SABIC Strike Hits the AI Supply Chain: PCB Resin Costs More, Lead Times Grow

A military strike on a petrochemical complex has become a problem not only for the energy sector but also for the AI hardware market. After the attack on the SABIC complex in Jubail in early April, production of the resin used to make printed circuit board laminates stopped. It is one of those materials that rarely make headlines, but without it you cannot build a server motherboard, an accelerator or networking gear for data centers.

The materials in question are epoxy and related resins used in PCB laminates, the base layer of a printed circuit board. Copper traces connecting chips, memory, power components and controllers are laid on top of it. For the AI market this is critical: modern clusters need huge numbers of boards, not only inside GPU modules but also in switches, power systems, network cards and storage racks. If one chemical input drops out, it can stall not just one line but several links in the chain at once.

The problem is compounded by the fact that AI servers need far from the simplest boards. They must withstand high component density, heavy thermal loads and complex signal routing between GPUs, memory, network interfaces and power circuitry. For such products the requirements for laminates and resins are higher than for mass-market consumer electronics: mechanical stability, predictable behavior under heat and high-frequency connection quality all matter. That shortens the list of interchangeable materials and makes the market sensitive even to local disruptions on the chemical side.

The first effects are already visible in the numbers. According to Goldman Sachs analysts, prices for the resin in question rose 40% in April alone. A South Korean supplier that works with the supply chains of Samsung and AMD reported that waiting times for epoxy resin have stretched from three to fifteen weeks. For electronics makers this is a sharp deterioration in planning: component purchases for servers and accelerators are usually synchronized on a rigid schedule, and even a few extra weeks on a single material can push back the output of finished systems by a quarter.

It is especially painful for AI infrastructure, where demand is already tight. Accelerator makers, OEM server assemblers and data center operators already live with long supply cycles, high factory utilization and competition for priority at contract manufacturers. Against that backdrop, pricier PCB materials hit more than the cost of a single board. The effect cascades into the price of the final system, the availability of spare parts and the timing of capacity expansion. Even if the GPUs and HBM memory themselves are in stock, without boards and substrates they cannot quickly be turned into a finished server.

Large buyers will likely try to buy up available volumes in advance and build safety stock, which may push prices up even further for smaller manufacturers. In this situation, companies with long-term contracts and supplier priority win, while assemblers that work on thinner margins and depend on the spot market lose. For startups and second-tier cloud providers, the risk is getting infrastructure later and at higher cost than planned.

SABIC is a major player in petrochemicals, and a failure at such a supplier quickly becomes a global problem because alternatives are unevenly distributed. Switching to another resin producer is not always possible: in electronics, certification, chemical consistency, flame resistance, electrical properties and compatibility with specific manufacturing processes all matter. Changing a material often requires repeat testing and sometimes retuning production. So the market cannot instantly make up the lost volume, even if demand for end electronics remains high.

The takeaway is unpleasant but important: the AI race depends not only on chips and clouds but on the most basic industrial materials. One strike on a petrochemical facility has already raised prices and stretched lead times in a segment that Samsung, AMD and the entire server ecosystem need. If the disruption drags on, it will add to inflation in AI hardware and remind the market that the main shortage may arise not at the level of chips at all, but at the level of resin for printed circuit boards.