US AI-related imports rise 62%: compute and networking capture almost all growth

US imports of AI-related goods reached $658.76 billion in January–August 2026, up 62.01% year on year. Compute hardware and networking supplied 99.67% of the $252.14 billion net increase. These figures are recalculated from the same vintage of monthly dollar series in the Minneapolis Fed AI Trade Tracker’s October 8 update.^[1][2]

The commercial priority is to identify suppliers delivering compute and network equipment and converting incremental sales into gross profit and cash, then examine power and cooling contracts individually. Global purchasing growth supports business opportunities but does not identify Chinese companies’ order shares, profit shares or value added in US end equipment.

Almost all the $252 billion increase comes from compute and networking

The table compares January–August in 2025 and 2026, in nominal USD billions. Growth and contribution rates use source monthly values with four decimal places; displayed figures use two. Contributions can be negative. Source rounding creates a very small difference between the seven components and the aggregate.^[2]

Compute contributes $214.09 billion of incremental imports and networking $37.20 billion. Electrical power adds just $2.55 billion; cooling/HVAC declines $0.99 billion. Incremental import dollars are highly concentrated. Applying the aggregate AI growth rate to power, cooling or construction suppliers would conceal their individual procurement cycles and revenue sources.

Two bridges separate import growth from supplier profit

The tracker starts with Census HS10 commodity data, classifying products by relevance to data-center construction and operation across seven infrastructure categories. It identifies relevant goods rather than confirming each final buyer or AI use. Some goods serve other industries, and model classification has errors. The baseline research also excludes HS27, 71, 98 and 99, so a share of its sample cannot be relabeled as a share of all US imports.^[1][3][4]

August AI-related imports were $99.05 billion, up 82.12% year on year and down 2.66% sequentially. Nominal dollars incorporate prices, equipment specifications and mix; the monthly series is not seasonally adjusted. A shift toward expensive systems can raise import value without proportionate server-unit growth, and cannot be converted directly into newly energized megawatts.^[2][4]

The first bridge links imports to a supplier’s actual deliveries. Systems vendors, distributors and end users may purchase at different times; inventory accumulation can precede deployment. The second links sales to gross profit and cash. Higher prices may largely reflect materials or expensive components rather than a larger retained margin. Inventory, receivables and expansion spending can then make cash lag revenue. This is a commercial inference from the procurement chain; aggregate trade data provide neither supplier margins nor collection terms.

Lower cooling imports do not establish weaker liquid-cooling demand

Cooling/HVAC imports fell 3.00%, but the category covers broad uses, omits US domestic production and does not isolate specific liquid-cooling orders. Falling imports can coexist with domestic substitution, mix changes or different project-delivery schedules. This result cannot establish a decline in US AI liquid-cooling demand or forecast any individual cooling vendor’s revenue.^[2][3]

Electrical-power import growth of only 2.62% likewise does not establish that power constraints have eased. Dollar flows measure goods; bottlenecks depend on required specifications, qualification, commissioning and delivery. A relatively cheap component without an accepted substitute can postpone system acceptance. Dollar rankings help screen industry scale, but cannot independently rank project risk.

On May 26, Schneider Electric announced phased delivery of more than $290 million in infrastructure solutions at TeraWulf’s Lake Mariner campus, including UPS systems, lithium-ion batteries, CDUs, in-rack manifolds and racks. The example illustrates coordinated power, cooling and rack delivery. The total provides neither country-level component sourcing nor shares for unconfirmed Chinese subcontractors. Each vendor’s revenue still depends on its contract and performance progress.^[7]

Aggregate tariffs conceal different equipment cost exposures

August’s effective tariff rate was 2.85% for AI-related goods overall, about 0.75% for compute, 9.55% for electrical power and 9.93% for cooling. These are realized category aggregates from the same data vintage, not statutory rates for an individual company or item.^[2]

Low aggregate AI duties do not make power and cooling procurement equally cheap. Project models must examine origin, classification, commercial terms and who bears the cost. A contract allowing pass-through may protect supplier margin while increasing project funding needs. A fixed-price contract requiring the supplier to absorb the cost would pressure profit first. These are contractual scenarios, not claims about any named company’s pricing rights.

Chinese exposure requires separate origin, ownership and value-added measures

Census import origin is the location of production or substantial transformation under the applicable rule, rather than shipping location or corporate headquarters. Goods made at a Chinese company’s foreign plant may qualify as local-origin goods, but foreign origin does not automatically establish Chinese ownership or the share of Chinese inputs.^[5]

Delton’s September 15 investor discussion said Thailand’s PCB supplier ecosystem was less mature than China’s. It uses group procurement and logistics, develops local suppliers and holds safety stock for key materials. This is a concrete example of manufacturing capabilities extending overseas. It does not identify a Thai-plant shipment into the United States or quantify Chinese material content. Procurement expertise and processes can extend across borders, while inventory and logistics costs grow with them.^[6]

Company research should separately track US customer orders and revenue, overseas qualification and yields, and material sourcing and cash commitments. Overseas group sales cannot simply be added to Chinese exports: intermediate inputs and final equipment may be counted twice. Chinese value added requires material and processing-stage evidence. The current trade series contains months and broad product categories, without country, firm or final-buyer dimensions, and therefore cannot provide a current Chinese-supplier share.^[2][4]

For a board maker or equipment vendor able to deliver across plants, customer qualification and supply-chain organization may improve access to orders. For a supplier lacking local service or reliable spares, a low factory price may not win a systems purchase. An investment case requires customer acquisition, delivery economics and collections; Chinese supply chains do not share one common margin.

US capacity additions change the relationship between imports and demand

On March 25, 2025, Schneider announced plans to invest more than $700 million in US operations through 2027, including switchgear and power-distribution expansion for data centers, utilities, manufacturing and other markets. The plan is not entirely AI spending and does not mean the capacity is already completed.^[8]

If those factories ramp successfully, slower import growth could reflect domestic substitution rather than weaker end demand. If construction, qualification or installation capacity lags, specifications can remain scarce even while aggregate imports rise. For overseas suppliers, a strong order window still has to generate adequate returns on invested assets before and after substitute capacity arrives.

Chinese-owned overseas plants face a similar return question. Longer logistics chains and safety stock consume cash; a maturing local supplier base may later improve lead times and reduce some cross-border procurement. Customer acceptance, effective capacity and cost determine the value of the overseas operation. Policy and origin labels alone do not guarantee profit.

Watch procurement mix and delivery returns over the next two to four quarters

First, track new months on a consistent definition: does compute and networking concentration persist, and how much reflects price and configuration? Power and cooling require company project orders, lead times and domestic-output evidence rather than an import growth rate alone. For Chinese companies, check completed qualification, shipments, overseas earnings and operating cash instead of merely counting foreign factory floor space.

Continued strong hardware purchasing accompanied by supplier margin and cash gains would show demand converting more clearly into business returns. If import growth comes chiefly with rising costs, stock and capital spending, revenue strength can coexist with weaker capital efficiency. For US projects, system acceptance and energization still matter. Equipment crossing the border completes only one procurement stage; it does not establish that the site can provide compute services.