The Capex Comeback Is One Line Wide
Business equipment investment is up 17.4% since the last quarter of 2024. One of the four lines in the national accounts supplied about 97% of that gain. Whether this is a capex cycle or a product cycle is a question the next GDP print can start to settle.
Figures are BEA national accounts (NIPA Tables 5.3.5 and 5.3.6) as revised in the second estimate of Q2 2026, released August 26, pulled September 20. The third estimate and annual update land September 30; the Q3 advance estimate lands October 29.
How to read it. Four lines, one origin. Each is a component of real business equipment investment from the national accounts, indexed so that the fourth quarter of 2024 equals 100; the vertical axis is that index, from 90 to 150, and the horizontal axis runs seven quarters to the second quarter of 2026. The heavy gold line is information-processing equipment; the black line is total equipment; the two faint lines are industrial and transportation equipment. Read the vertical distance between lines at the right edge: the gold line ends at 141.9, the total at 117.4, and the two faint lines at 110.0 and 100.5. The one thing it shows is a single component leaving the others while the total follows it.
The verdict, first
Two claims about American capital investment are in circulation this month, and they point in opposite directions. One says the capex comeback is broadening out across the economy. The other says capital is starting to leave the United States for other destinations. Both are argued from totals. Neither uses the one table that tests them, which is the Bureau of Economic Analysis's breakdown of what businesses actually bought.
Why we chose this question. It decides both claims at once, and it is falsifiable on a date. "Broadening" is a statement about the three lines other than information processing, and the Q3 advance estimate on October 29 will report all four. "Capital leaving" is a statement about whether foreign money is still arriving, and the answer, from the Treasury's own capital-flow data, is that foreign investors bought more American equities and corporate bonds in the year to July than in the year before. A claim about breadth can be scored on a date. The larger claims on either side cannot.
What is said: equipment investment is up strongly since the change of administration, and the recovery has spread beyond its origin. Or, from the other side, that investors are cooling on America.
What the accounts show: real business equipment investment rose 17.4% between the fourth quarter of 2024 and the second quarter of 2026. That is a large number and it is real. But the accounts split equipment into four lines, and one of them, information-processing equipment, rose 41.9% and accounts for roughly 97% of the total gain. Industrial equipment rose 10%. Transportation equipment rose half a percent. The residual category fell. Take information processing out and the rest of equipment investment is up about 1% in six quarters.
What the flows show: on the second claim, the Treasury's own capital-flow data answer directly. Foreign investors' net purchases of long-term American securities were $1.75 trillion in the twelve months to July, up from $1.47 trillion the year before. What changed is the mix: $1.39 trillion of it went into equities and corporate bonds, while net buying of Treasury notes and bonds fell to $247 billion from $456 billion. The case that capital is leaving has its own facts on the same table. In July alone private foreign investors were net sellers, of $29 billion of Treasury notes and bonds and $11 billion of equities, and official institutions were the buyers; the Treasury's bottom-line total, which adds short-term paper and bank liabilities, fell to $1.31 trillion from $1.68 trillion over the twelve months; and American purchases of foreign securities rose to $478 billion from $286 billion. Twelve months of long-term buying say the money is still coming; the latest month says it paused. This note reads the year: capital is not leaving, and within the year it has moved from the government's paper to equities and corporate bonds.
The overall point, stated once. Since the end of 2024 the United States has had an investment boom in one category of hardware: the computers, servers, networking gear and data-center equipment of the AI buildout. It has not had a capex cycle in the ordinary sense, where firms across industries buy machines, vehicles and plant. The boom is large, real, and one line wide. That reading answers both circulating claims, and flatters neither: the money is being spent here, by foreigners among others, and it is being spent on one thing.
The case that this is a broadening capex cycle, in its strongest form. Four arguments, and they are serious.
AI hardware is capex. Servers and networking gear are capital goods bought by businesses to produce output. A cycle led by one category is still a cycle; the 1990s equipment boom was also led by computers, and it was the largest sustained investment expansion of the postwar period.
The second line is moving. Industrial equipment rose 10% in real terms over the span, and its most recent quarter, up 6.3% on the quarter, was by far its strongest. Transportation rose 6.2% in the same quarter. Information processing supplied about a quarter of that quarter's gain, not 97% of it. A broadening cycle starts in one line and shows up in the others later, and the latest quarter is what that would look like.
