 trillion won for SK hynix in the following quarter; these figures were presented as market estimates. Even if a stronger won weighs on exporters, rising memory prices and a larger HBM4 share could partly offset the impact, he said.

Competition between Google’s TPUs and Nvidia accelerators was also cited as a factor expanding HBM demand. Google plans to significantly increase TPU production and sales by 2028, and because TPUs also use HBM, memory demand could hold up even if reliance on any one AI accelerator provider declines. Noh said that if external sales of Google’s TPUs and improved cash flow become more visible, doubts about the sustainability of AI investment could ease and semiconductor stocks could be valued differently.

Noh said SK hynix may consider production and partnership options in the United States, but the economics of U.S. factories remain uncertain. The U.S. government has called for memory production to be localized, and companies’ decisions could vary depending on policy conditions such as tariffs and subsidies. He said a U.S. expansion could be viewed as one of the scenarios already under consideration, but should be distinguished from a finalized business plan.

Intel is incurring heavy losses in its foundry business and may find it difficult to secure products to manufacture even though it has facilities. As more data center companies develop their own Arm-based CPUs, there may be room for cooperation with SK hynix to increase utilization of Intel’s factories, according to the analysis. However, given the high price volatility of the memory business, he added, it remains to be seen how Intel might participate.

Despite orders related to Tesla and expectations that the Taylor plant will begin operations, securing customers for high-performance data center accelerators was identified as the key challenge for Samsung Foundry. TSMC generates a large share of its revenue from data centers and has a high market share, while Samsung currently has a relatively larger share of volumes tied to automotive, smartphones and memory, according to the assessment. Having advanced process capabilities alone is not enough; meaningful customers and volumes are needed to deliver profitable growth.

One proposal was for Samsung Electronics to build experience producing data center chips through collaboration with domestic fabless customers such as Rebellions, and to broaden its customer base by offering competitive pricing. Tesla’s AI chips in the United States and the automotive semiconductor market could present growth opportunities, but winning over TSMC customers would require validation of reliability and supply capabilities. A return to profitability for the foundry business remains possible, but it is too early to assume a substantial near-term earnings improvement, the assessment said.

Park Se-ra, an analyst at Shinyoung Securities, said that since publishing her first report on the construction industry in 2015, she had identified difficulty forecasting earnings, a lack of differentiation among major companies and inadequate shareholder returns as weaknesses in the sector. More recently, more construction companies have resolved troubled projects from the past and recognized losses on their financial statements, while avoiding unsustainably low bids and participating from the design stage have become more common. The sector has regained some credibility as companies such as Samsung E&A have demonstrated their ability to manage costs through their results, she said.

Construction companies are moving to selectively expand their businesses in areas such as nuclear power, plants and data centers. Some companies may serve as lead contractors on nuclear projects, while others participate in consortia as non-lead contractors; GS E&C is building its capabilities in data center development and operations. Park’s central assessment was that the construction sector’s growth story should extend beyond the housing market and Middle Eastern plant projects to include the development of global AI infrastructure.

If competition in AI shifts from model development to how quickly data centers and power facilities can be built, Korean construction companies could become execution partners for overseas projects. U.S. investment-related frameworks and projects could support Korean companies in establishing local entities or joining joint ventures, while also creating opportunities for construction companies to deploy their capital in overseas infrastructure development. However, given the significant construction cost, labor and management risks at U.S. sites, she stressed the importance of structures in which contractors and clients share risks and returns.

Market participants have been associating Samsung C&T, Hyundai Engineering & Construction and Daewoo Engineering & Construction with nuclear power; GS E&C and DL E&C with data centers; and Samsung C&T and Samsung E&A with U.S. semiconductor factories. Park said individual companies’ businesses are not limited to a single theme, but their share-price reactions may differ depending on the labels the market assigns when a particular issue gains prominence. She also pointed to the possibility of rotational buying spreading from large companies to mid-sized construction firms and building-materials companies.

Park said postwar reconstruction demand could extend beyond repairing destroyed buildings to reshaping energy supply chains and power generation and transport infrastructure. If war undermines the stability of existing energy supplies, investment may increase in new oil fields and supply routes, nuclear power and alternative energy, according to her analysis. For actual reconstruction projects, construction companies already operating in the region and equipped with workers and materials have an advantage; companies with a large presence in the Middle East could be mobilized relatively quickly, she said.

The housing market is not yet in a broad-based recovery; rather, projects with proven viability are beginning to move amid a supply shortage. Orders for urban redevelopment projects won by major construction companies totaled less than 20 trillion won in 2024 but rose to about 50 trillion won last year. A forecast suggested the total had already surpassed that level this year and could reach 80 trillion won for the full year. Large projects in Seoul are moving first, and the situation has some similarities to earlier housing cycles in which project viability recovered after a prolonged supply slump, she said.

As an order-driven industry, construction tends to see business expectations reflected in share prices before specific earnings figures emerge. Conversely, by the time order volumes and earnings are confirmed, expectations may already be priced in. Park proposed looking beyond immediate catalysts to issues likely to come into focus about three months later. She forecast that, around the turn of the year, reassessments of the following year’s earnings and order levels could make semiconductor factory construction and AI infrastructure contracts key variables for the sector.

There was also a view that if investor interest in the construction sector as a whole grows, the benefits could spread beyond major companies to specialist contractors and building-materials companies. However, given the shortage of long-term investors in the sector, thematic rallies alone are unlikely to sustain a revaluation; predictable earnings, clear business competitiveness and shareholder returns are also needed. A key factor in long-term valuations will be whether construction companies can use their accumulated cash for equity investments and development participation in growth infrastructure such as data centers, rather than spending it solely on land purchases.

Memorizing the names of Fed officials or all their remarks does not necessarily lead to investment success. The more important task is to understand the broader direction of interest rates and markets, and the stage of the cycle, according to this view. Individual news items can help confirm that trend, but investors should not be swept up in short-term noise.

Construction stocks will struggle to earn the trust of long-term investors if they remain companies that react only to one-off themes such as reconstruction in the Middle East. One commentary argued that they need to demonstrate stable earnings and a long-term growth vision, while strengthening shareholder returns in line with an industry structure that generally requires less capital spending on factory equipment. The argument was that cash should be used efficiently for business investment or shareholder returns rather than simply accumulated.