AI infrastructure boom reshapes electronics supply chain, says in4ma

"The current shortages in printed circuit boards, laminates and specialty raw materials are not a cyclical blip. They are the direct consequence of a capital expenditure wave from a small group of hyperscalers," in4ma reported.

According to the analysts, capital expenditure planned by the four largest US hyperscalers — Amazon, Alphabet, Meta and Microsoft — is expected to reach approximately USD 700–725 billion in 2026, compared with around USD 410 billion a year earlier. The figure exceeds the annual revenue of the global EMS industry, illustrating the scale at which AI infrastructure is reshaping the electronics market.

While the four technology giants dominate spending, they are far from the only companies expanding AI infrastructure. in4ma also points to Oracle, xAI, OpenAI through the Stargate initiative, Anthropic, CoreWeave, Nebius, Lambda and Apple as significant contributors. At the same time, Chinese hyperscalers including Alibaba Cloud, Tencent, ByteDance, Baidu and Huawei Cloud are expanding their own AI infrastructure, placing additional pressure on Asian PCB and substrate suppliers that also serve international customers.

According to in4ma, this concentration of investment means that production capacity increasingly follows the most profitable applications, leaving traditional electronics sectors competing for the remaining manufacturing resources.

The bottleneck extends beyond semiconductors

The report argues that the industry's biggest challenge is no longer access to advanced chips alone. Instead, constraints are spreading across the entire supply chain, affecting advanced laminates, ABF substrates, copper foils, glass fabrics and printed circuit boards required for AI servers.

High-performance AI systems typically require PCBs with more than 20 layers, low-loss M8 and M9 laminates and advanced substrate technologies. These rely on the same manufacturing lines and raw materials used by automotive, industrial and consumer electronics products.

"When four companies commit more than USD 700 billion in a single year, marginal capacity follows the money," the analysts note.

According to in4ma, some PCB manufacturers have already informed existing customers that they will discontinue production of standard boards in favour of AI server projects, where margins are considerably higher.

The pressure is also visible further upstream. Demand for specialised glass fabrics, including 1080, 106 and 104 styles, together with T-glass materials and very-low-profile copper foils, continues to increase as manufacturers seek higher-performance materials capable of supporting AI server designs.

Production issues at resin suppliers have added further strain to the market. in4ma notes that disruptions affecting SABIC, which accounts for roughly 70% of global high-purity PPE resin supply, have contributed to higher manufacturing costs and helped push PCB prices up by as much as 40% during the first half of 2026.

Massive projects illustrate the scale

To illustrate the scale of current investment, the report highlights plans for a proposed 10-gigawatt AI campus in Pike County, Ohio. The project, developed by SB Energy, a SoftBank subsidiary, is estimated to require investments exceeding USD 500 billion once fully completed.

According to the report, the first 800-megawatt phase is targeted for 2028 under a long-term lease agreement with OpenAI. Nvidia has reportedly been involved in discussions regarding financing arrangements for both construction and future chip purchases, although no final agreements have been announced.

The analysts argue that projects of this magnitude demonstrate that the industry's limiting factors are shifting. Rather than semiconductor availability alone, future expansion will increasingly depend on guaranteed power generation, direct-to-chip liquid cooling technologies and access to financing on an unprecedented scale.

Taiwan dominates AI server manufacturing

While hyperscalers are driving demand, the companies manufacturing AI server infrastructure are concentrated elsewhere.

According to in4ma and EMSNOW, the market is overwhelmingly dominated by Taiwanese manufacturers, including Foxconn, Wistron, Quanta Computer, Wiwynn, Inventec and Accton Technology. Among Western manufacturers, Celestica is identified as one of the few companies outperforming the broader market.

However, the report argues that spectacular revenue growth does not automatically translate into stronger profitability.

Foxconn, for example, has benefited from surging AI server demand, yet the company continues to operate with a net margin of just 2.3%. According to in4ma, inventories increased by 31% year-on-year while total debt rose by 34%, illustrating the working-capital intensity associated with AI infrastructure projects.

Wistron recorded revenue growth of more than 108%, with approximately 62% of its sales now generated by data centre products. Despite this expansion, the company's net margin remains only 1.25%, which in4ma describes as the lowest among its closest competitors.

Inventec presents a different picture. Although revenue increased by a comparatively modest 6.9%, pre-tax profit rose by more than 30%, suggesting a strategy focused on profitability rather than market share.

Longer lead times and changing payment terms

The tightening supply environment is also changing commercial practices throughout the electronics industry.

According to in4ma, suppliers are increasingly requesting partial or full prepayments before accepting orders. The analysts believe these measures serve two purposes: discouraging speculative purchasing while simultaneously improving suppliers' liquidity during a period of prolonged capacity constraints.

Whether stricter payment terms will also reduce the bullwhip effect that typically accompanies allocation periods remains unclear, the report notes.

Looking ahead, in4ma expects AI-driven demand to remain strong through the end of the decade. Supply constraints for advanced PCB technologies are likely to persist until around 2028, supporting elevated pricing and stronger margins for manufacturers serving the AI market before competitive conditions gradually normalise after 2030.

Readers interested in the latest market outlook from in4ma will have an opportunity to hear directly from the company's analysts during upcoming Evertiq Expo events. Mareike Haass will present the latest industry analysis at Evertiq Expo Gothenburg on 17 September, while Dieter G. Weiss and Anna Żurek-Kasiewicz will speak at Evertiq Expo Warsaw on 22 October.