The other lines were held down by things that have passed. Transportation equipment was flat for five quarters while tariffs raised the cost of vehicles and of the steel and aluminum that go into aircraft, and rates stayed high through 2025, then rose 6.2% in the second quarter of 2026. Those were headwinds to the non-AI lines, not evidence that firms outside technology do not want to invest.
Orders point up. Manufacturers' orders for nondefense capital goods excluding aircraft, the monthly series that leads equipment investment, rose from $79.3 billion in December 2025 to $85.7 billion in July 2026. That series includes information-processing goods, but it also includes machinery, and it has not turned.
The test below is written against these four. The first is a definitional point, and the note concedes it: this is investment. The second and fourth are the ones the next two prints can score. The third cannot be settled from the accounts, and we say so. The 1990s comparison inside the first cannot be run in real terms, because the component series begin in 2007; section 2 runs it in nominal terms instead.
The four lines
The BEA's equipment category has four components. Here is each one from the last full quarter of the previous administration to the latest reported quarter, in real terms, with the nominal figure alongside.
| Equipment line | Q4 2024 | Q2 2026 | Change | Real % | Nominal % |
|---|---|---|---|---|---|
| Information processing | 530.2 | 752.1 | +221.9 | +41.9% | +53.8% |
| Industrial | 256.8 | 282.4 | +25.6 | +10.0% | +15.9% |
| Transportation | 301.0 | 302.6 | +1.7 | +0.5% | +2.9% |
| Other (residual) | about −19 | falling | |||
| Total equipment | 1,320.6 | 1,550.5 | +230.0 | +17.4% | +24.1% |
Billions of chained 2017 dollars. Chained-dollar components do not sum exactly, so the residual and the shares are approximate; the direction is not in doubt.
Two qualifications belong next to the table. In nominal dollars the split is less extreme: information processing supplied about 77% of the nominal gain, and the rest of equipment is up about 8.6%; the difference between the two conventions is the price deflator on information-processing goods. And the latest quarter alone had a different shape. In the second quarter of 2026 industrial equipment rose 6.3%, transportation rose 6.2%, and information processing rose 1.7%, so that the one line supplied about a quarter of the quarter's real gain. Six quarters say one line; the most recent one says otherwise. Both are in the exhibit.
Real business equipment investment excluding information processing, fourth quarter of 2024 to second quarter of 2026. The total rose 17.4% over the same six quarters. The difference is one line.
Three things follow. First, the headline is honest: 17.4% real growth in six quarters is a fast pace by any recent standard. Second, the composition does not support it: one line supplied nearly all of it, and the two lines that a broad capital-spending cycle would show, industrial and transportation, together added about 27 billion chained dollars against information processing's 222. Third, the nominal figures are higher than the real ones on every line, which means equipment prices rose over the span. Information-processing prices, which usually fall, rose too.
Equipment is one of three legs of business fixed investment. Over the same six quarters structures fell 7.1%, intellectual-property products rose 14.0%, and the three together rose 10.5%. A reader who counts software and research as capital spending will find some of the broadening there; that leg is not this note's test. The note stays on the equipment leg because it is where the "comeback" claim is made and where the accounts are most specific.
The reference class
Concentration on its own proves little. Every equipment cycle since the accounts began has had a leading line. The question is how concentrated this one is against the others the same data can measure. The component series begin in 2007, which allows three earlier six-quarter windows, each starting at a turn in the cycle.
| Six-quarter window | Total equipment | Information processing | Share of the gain from information processing |
|---|---|---|---|
| Q4 2010 to Q2 2012, after the financial crisis | +19.5% | +11.7% | 18.6% |
| Q4 2017 to Q2 2019, after the 2017 tax act | +4.9% | +8.4% | 57.2% |
| Q4 2020 to Q2 2022, after the 2020 recession | +4.4% | +11.5% | about 103% |
| Q4 2024 to Q2 2026, this note | +17.4% | +41.9% | 96.5% |
How to read it. Four rows, one span each. Every row is a six-quarter window of real business equipment investment starting at a turn in the cycle; the first two figure columns are the percentage change in total equipment and in information-processing equipment over that window, and the last is information processing's chained-dollar change divided by the total's, stated as a percentage and approximate. Compare down the last column: the lower the share, the broader the window. The one thing it shows is that the current window is the only one with both a large total and a share near 100.
