The FRED Blog recently discussed who holds Treasury securities in the US and abroad. Today, we answer a related question: How do Treasury securities fit into the overall portfolio of US financial assets held overseas?

Our FRED graph above uses data from the Treasury International Capital (TIC) system to show the class shares of US financial assets held overseas. In April 2026, those asset classes were (in descending order)

1. Equities (blue area). Common stock, preferred stock, and fund shares of US corporations made up 59.4% of the overall portfolio of US financial assets held overseas
2. Long-term Treasury securities (pink area). Bonds issued by the US Treasury and maturing more than a year into the future amount to 19.9%
3. Long-term corporate bonds (purple area). This category also includes state and local governments bonds (including municipal bonds) and it represents 13.3%
4. Short-term Treasury securities (orange area). Bills issued by the US Treasury that mature in less than one year amount to 3.8%
5. Long-term agency bonds (green area). Those bonds are issued by US federal agencies Fannie Mae, Freddie Mac, and Ginnie Mae, representing 3.6% of the overall portfolio of US financial assets held overseas

The graph also shows a shifting mix of asset holdings. Between 1984 and 2008, equities made up roughly one third of foreign portfolios of US assets. But the 2007-2009 Great Recession greatly boosted holdings of risk-free Treasury securities, and those peaked at 36.5% in 2009. Between then and the time of this writing, equities gradually reclaimed their lion’s share of foreign portfolios of US assets, amounting to more than half of their overall value.

To learn more about this topic, check out this May 2026 FEDS Note.

How this graph was created: Search FRED for and select “Foreign Portfolio Holdings of U.S. Equity Securities: All Countries.” Click on the “Edit Graph” button and select the “Add Line” tab to search for “Foreign Portfolio Holdings of U.S. Long-Term Agency Bonds: All Countries.” Don’t forget to click on “Add data series.” Repeat the last two steps to add data on “U.S. Long-Term Corporate Bonds,” “U.S. Long-Term Treasury Securities,” and “U.S. Short-Term Treasury Securities.” Next, select he “Format” tab and select “Graph type: Area” and “Stacking: Percent.”

Suggested by Diego Mendez-Carbajo.

The takeaway

When tariffs were relatively stable, prices for durable goods such as appliances, electronics, and furniture were declining by as much as 3% year-over-year. But in 2025, the effective tariff rate surged to over 11% and durable goods prices began increasing by 2% to 3%.

Dramatic shifts in durable goods prices and trade policy

In our FRED graph above, the solid blue line shows the year-over-year percent change in durable goods prices, and the dashed green line shows the effective tariff rate on imports. (Btw, the effective tariff rate is total tariff revenue collected by the government divided by the total value of all imported goods.)

For most of the period shown, durable goods prices were actually falling. Deflation was the norm from mid-2023 through early 2025, with prices declining by as much as 3% year-over-year. Meanwhile, effective tariffs remained relatively stable at around 2.5% through 2024.

The picture changed sharply in 2025. The effective tariff rate surged from roughly 2.5% to over 11%, more than quadrupling in about a year. Shortly after, durable goods deflation reversed course, from declines to increases of around 2% to 3% by early 2026.

While the timing is striking, this relationship is complex: Durable goods prices reflect many factors beyond import tariffs. Still, the coinciding tariff spike and price acceleration suggest import taxes may be playing a role in ending the deflation consumers experienced with durable goods such as appliances, electronics, and furniture.

The current trend

Notably, effective tariffs appear to have peaked in late 2025 and have since begun declining. If this trend continues, we could see a corresponding moderation in durable goods inflation and potentially even a return to the price declines that characterized the earlier period.

How this graph was created: Search FRED for and select “Personal consumption expenditures: Durable goods (chain-type price index).” Click “Edit Graph” and change the units to “Percent Change from Year Ago.” Next, click “Add Line,” search for and select “Federal government current tax receipts: Taxes on production and imports: Customs duties,” and click “Add Data Series.” Click “Edit Graph,” use “Customize data” to search for “Current payments to the rest of the world: Imports of goods,” and click “Add.” Input the formula a/b*100 and click “Apply.”

Suggested by Maximiliano Dvorkin and Melanie LeTourneau.

The takeaway

Although data center construction and semiconductor purchases are both related to the AI boom, changes in their prices may be occurring at different stages of the investment process.

Producer price index

Our FRED graph above shows the producer price index for semiconductor and other electronic component manufacturing. This index measures prices received by domestic producers in that industry. For much of the recent past, the index had changed slowly, even during periods when semiconductor shortages received wide attention. But that pattern has changed: The index rose from 61.6 in January 2026 to 73.1 in May 2026, an increase of about 19% in four months.

A question of timing

Increased data center construction has been expected for some time. A study by Kalyani and Li (2026) finds that US business spending related to AI grew substantially in 2025: Information processing equipment, software, and data center construction accounted for one-third of total business investment in the third quarter of 2025, the highest share since 1947. In other words, the boom in AI-related capital spending appears to have materialized and is no longer merely expected. But if firms were already planning large investments in AI infrastructure, why would semiconductor prices rise only now?

One possibility is that data center construction and semiconductor demand may arrive at different points in the investment process. Early spending on data centers includes land, buildings, cooling systems, power connections, and other physical infrastructure—costs that aren’t measured by this semiconductor price index.

Semiconductor price pressures may appear later, when planned data centers are closer to being equipped and firms have a better sense of which inputs are scarce. At this stage, firms need processors, memory, networking equipment, power-management chips, and many other electronic components. Recent earnings reports from semiconductor firms suggest that AI data center demand has started to reach a wider set of chip producers and not only firms that make the most advanced AI processors.

This interpretation helps explain why semiconductor prices may not have moved much during earlier stages of the AI boom.

Also, supply bottlenecks and capacity constraints in the investment and construction process aren’t always obvious in advance. As projects move forward and orders become more concrete, firms can learn more about where supply can expand easily and where price pressures may emerge for specific components.

Inventories

A surge in expected demand doesn’t necessarily raise prices immediately if suppliers can increase production or if inventories are available. In fact, some semiconductor markets were working through excess inventories after the pandemic-era cycle. Once these inventories have fully normalized and data center orders have strengthened, the boom could begin to fully and clearly show up in the price index.

How this graph was created: Search FRED for “Producer Price Index by Industry: Semiconductor and Other Electronic Component Manufacturing.” Select the monthly, not seasonally adjusted series. The series ID is PCU33443344.

Suggested by Aakash Kalyani.