Is AI boom redrawing the global air freight map?

Over the past few years, if there is one thing the world has learned, it is how much we depend on supply chains, and how quickly they can be disrupted.

We saw vaccines, PPE kits, masks, medicines and medical equipment being moved across borders during the pandemic. Then came wars and the movement of arms, grains and food; the surge in cross-border e-commerce parcels; tariff-driven frontloading; the closure of the Strait of Hormuz; and continuing tensions in West Asia. Through all of this, air cargo has remained an important part of the response, alongside its traditional flows of general cargo, perishables, flowers and industrial equipment among others.

That got me thinking: what is driving air cargo now, and more importantly, what could drive it next?

So I asked my AI assistant.

And honestly, the answer caught me by surprise as the AI agent suggested that the next wave of freight growth may not come from consumer goods, retail replenishment or e-commerce. It may come from something far more complex: the physical infrastructure required to power artificial intelligence.

This sentiment is also reflected among industry experts. “While e-commerce demand is cooling, AI-driven freight is booming, particularly on the Transpacific. I cannot see the e-commerce growth engine being revived. There will always be a consumer demand for cheap goods manufactured in Asia, but the extraordinary demand growth of recent years will not be sustained. E-commerce was air freight's single biggest growth pillar, but that is no longer the case,” stated Niall van de Wouw, Xeneta Chief Airfreight Officer, in an official statement in July.

The cargo behind this shift is also different. GPUs (graphics processing units, which are specialised chips required to do many calculations in parallel, especially for graphics and video), servers, semiconductor components, networking equipment, memory chips, power systems, cooling equipment and fully assembled racks are moving across Taiwan, South Korea, Southeast Asia, India, Mexico, North America and Europe. Much of this cargo is high-value, oversized, export-controlled or time-critical.

The importance of air cargo to this emerging trade was evident in 2025. According to IATA, more than two-thirds of the value of AI-related trade was carried by air. Air cargo consignments of AI-related goods grew 20% year-on-year, while AI-related goods accounted for 53.5% of the total value of air-transported trade despite representing just 7% of its volume.

The momentum continued into June 2026. Xeneta reported 7% year-on-year growth in global air cargo demand, driven by semiconductor and AI-related hardware. Spot rates rose 38% year-on-year to $3.40/kg, although growth eased from 41% in May. On Asia-Pacific–North America routes, AI volumes continued to offset weaker e-commerce demand, with rates on Northeast Asia and Southeast Asia routes 41% and 42% higher, respectively, than in late February.

As Kathy Liu, VP, Global Sales and Marketing at Dimerco Express Group, puts it, “What we’re seeing is a market split in two. AI demand out of Taiwan just keeps climbing, while the e-commerce base that carried Europe is gone with the de minimis change.”

AI demand is tightening key airfreight lanes
Dimerco's August 2026 Asia Pacific Freight Report shows pressure building across key Asian airfreight markets. Taiwan's airfreight capacity remains tight, with rising rates to Asia and both US coasts as demand for AI servers, semiconductors and other high-tech products increases. South Korea is seeing similar pressure, with tight capacity to Asia and the US and rising rates across major corridors.

Dimerco, which works with Cathay Cargo, Air China Cargo, China Airlines, EVA Air, Korean Air, Starlux Airlines and Singapore Airlines for AI infrastructure movement, reports that load factors on Asia-US lanes have reached approximately 90%, with AI and semiconductor shipments replacing e-commerce as the primary capacity driver. This demand is being generated by a wide range of equipment required to build and operate data centres. “A data centre needs around nine categories of hardware, compute, networking, storage, racks, power, cooling, facility systems, security and AI-specific components, and they have little in common,” Liu says.

That variety creates different logistics requirements. GPUs are relatively small but extremely valuable and subject to export controls. Transformers and chillers are oversized project cargo, while UPS batteries are classified as dangerous goods. Fully outfitted server racks present another challenge. Liu says a fully outfitted rack costs close to $4 million and takes up significant airfreight volume once crated. Dimerco recently handled 320 server racks on a tight timeline, which required the main deck capacity of multiple B747 freighters.

Planning begins before the cargo reaches the airport
Transportation planning starts before cargo is handed to an airline. Dimerco states that GPUs, control systems and fully assembled racks move by air, while cabling and empty racks can move by ocean. For oversized racks, pallet configuration and ULD build-up are planned before booking. But, all these shipments are highly delicate in nature. So, Dimerco conducts inspections before pickup, verifies packing lists and records photographic documentation.

“AI demand out of Taiwan just keeps climbing, while the e-commerce base that carried Europe is gone with the de minimis change.”

Kathy Liu, Dimerco Express Group

While ShockWatch and tilt indicators are used during transit, planning continues at the destination too. So, regional hubs in Singapore and San Francisco allow equipment to arrive according to a data centre's build sequence. Liu says these requirements are planned at the booking stage.