The 2010 window is what a broad recovery looks like in this data: total equipment up almost 20%, with information processing supplying less than a fifth of it and industrial and transportation the rest. The 2017 window, the last time the phrase "capex comeback" was in general use, produced a 4.9% rise with information processing supplying more than half. The pandemic window produced a similar total with information processing supplying all of it and the other lines net negative.
The 1990s cannot be added to the table, because the real component series begin in 2007, but the nominal series reach back and give the proponents' reference class a number: from the end of 1994 to mid-2000 nominal equipment investment rose 54%, and information processing supplied 44% of the gain. That was a computer-led cycle in which the other lines still carried more than half.
The current window matches the 2010 window on size and the pandemic window on shape. It is the only one of the four with both a large total and near-total concentration. That combination is what an AI-hardware cycle would look like in the national accounts, and it is not what a broadening one would look like.
What it changes
The reading bears on two live questions.
The first is what the Federal Reserve's rate increase on September 16 does to the build. The statement named robust capital investment as one reason the rate could rise. If that investment is one line wide, then the sector the hike leans on is the sector with the deepest pockets and the most contracted demand, and the sectors a hike would ordinarily slow, machinery and vehicles, are already flat. The hike is less of a brake on the boom than the aggregate suggests, and more of a brake on the parts that are not booming. Our note on the hike, One Rate, Three Landings, takes up the issuer side of that argument.
The second is what the flows in the verdict mean for the Treasury market. Foreign investors in aggregate, who are funding American equities and corporate bonds, have cut their net purchases of Treasury notes and bonds by almost half in a year. The claim that capital is leaving America is wrong on foreign purchases of American securities over the year and right about one market: foreigners bought $209 billion less of Treasury notes and bonds and $344 billion more of equities. The table does not say these are the same investors, or that the equities bought belong to the companies doing the building; on its own split, over two-fifths of the added equity buying came from official institutions, which were not the Treasury buyer that was lost.
Verdict and tripwires
The question was whether the post-2024 rise in equipment investment is a capex cycle or a product cycle. From the accounts: a product cycle, so far. The total is up 17.4%. Information processing is up 41.9% and supplied roughly 97% of the gain. Everything else is up about 1%. The one prior window of similar size in this data, after 2010, had the opposite shape. That is a description of six quarters, not a forecast. The latest of the six reads the other way, with the other lines up about 6% and information processing supplying a quarter of the gain, and it is the strongest argument that the shape is changing. The tripwires are written to catch it if it does.
Tripwires (armed):
- FOR BROADENINGCX-1 The other lines move. In the Q3 advance estimate on October 29, real industrial plus transportation equipment rises at an annualized rate of 4% or more, and does so again in the Q4 advance estimate in late January 2027.
- FOR ONE LINECX-2 The line turns and the total follows. Real information-processing equipment falls quarter on quarter and total equipment falls with it, while the other lines are flat.
- FOR BROADENINGCX-3 The line turns and the total holds. Real information-processing equipment falls quarter on quarter and total equipment still rises.
- FOR ONE LINECX-4 Orders roll. Nondefense capital goods orders excluding aircraft decline for three consecutive months from the July 2026 level of $85.7 billion.
CX-1 requires two prints and CX-2 and CX-3 one, because a single quarter of the two smaller lines is noisier than a single quarter of the largest; the second quarter of 2026 is the case in point, and it is why CX-1 is written to need a second. If Q3 and Q4 both print with information processing rising and the other lines flat, none of these fires; the reading stands unscored and is re-scored at the Q4 advance estimate. CX-4 is a nominal series that includes information-processing goods, so it is a weaker test than the quarterly accounts and is scored second; a rise in orders is not scored the other way for the same reason, since it cannot say which line is ordering.
Not a tripwire: the total equipment figure on its own. A large total is consistent with both readings; it is the split that decides.
What would change our view. We abandon "one line wide" if CX-1 or CX-3 fires, since either would show the other three lines carrying growth on their own. We would also revisit the note before either print if the BEA's annual update on September 30 revises information processing's share of the six-quarter gain below 85%, which would mean the concentration in the second estimate was partly a measurement artifact. In the other direction, CX-2 would confirm the reading but would not by itself say anything about the AI buildout's economics; a category that stops growing is not a category that failed. On the flows, we would revise "capital is not leaving" if the next two Treasury capital-flow releases, for August in mid-October and September in mid-November, both show net foreign sales of long-term American securities, or if the twelve-month figure falls below the prior year's $1.47 trillion. This note does not take a view on whether the one line is well spent. Our AI-capex series does.
