Component Procurement Supply Chain Under AI Pressure: What Will Be in Short Supply After Memory Chips

Related Vendors

Taiwan Machine Tool & Accessory Builders' Association (TMBA)

SIMTOS (Seoul International Machine Tool Show)

It's not just memory chips that are in short supply: AI infrastructure expansion is affecting the entire supply chain. CPUs and MLCCs are coming under pressure, with one common result: longer delivery times, rising prices and an increasingly reactive procurement market.

The expansion of the AI infrastructure has dominated the semiconductor debate for years, and with it the demand for GPUs and memory chips in particular. However, the bottlenecks created by the AI boom now extend further. CPUs and passive components such as MLCCs are under increasing supply pressure. The consequences for developers and purchasers are similar: longer delivery times, rising prices and a procurement market that is increasingly reactive rather than predictable.

CPUs: from Marginal Topic to Bottleneck Component

For a long time, CPUs were considered the rather unspectacular backbone of PCs and servers. Growth was manageable, and there were no major leaps either in terms of economics or the pressure to innovate. This has changed, as Nikkei Asia reports with reference to AMD CEO Lisa Su.

The AMD boss puts the expected annual growth of the server CPU market at more than 35 percent for the coming years. This growth will be driven primarily by AI inferencing and agent-based AI applications, which are increasingly dependent on CPU computing power. "The overall demand in the CPU market today is significantly higher than any of us predicted a year ago," Su is quoted as saying.

Nvidia is also pushing into this market. With Vera, the GPU market leader is launching its own CPU platform on the market, which is aimed in particular at agent-based AI applications. CFO Colette Kress announced the prospect of CPU sales of almost 20 billion US dollars for the current year during an earnings call and formulated the ambition of leading Nvidia to the top of the CPU market. This claim is unlikely to have escaped the attention of established suppliers Intel and AMD.

Intel is responding to the increasing pressure with a prioritization strategy, as Nikkei Asia reported in May 2026: Older chips based on the Intel 7 process node—including Alder Lake and Raptor Lake - are therefore being reserved primarily for server and industrial applications that generate higher margins. A similar logic can already be seen among memory chip manufacturers: here, too, capacities are increasingly being reallocated to AI-related products, while legacy applications have to deal with scarcer allocations.

This has consequences for PC and notebook manufacturers. According to Nikkei Asia, companies that continue to work with older Intel 7 platforms are reporting that orders are only being partially fulfilled. At the same time, Intel is pushing them more strongly towards newer processors based on the 18A process. "We recently reordered 100 Intel 7 CPUs. We received 30, 10 of which were based on 18A," a PC manufacturer told Nikkei Asia. "We were told that if we don't take the 18A CPUs, they will go to other customers."

For developers, this means that anyone planning on older CPU platforms must expect limited availability. Switching to newer processors can result in redesigns—including adapted peripheral components such as displays, sensors or power supplies.

MLCCs: AI Server As the New Main Customer

The picture is similar for multilayer ceramic capacitors, or MLCCs for short, as reported by eeNews Europe. According to Samsung Electro-Mechanics, AI servers require ten to fifteen times as many MLCCs as standard servers. An Nvidia Grace Blackwell GB200 board has around 6,500 MLCCs; the upcoming Rubin architecture is expected to increase the requirement to around 12,000 components per board. At the same time, technical requirements are shifting: High-capacity MLCCs in miniaturized housings that can stabilize high currents at low voltages of around 0.8 volts are increasingly in demand.

The major manufacturers—including Murata, TDK, Taiyo Yuden, Kyocera AVX, Samsung Electro-Mechanics and Yageo—have restructured their capacities accordingly. Production lines that previously supplied consumer MLCCs are being geared more towards high-end components for AI applications. As a result, standard MLCCs for automotive and industrial applications are becoming scarcer and delivery times for high-capacitance types in housing sizes 1206 and 1210 are already exceeding 20 weeks.

In April 2026, Taiyo Yuden was the first company to raise prices for consumer and automotive MLCCs by six to 13 percent. Yageo and Walsin negotiate price adjustments on a product-by-product basis, without formal announcements. Murata and Samsung Electro-Mechanics have not yet officially positioned themselves, but the market trend is clearly pointing upwards.

Analysts at Astute point out that the bottlenecks are not primarily due to a shortage of raw materials, but to a deliberate capital allocation strategy on the part of manufacturers: "Suppliers are reluctant to build up new capacity for lower-margin components, despite clear signs of recovery." For developers outside the AI server segment, this means that the availability of standard MLCCs can quickly deteriorate in the event of a sudden increase in demand, for example due to an automotive recovery.

New product generations at least address the technical requirements. Murata claims to be the first company in the world to produce MLCCs in 0402 format with 47 µF for AI applications. Kyocera and Taiyo Yuden have announced comparable products in the same housing. Silicon capacitors are also increasingly positioning themselves as an alternative to MLCCs: Analog Devices has announced its intention to acquire the specialist Empower Semiconductor for USD 1.5 billion. Its embedded silicon capacitors are intended to enable higher capacitance densities directly at package level.

Structural Shift, Not A Temporary Outlier

What is emerging is not a classic bottleneck situation like during the 2021 crisis, when an external shock hit the supply chain. The current pressure is caused by a structural shift in demand. AI infrastructure consumes a disproportionately high number of components in specifications that were previously rather niche products. The boom in AI is drawing production capacity away from the consumer and industrial segments.

For developers and purchasers, this results in a change in procurement logic: just-in-time strategies are working increasingly poorly for critical passive components and certain CPU platforms. Long-term purchase agreements and early design decisions—especially when choosing CPU platforms and MLCC specifications—are becoming increasingly important. The question is not so much whether the situation will ease, but when. It is also debatable whether the next surge in demand, for example due to autonomous driving or 6G infrastructure, will occur before normalization. (sb)

AI in the industry

ETMM Dossier – Artificial Intelligence