“Reliability now matters more than cost. A two-day delay to a data centre launch can represent millions in lost opportunity,” Liu adds.

How is the AI supply chain moving?
Reports show that Asia-to-North America remains a major flow for these kinds of shipments, but finished equipment represents only a part of the demand. Components move between semiconductor fabs, outsourced semiconductor assembly and test (OSAT) operations in Southeast Asia and Taiwanese server integrators.

Kuehne+Nagel provides another indication of the shift. Its Air Logistics business reported 20% higher Q2 net turnover, supported by semiconductor demand, cloud infrastructure shipments and Southeast Asian exports excluding China. Airfreight volumes reached 1.1 million tonnes in H1 2026, while the company also managed Google cloud infrastructure shipments from Asia to the US.

“When you look at the hyperscalers such as Meta, Google, Microsoft and Amazon, they're talking about investing hundreds of billions of US dollars every year in AI infrastructure,” said James Evans - General Manager Cargo Commercial, Cathay Cargo.

The numbers point to a market that is becoming structurally more specialised. Mordor Intelligence estimates the semiconductor logistics market at $79.32 billion in 2025, rising to $86.55 billion in 2026 and $133.87 billion by 2031, representing a 9.12% CAGR.

However, a closer view at the geographical distribution of AI hubs reveals that Taiwan and South Korea have become important AI infrastructure manufacturing hubs outside the US. Liu says Dimerco has seen significant growth in the Korea-to-Taiwan trade lane, driven largely by memory chips and other semiconductor components used in AI servers. Samsung and SK Hynix supply high-value memory products, including high-bandwidth memory (HBM) and DRAM, to Taiwan, where server manufacturers integrate them into finished AI infrastructure for export.

“Chip shortages are expected to reduce smartphone, laptop, and consumer electronics volumes by 8–10%, not from lack of demand but from lack of chips to build the products.”
Asok Kumar, Expolanka Group and Morrison Express

“The trend highlights that AI-related logistics demand is not only reflected in outbound shipments from Asia to North America, but also in the increasing movement of components within Asia, particularly between Korea and Taiwan, before final assembly and export,” Liu says.

Reports show that Southeast Asia is increasingly gaining importance, with Malaysia, Vietnam, Singapore and the Philippines attracting assembly, testing and packaging investment and developing data centre markets. Thailand is among the region's tightest airfreight markets, with capacity tight and rates rising across Asia, Europe and both US coasts. Singapore has backlog conditions to Europe, while Kuala Lumpur and Penang remain tight on Asia- and US-bound lanes. India is seeing rising pressure to Europe and North America, while Mexico is emerging as a nearshore manufacturing and logistics leg into the US.

Clearly, this scale of AI shift is also increasingly gaining visibility in airline results. For instance, Korean Air reported Q2 2026 cargo revenue of KRW 1.5419 trillion ($1.07 billion), up 46% year-on-year, with AI-related cargo, including chips, server racks and data-centre infrastructure, replacing e-commerce shipments from China as a primary growth driver. Cathay Group’s 2026 interim results also mentioned, “Regional (North Asia) semiconductor shipments continued to underpin growth.”

From Cathay Cargo's perspective, the strongest AI-related cargo flows currently connect Southeast Asia and Northeast Asia with the Americas, particularly markets such as Taiwan and Japan. Explaining the scenario, the Cathay spokesperson added that while "China +1" has existed as a strategy for many years, the speed of manufacturing shifts over the past year has accelerated dramatically. Production has expanded across Thailand, Vietnam, Malaysia and other Southeast Asian markets, requiring airlines to react quickly. "We've rapidly redeployed capacity, restored freighter services, upgraded passenger aircraft and increased capacity into Southeast Asia."

“This is diversification rather than relocation. Companies are not leaving China. They are building parallel capability across Taiwan, Southeast Asia, India and Mexico,” Liu added.

Evans further pointed to market data illustrating just how quickly the sector is expanding. "High-tech exports from Taiwan grew around 60% between 2024 and 2025, and during the first five months of this year growth approached 120%. Our customers have experienced that growth directly, and we've had to reposition our network accordingly.”

The changing trade map is becoming visible across the region. Reuters reported that demand for advanced HBM chips and processors is extending two to three years into the future. Japan Airlines estimates technology products accounted for around 80% of the increase in air exports from Asia excluding China over the past year. At Singapore's Changi Airport, freight throughput increased 8.7% year-on-year in H1 2026, driven by strong semiconductor demand.