Sources & method
Accounts. Bureau of Economic Analysis, National Income and Product Accounts, Table 5.3.5 (private fixed investment by type, billions of dollars) and Table 5.3.6 (real private fixed investment by type, billions of chained 2017 dollars), as published in the GDP second estimate for Q2 2026 on 2026-08-26. Series retrieved through FRED on 2026-09-20: total equipment (Y033RX1Q020SBEA, Y033RC1Q027SBEA), information-processing equipment (Y034RX1Q020SBEA, Y034RC1Q027SBEA), industrial equipment (A680RX1Q020SBEA, A680RC1Q027SBEA), transportation equipment (A681RX1Q020SBEA, A681RC1Q027SBEA), nonresidential fixed investment (PNFIC1), structures (B009RX1Q020SBEA), intellectual-property products (Y001RX1Q020SBEA). Retrieval script and outputs archived with the note.
Shares. Chained-dollar components are not additive. A component's share of the gain is its chained-dollar change divided by the total's chained-dollar change, and the "other" line is the residual; both are stated as approximate. The ex-information-processing figure is the change in total minus information processing, on the same basis.
Comparison windows. The same six-quarter length applied after three earlier turns: Q4 2010, Q4 2017, Q4 2020. The real component series begin in 2007, which rules the 1990s out of the table; the 1994 to 2000 figure is nominal (Table 5.3.5). Latest-quarter and nominal splits are computed in a second archived script.
Orders. Census Bureau, manufacturers' new orders, nondefense capital goods excluding aircraft (FRED NEWORDER), monthly, nominal, through July 2026.
Capital flows. Treasury International Capital data for July 2026 (2026-09-16 release), table "TIC Monthly Reports on Cross-Border Financial Flows," columns "12 Months Through Jul-25" and "Jul-26": line 3 (domestic securities, net U.S. sales) for the totals; lines 5 and 10 for Treasury bonds and notes, 7 and 12 for corporate bonds, 8 and 13 for equities, private and official summed; line 16 (sign reversed) for American purchases of foreign securities; line 30 for the bottom-line total; the July column for the single-month figures. The table and the parse script are archived with the note. Federal Reserve statement of 2026-09-16 for the capital-investment sentence.
What this note does not do. It does not value any company, name any issuer, or forecast equipment investment. It does not assess whether AI hardware spending will pay back; the house's AI-capex notes do that.
Disclosures
Information only. TON618 Capital. This report is for information purposes only. Nothing here is an offer to sell or a solicitation of an offer to buy any security, fund interest, or digital asset, and nothing here is personalized investment advice or a recommendation regarding any instrument.
Publisher's exclusion. All research is published solely as general, impersonal information of regular circulation. It is not tailored to the objectives or circumstances of any individual and is not issued in connection with compensation from any client. The Fund has no clients and distributes all research free of charge. On that basis it publishes in reliance on the publisher's exclusion from the definition of "investment adviser" under the Investment Advisers Act of 1940 (§202(a)(11)(D); cf. Lowe v. SEC, 472 U.S. 181 (1985)).
Registration & conflicts. TON618 Capital is not registered as an investment adviser or broker-dealer in any capacity. The Fund is a Bitcoin fund and may hold or transact in the securities or digital assets it discusses; material conflicts are disclosed where they exist. Ownership: this report names no issuer; the Fund holds no position in any instrument it discusses as of the report date, and none of the series discussed is a Fund holding or a BTC-correlated instrument. Compensation: the Fund received no compensation from any party in connection with this report and charges nothing for it.
Use of AI. Artificial intelligence is used in the creation of this research. All methodology and data integrity are reviewed and approved before publication by TON618 Capital's Chief Investment Officer, Keyth Beck; errors may nonetheless occur, and readers should verify independently.
CFA. This report was prepared to align with CFA Institute analytical standards (methodology only). CFA® and Chartered Financial Analyst® are registered trademarks owned by CFA Institute. That reference describes the analytical framework applied; it does not imply the report was prepared, reviewed, or authored by a CFA charterholder, and the report is not issued, reviewed, endorsed, certified, or approved by — nor affiliated with — CFA Institute.
Risk & feedback. Past performance is not indicative of future results. Digital assets and equities are volatile and may result in total loss of capital. Corrections and feedback are welcome — please direct them to CIO Keyth Beck at keyth@ton618capital.com. Version 1.0.