Everything About AI
NextFin News - The artificial-intelligence boom is doing more than lifting trade: it is making global commerce depend on a narrower set of products, factories and power systems. World merchandise trade expanded 4.6% by volume in 2025, according to the World Trade Organization, but AI-enabling goods generated 42% of that growth while representing about 17% of merchandise trade by the end of the year. The tension is straightforward: the same investment wave supporting trade is concentrating the points at which trade can fail.
The concentration is visible in the geography. Asia supplied 62% of AI-enabling goods trade in 2025, compared with 16% for North America, the WTO said. The region’s lead reflects the clustering of semiconductor production and electronics supply chains, not simply faster consumer demand. In the United States, imports of automatic data-processing machines, including servers and high-performance computing equipment, accounted for about three-quarters of the 4% growth in merchandise imports in 2025, UN Trade and Development said. In China, growth in the same category almost offset the net decline across more than 5,000 other imported product categories.
That makes headline trade data unusually dependent on a capital-spending cycle. The International Energy Agency said capital expenditure by five major technology companies exceeded $400 billion in 2025 and is expected to rise another 75% in 2026. Data-center electricity demand grew 17% in 2025, while AI-focused data-center use rose 50%. The bottleneck is moving downstream: from chips and advanced packaging to grid connections, transformers, generation and construction capacity.
The trade rebound is therefore less broad than the aggregate number suggests. UNCTAD estimates merchandise-trade growth at 4.7% in 2025 but projects a slowdown to between 1.5% and 2.5% in 2026 as demand, investment and the AI surge cool. A boom concentrated in a few strategic industries can keep trade totals high while making the system more sensitive to a shipping disruption, export-control decision, power shortage or valuation shock.
The key question is not whether AI supports trade. It clearly does. The question is whether the trade system can diversify faster than AI demand concentrates it.
The Headline Boom Is Narrower Than the Trade Number
The first judgment is that AI has become a powerful but narrow engine of global commerce. The WTO’s 2025 figures show both sides of the story: world merchandise trade grew much faster than the organization’s earlier 2.4% forecast, yet a large share of the upside came from goods tied to one investment theme. AI-enabling goods rose 21.9% in value to $4.18 trillion from $3.43 trillion in 2024. Their share of global merchandise trade climbed from roughly 13% in 2022 and 2023 to nearly 17% at the end of 2025.
That rise contains a cyclical element. Electronics demand weakened in 2022 and 2023, so part of the rebound reflects inventory normalization and the recovery from a semiconductor downturn. But the WTO also identifies a structural expansion in demand for semiconductors, processors and other AI-enabling goods. The two forces matter because a cyclical rebound can reverse when inventories normalize; structural demand can keep capacity tight even after the first wave of buying fades.
The composition of the trade growth matters more than the aggregate. Asia contributed 3.2 percentage points of the WTO’s 4.6% global trade-volume increase in 2025, or 71% of the total. It held the largest share of AI-enabling trade because chip manufacturing, electronics assembly and related components are concentrated there. North America recorded faster AI-enabling-goods growth at 32.4%, versus 24.3% in Asia and 10.6% in Europe, but North America’s smaller base meant a 16% share of AI-enabling trade.
This is the first transmission channel: corporate capital expenditure becomes cross-border demand for servers, accelerators, memory, networking equipment and advanced manufacturing inputs. The second channel is geographic. The spending does not diffuse evenly across countries; it travels through existing clusters. That produces a larger trade impulse per dollar of investment, but it also creates a larger payoff to disruption at the cluster.
UNCTAD’s import data shows why this matters for interpreting economic resilience. In the United States, automatic data-processing machines accounted for about three-quarters of merchandise-import growth in 2025. In China, strong growth in the same category nearly offset declines across thousands of other categories. A trade expansion driven by servers can look healthy in the headline while leaving traditional goods, commodities and intermediate industries much less dynamic.
There is a misleading comfort in the breadth of the customs data. Thousands of products may cross borders, but the marginal growth dollar is increasingly tied to a smaller group of technologies.
Why the Chokepoints Are Multiplying
The second judgment is that AI is converting a semiconductor bottleneck into a network of linked bottlenecks. A high-end computing system needs more than a leading processor. It needs advanced packaging, high-bandwidth memory, networking, data-center construction, cooling, electricity and a grid connection. If any one of those inputs is delayed, the value of the other inputs can remain idle.
The IEA’s energy figures show the scale of the new constraint. Five major technology companies spent more than $400 billion on capital expenditure in 2025, with another 75% increase expected in 2026. Data-center electricity demand grew 17% last year; AI-focused demand grew 50%. The IEA expects total data-center electricity consumption to double by 2030 and AI-focused use to triple. Efficiency per AI task is improving, but usage and model scale are growing faster than efficiency in the near term.
“AI-related sectors remain dynamic. But the broader trade rebound already appears to be losing momentum,” UN Trade and Development said in its 2026 assessment of the global economy.