As a result, forwarders are having to adapt just as quickly. DHL Global Forwarding Japan and Kitakyushu Airport handled approximately 180 tonnes of semiconductor manufacturing equipment from Amsterdam Schiphol to Kitakyushu on a dedicated large freighter at the end of 2025. The operation, announced in March 2026, involved measures to manage temperature, humidity and vibration, with DHL having worked with Korean Air and local stakeholders since 2023 to develop the capability.

Supply chain diversification creates new logistics requirements
As manufacturing and assembly spread across more countries, logistics providers must manage additional trade lanes, customs regimes and production handoffs. AI infrastructure is particularly complex because, unlike conventional supply chains, data centre operations require tightly sequenced deliveries across multiple locations, often spanning Europe, the United States, Mexico, and Asia-Pacific, with hardware that demands specialised handling and zero-margin delivery windows, stated GEODIS.

And this is not simply about moving goods from one point to another. A single AI project can involve memory chips moving from South Korea to Taiwan, components passing through Southeast Asia, finished servers moving by air, oversized equipment moving by ocean and final delivery being synchronised with a data centre's construction schedule.

As Liu puts it: “Our role is to make the supply chain flow predictable and resilient, because delay is an expensive risk on these projects”. As a result, trade compliance is increasingly incorporated into supply chain design. Dimerco manages BIS requirements, restricted-entity screening, classification, documentation and bill of lading structuring to reduce inspection and compliance risks.

“A small classification error can mean delay, penalties or seizure,” Liu says. Bonded and free trade zone inventory provides flexibility, allowing equipment to be redirected between projects without unnecessarily triggering duties.

Warehousing becomes part of the logistics strategy
Dimerco supports the AI supply chain through semiconductor-grade warehouses in Taipei, Hong Kong, Singapore and San Francisco, alongside bonded and free trade zone facilities. Its vendor-managed inventory model, operated since 2008, is now being applied to AI projects, while MyDimerco provides visibility across more than 150 offices and 80 logistics facilities.

Together, these capabilities allow AI equipment to be managed from manufacturing and storage through transportation and final data centre delivery. The challenge is coordinating components, inventory, aircraft capacity, compliance, handling and final installation across multiple countries and timelines.

Tariffs are changing the trade compliance equation
However, the AI infrastructure boom is changing more than freight volumes. It is also changing how companies think about tariffs, compliance and accountability. Liu says the US regulatory environment is moving towards greater enforcement, including tighter qualification requirements for importers of record (IORs).

"High-tech exports from Taiwan grew around 60% between 2024 and 2025, and during the first five months of this year growth approached 120%.
James Evans, Cathay Cargo

That makes previously overlooked costs harder to ignore. “Duty drawback is more complicated, and when the duty amount involved is not yet at a level that justifies the effort, some companies choose to skip it. Now, we feel it may be more reasonable to revisit it as a way to help reduce cost,” Liu says. Dimerco recommends that companies revisit their trade compliance processes and assess whether establishing a US entity could make sense as requirements evolve. “Most importantly, revisit your company's trade compliance processes and setup, and consider involving professionals where necessary,” she says.

AI boom or another bubble?
While all the reports show growth of AI-infrastructure-related shipments, the question now is how sustainable that growth is. At transport logistic and air cargo Shanghai 2026 on June 24, Asok Kumar of Expolanka and Morrison Express said companies across cooling systems, data connectivity, server-rack assembly and semiconductor manufacturing were reporting production capacity booked two to three years ahead. But he also asked the question that is becoming harder to ignore, pointing to Taiwan's projected 9.8% economic growth in 2026.

“It scares me as well,” Kumar said. “This is not the growth of a mature economy. This is the growth of an emerging market. Taiwan is not an emerging market. Is all this leading to a bubble?”

Based on industry conversations, he suggested the market could remain relatively safe at least until the end of 2027, while warning that chip shortages could reduce smartphone, laptop and consumer-electronics volumes by 8–10%.

The pressure is also reshaping airports. Hactl is responding to competition from mainland Chinese gateways by retaining Hong Kong's focus on high-value, sensitive and pharmaceutical cargo while developing a lower-cost, barge-connected facility in Dongguan for standard freight.

And this brings me back to the question that started this story. We have seen the dot-com bubble, followed by the e-commerce boom and correction. AI is now driving another enormous investment cycle in chips, servers and data centres. But, as investment in AI infrastructure accelerates, so does public debate. Across multiple countries, people are increasingly opposing new AI data centre developments, citing concerns over energy and water consumption, environmental impact and the long-term sustainability of this rapid expansion. On the other hand, GEODIS also mentioned in a blog that reducing emissions in transportation, warehousing, and packaging has become a shared business priority.

For air cargo, the opportunity is clearly real. But so is the question hanging over it: how long can this extraordinary AI-driven growth continue, and when — or whether — will the bubble burst? If it does, the impact will not be limited to technology companies. It could reshape the freight networks, capacity decisions and logistics infrastructure being built around AI today.