Census Bureau reports July capital-goods imports of $140.117 billion
Key takeaways
- The Census Bureau's July advance report puts the goods deficit at $118.806 billion and capital-goods imports excluding automotive vehicles at $140.117 billion.
- July capital-goods imports were 15.20% above the January-June 2026 monthly average of $121.6265 billion. The category includes far more than AI equipment.
- Amazon funded $96.3 billion of first-half 2026 capex in cash, while J.P. Morgan Asset Management estimates $194 billion of hyperscaler bond issuance in the first half.
- The next test is whether August and later releases keep the import category elevated while financing demand remains absorbable.
The U.S. Census Bureau's advance report, published August 27, puts July goods imports at $318.177 billion, up from $306.813 billion in June. Capital-goods imports excluding automotive vehicles rose to $140.117 billion from $125.886 billion. The release measures goods flows, not AI demand, domestic data-center spending or revenue.
That distinction changes the underwriting question. As of this issue, the latest completed pre-July comparison was the January-June total of $729.759 billion, or a monthly average of $121.6265 billion. July was $18.4905 billion, or 15.20%, above the latest available baseline and 46.9% above July 2025. The figures establish an equipment bill; they do not identify its AI share.
These are different measures and should not be combined into an AI-spending total. Amazon's $96.3 billion of first-half cash capex is the quantified funding headroom for one buyer, not an industry-wide buffer.
The trade category is broader than the AI thesis
The supplied $88.6 billion goods-and-services deficit is not reconciled with the official July goods-only release and is excluded. Census reports no AI-only cut; the category includes computers, semiconductors, telecom equipment and other capital goods.
| Measure | Value | Basis |
|---|---|---|
| Goods deficit | $118.806bn | July 2026 goods balance |
| Capital-goods imports ex automotive | $140.117bn | July 2026 broad end-use category |
| Category baseline | $121.6265bn/month | January-June 2026 average |
| Amazon cash capex | $96.3bn | First-half 2026 company spending |
The release measures physical purchasing, not installation, utilization, power availability or willingness to pay for the resulting compute.
Financing absorbs the next phase
J.P. Morgan Asset Management reports $194 billion of hyperscaler bond issuance in the first half of 2026 and estimates $279 billion for the year. It cites technology spreads above broad investment grade and a later 1.4-times order book versus 3.5-to-4.5-times earlier. That is changing absorption, not a market-wide seizure.
The Bank for International Settlements describes financing through special-purpose vehicles, leases, offtake commitments and guarantees; risk can migrate across developers, funds, insurers, banks and hyperscalers. The U.S. Treasury's September 2 curve showed 4.39% for two-year, 4.79% for ten-year and 5.27% for thirty-year par yields.
Amazon's cash capex is a material buffer for one buyer, but it includes fulfillment and cannot serve as an industry offset. Federal Reserve Governor Christopher Waller said data-center plans point to continued rapid AI-related investment, with policy still conditional on inflation.
The next releases decide whether the bill is durable
The closest historical reference is Waller's 2021-22 forward-guidance episode, when the two-year Treasury yield rose about 200 basis points by March 2022. It is a duration precedent, not a resolved AI-financing comparison.
Three paths are investable. If continued buildout keeps at least three of the next four capital-goods readings 10% above $121.6265 billion and identifies a meaningful computer, semiconductor or telecom contribution, equipment suppliers retain the demand signal. If two successive readings fall below the baseline while hyperscalers rely mainly on operating cash, financing intermediaries lose urgency. If technology spreads widen and new deals draw order books near or below the reported 1.4-times comparison, long-duration projects face absorption stress.
Treasury investors carry duration exposure; semiconductor and utility suppliers carry the demand read-through; levered data-center developers and long-duration software carry refinancing risk. CTOL desk analysis assigns a 60% probability to the continued-buildout path through December 31, 2026. The September 30 advance report, covering August, is next.