The mechanism is not simply “more AI uses more power.” It is that capital is arriving faster than physical infrastructure can be permitted, built and interconnected. Chips can be shipped internationally; electricity cannot. A server can move from one country to another, but a data center still needs local generation, transmission, cooling water and a connection queue. This makes power a regional trade constraint even when the hardware market is global.
The result is a cross-asset transmission. A shortage of advanced chips raises equipment prices and encourages inventory accumulation. A shortage of electricity raises the cost of data-center operation and pushes developers toward gas, renewables, storage and transmission investment. That can lift demand for energy infrastructure while increasing exposure to fuel prices, construction delays and regulatory approvals. The shock moves from technology equities into utilities, natural gas, industrial equipment and local public finance.
The same architecture also raises the cost of policy fragmentation. Export controls may slow a rival’s access to advanced processors, but they can redirect orders toward substitute suppliers, accelerate domestic capacity spending and increase the price of the next available unit. Trade policy therefore affects not just the volume of commerce but the redundancy of the network.
Asia’s 62% share of AI-enabling trade is not evidence of weakness by itself. Clusters are productive precisely because suppliers, engineers and customers are close enough to reduce coordination costs. The vulnerability comes when efficiency is mistaken for resilience. A system optimized around a few clusters carries less spare capacity and fewer substitutes.
The Structural Shift Is Real, but the Spending Wave Is Cyclical
The correct call is a structural risk layered onto a cyclical boom. Treating the entire episode as a temporary chip cycle understates the durable shift in trade geography and electricity demand. Treating every AI-related shipment as permanent demand overstates the durability of current valuations and capital spending.
The cyclical case has three pieces. First, the WTO explicitly links part of the increase in AI-enabling trade to the rebound from weak electronics demand in 2022 and 2023. Second, UNCTAD expects merchandise-trade growth to slow from 4.7% in 2025 to 1.5%-2.5% in 2026 as aggregate demand and investment weaken and the AI surge cools. Third, the IMF has estimated that a moderate correction in AI stock valuations combined with tighter financial conditions could reduce global growth by 0.4 percentage point in 2026 relative to its baseline. A valuation reset can therefore reduce the financing available for the projects that created the trade impulse.
But the structural case survives a slowdown. The WTO says AI-enabling goods have grown from about 13% of global merchandise trade in 2022-23 to nearly 17% by the end of 2025. The IEA expects data-center electricity consumption to double by 2030 even while energy efficiency improves. Governments are also treating advanced computing, energy security and supply-chain resilience as strategic priorities. That changes procurement decisions, subsidy programs and export controls in ways that do not reverse automatically when quarterly demand softens.
The distinction matters for investors and policymakers because the cycle and the structure point in different directions. A cyclical cooling could reduce chip orders, freight volumes and capital-equipment utilization. A structural reconfiguration could keep demand for advanced packaging, grid equipment and regional manufacturing capacity elevated. The winners of the first phase may not be the winners of the second.
IMF data captures this asymmetry across countries. The top four net exporters of AI-related hardware—Taiwan Province of China, Korea, Thailand and Malaysia—recorded an average seasonally adjusted annualized growth surprise of 4.4 percentage points in the July 2026 update, while the rest of the world recorded a negative 0.3 percentage point surprise. That gap is too large to describe as a uniform global technology dividend. It is a concentration effect.
The deeper mechanism is capital allocation. AI revenue growth encourages hyperscalers and governments to build capacity; capacity orders reinforce the revenue outlook of suppliers; stronger supplier earnings attract more capital; and that capital chases the same constrained nodes. The loop can be self-reinforcing on the way up and self-reversing on the way down. The physical assets remain, but the marginal order disappears when expected returns fall.
This is where the second-order risk appears. The obvious risk is that a chip shortage delays AI deployment. The less obvious risk is that a correction in AI demand simultaneously weakens export growth in Asian manufacturing hubs, reduces orders for industrial equipment, pressures energy-infrastructure projects and tightens fiscal conditions in regions that expanded capacity for the boom. A single theme can therefore reverse through several national-income channels at once.
The trade system is not merely exposed to AI demand. It is becoming synchronized with it.
The Counter-Thesis: Concentration Can Be the Route to Resilience
The strongest counter-thesis is that concentration is not necessarily fragility. Specialized clusters can absorb shocks better than dispersed production because they have deeper supplier networks, denser technical labor markets and faster learning curves. The WTO’s trade data supports part of this view: Asian economies contributed 71% of global merchandise-trade growth in 2025, and the region’s semiconductor and electronics ecosystem allowed supply to scale quickly enough to support the AI investment surge despite tariffs and geopolitical uncertainty.
On this argument, today’s concentration is transitional. High profits draw in competitors. New fabs are being built in North America, Europe and elsewhere; cloud providers are diversifying suppliers; custom chips are reducing dependence on a single accelerator architecture; and governments are funding domestic capacity. Over time, the very rents created by bottlenecks should finance redundancy. The market’s answer to scarcity is more capacity.
That counter-thesis deserves more than a dismissal. Physical capacity is expanding, and efficiency improvements can materially reduce the electricity required per AI task. The IEA says power consumption per task is declining rapidly. If that trend persists, the forecast doubling of total data-center electricity use could coexist with a lower energy intensity of computing. Meanwhile, if AI productivity broadens beyond a few hyperscalers, demand could become more diversified across industries and regions.
But the counter-thesis assumes that capacity arrives before the boom cools and that new capacity is substitutable. Neither is guaranteed. A new fabrication plant does not instantly replace the process technology, packaging ecosystem or engineering base of an established cluster. A new data center without transmission capacity is not usable capacity. A domestic supply chain may reduce geopolitical exposure while increasing costs, which can slow adoption and shift demand into a smaller number of projects.
The falsifying signal for the structural-risk thesis is specific: if the share of AI-enabling goods in global merchandise trade falls back toward 13%—its approximate 2022-23 level—while data-center electricity growth slows below 10% annually for two consecutive years, then the concentration would look more like a temporary post-downturn rebound than a durable regime change. A sustained decline in the share of Asia in AI-enabling trade would also show that diversification is arriving faster than concentration.
Until those signals appear, the stronger reading is that diversification is occurring at the margin while the core bottlenecks remain concentrated. New capacity can reduce the probability of a single-point failure without eliminating the network’s dependence on a few strategic nodes.
What the Trade Data Means for Markets and Policy
The immediate market implication is not a simple “AI up” or “AI down” call. It is a change in sensitivity. When AI-enabling goods account for 42% of global trade growth, a change in AI capex expectations can move freight, industrial orders, semiconductor demand and regional export earnings together. That correlation can remain low during the boom because revenue surprises dominate; it can rise during a correction because the same capital-spending assumption sits underneath several asset classes.
Companies closest to bottlenecks may retain pricing power in the medium term, especially in advanced manufacturing, packaging, networking and grid equipment. But their exposure is two-sided. Scarcity supports margins until customers delay projects or redesign systems. Firms further downstream may benefit from cheaper components after a correction, but only if power and financing remain available.
For countries, the asymmetry is sharper. Exporters embedded in AI hardware chains gain from foreign investment and trade volumes, as the IMF’s 4.4-percentage-point growth surprise for the top four hardware exporters illustrates. Countries outside those chains can experience the opposite effect: weaker commodity demand, higher energy costs and less foreign investment as capital concentrates in technology hubs. The AI boom can widen the gap even while it lifts aggregate trade.
For policymakers, the challenge is to distinguish resilience from duplication. Building a second source for a critical chip or transformer can lower systemic risk. Requiring every country to reproduce every step of the value chain can raise costs without creating true substitutes. The most effective policy target is not maximum self-sufficiency; it is credible redundancy at the nodes where a failure would stop the network.
That means tracking physical lead times, grid-connection queues, transformer availability, advanced-packaging capacity and the geographic share of AI-enabling trade—not just chip company revenue. Financial markets often see the demand signal first. Trade resilience depends on the supply signal.
Outlook: Three Paths Through the Next Phase
In the short term, sentiment and liquidity will remain tied to the capex plans of the largest technology companies. The base case is a cooling in the rate of increase rather than an abrupt collapse: AI-enabling trade remains above its pre-boom share, but global merchandise-trade growth moves toward UNCTAD’s 1.5%-2.5% 2026 range as nontechnology trade stays soft. The trigger is a moderation in hyperscaler capital expenditure without a broad cancellation of power and data-center projects.
The upside case is a productivity-led expansion. If energy efficiency lowers power use per task faster than demand rises, and if custom silicon and new manufacturing capacity broaden the supplier base, AI investment could spread beyond the current hubs. The measurable confirmation would be two consecutive years in which AI-enabling trade grows while its regional concentration falls and data-center electricity growth remains below total compute growth.
The downside case is a synchronized capex reversal. A valuation correction, higher financing costs or a major power and shipping disruption could cut orders across chips, servers, industrial equipment and construction. The IMF’s 0.4-percentage-point growth sensitivity provides a scale marker for how financial conditions can turn a technology correction into a macro event. The warning signal would be a year-over-year decline in the value of AI-enabling-goods trade combined with a fall in the capital expenditure of the five largest technology companies.
Over the medium term, the beneficiaries are likely to be firms and regions that solve bottlenecks rather than merely sell into the first wave: grid equipment, power generation, cooling, advanced packaging and specialized manufacturing. The exposed are suppliers whose revenue depends on one spending cohort, and countries whose export growth relies on a narrow set of technology shipments.
Over the long term, the structural question is whether diversification becomes real redundancy. The relevant test is not how many new facilities are announced but how quickly a buyer can switch suppliers without losing performance, and how quickly a data center can secure power without displacing other users. Those are operational measures, not slogans about reshoring.
Data and projections cited were available as of Nov. 12, 2026. The AI boom has made global trade stronger on the surface and more concentrated underneath. The next phase will show whether capital can turn concentration into redundancy before the cycle turns.
The AI boom is not merely expanding world trade; it is tying more of world trade to the same few points of failure.
Explore more exclusive insights at nextfin.ai.
