Nomura refutes the “semiconductor peak theory”: an “epic shortage” is coming, with price increases and earnings revisions still the biggest catalysts.

```

Although the AI semiconductor sector has recently seen a pullback, the industry investment cycle has not reached its peak.

According to Chase Wind Trading Desk, Nomura's latest report points out that the capital expenditure of cloud service giants will continue until 2027. An unprecedented supply gap for components is approaching, and sustained price hikes and upward revisions in earnings forecasts will become the biggest catalysts driving the market higher.

The market will encounter the most severe supply chain mismatch in history in the second half of 2026. Although TSMC is aggressively expanding its wafer-level packaging capacity, the real supply bottleneck will shift to wafer-level substrates (WoS) and smaller components such as PCBs and CCLs. This structural shortage will directly intensify short-term price fluctuations, but it also confirms the long-term sustainability of this cycle.

Driven by strong demand and cost inflation, Nomura has significantly raised its expectations for the global server market, forecasting that AI server revenues will grow by 78% and 76% in 2026 and 2027 respectively. Based on this, the institution reiterates its "buy" rating for nine Asian AI technology companies, including TSMC, ASE, and MediaTek, raising target prices across the board and recommending investors to buy on dips when the market is weak.

As tech giants such as Nvidia and Google intensify their competition over advanced packaging capacity, the competitive landscape for AI chips faces restructuring. Meanwhile, the rise of Agentic AI is unexpectedly driving demand for traditional server CPUs, opening up new market opportunities for the semiconductor testing and packaging industry.

“Epic” supply chain mismatch intensifies, price hike expectations clear

The semiconductor hardware supply chain is facing an unprecedented supply-demand imbalance. Nomura points out that many component suppliers have severely underestimated the order-upside potential brought by AI in their capacity expansion plans. In addition to the well-known shortages of advanced processes, advanced packaging, memory, and CPUs, PCBs, CCLs, IC substrates, high-end capacitors, power management ICs, and optical components are all currently in shortage.

As Nvidia's Rubin architecture and AWS's Trainium 3 begin mass production in the second half of 2026, the supply-demand situation will worsen further. Since newly built greenfield capacity usually takes two years, this means that supply will remain constrained until 2027. This widespread component shortage will not only limit the growth of AI servers but also severely squeeze the supply chains of non-AI sectors such as consumer electronics and automobiles. Against the backdrop of intensifying shortages, the price-uptrend across the supply chain is expected to continue or even expand further.

Data center construction accelerates, giant capital expenditures support long-term demand

Global tracking data of newly built data centers provides leading demand indicators for the hardware supply chain. Nomura’s proprietary data shows that the number of global data center projects has increased from the previously tracked 240 to 280, with the number of gigawatt (GW)-level projects rising to about 50. It is expected that newly deployed computing power will reach 32 GW in 2027, and there is currently visibility for 23 GW in 2028.

These demands are mainly driven by the capital expenditures of hyperscale cloud service providers such as Microsoft, Google, Meta, AWS, and emerging neoclouds such as CoreWeave. Additionally, according to Bloomberg News, the Chinese government has drafted an unprecedented national AI computing network plan, planning to invest $295 billion over the next five years to achieve nationwide distributed data center interconnection by 2028. These ongoing projects provide solid support for hardware demand in the next two to three years.

Giants vying for advanced packaging capacity, market landscape faces restructuring

In the advanced packaging field, TSMC’s approach has become aggressive. TSMC Chairman C.C. Wei stated in the earnings conference call that the company is striving to meet all demand and will not relinquish any business opportunity. Nomura expects TSMC’s CoWoS capacity target to reach 2 million units in 2027. However, due to bottlenecks in WoS and other small components, actual CoWoS output in 2027 may only be 1.8 million units.

Against the backdrop of limited capacity, competition among AI chip giants for resources will be extremely fierce. Nomura predicts Nvidia will still occupy about 55% of TSMC’s CoWoS capacity in 2027, while Google’s TPU share will jump from 23% in 2026 to 27%, making it the fastest-growing AI logic chip. This "when elephants fight" situation will severely squeeze capacity space for other manufacturers such as AMD and AWS.

Meanwhile, Intel's EMIB-T technology is becoming TSMC's biggest potential threat in advanced packaging. Google’s next-generation TPU v9 project is collaborating with MediaTek and plans to use Intel’s EMIB-T packaging. In order to maintain its leading position, TSMC is accelerating the mass production of SoIC and CoPoS technologies to meet the packaging demands of even larger chips in the future.

The rise of agentic AI, server CPUs see unexpected boom

Besides AI accelerators, the boom in agentic AI is driving rapid growth in server CPU demand.

Nvidia CEO Jensen Huang pointed out that the future AI economy is token-based, and the Vera CPU is designed specifically for agentic AI as a low-latency, high-bandwidth processor.

AMD CEO Lisa Su also stated that agentic AI needs CPUs for orchestration and data movement, predicting that the total potential market (TAM) for server CPUs will exceed $120 billion by 2030. Arm CEO Rene Haas likewise believes that the data center CPU market will see a $100 billion-scale opportunity.

The surge in CPU demand directly benefits outsourced semiconductor assembly and test (OSAT) firms. Additionally, the boom in CPU servers has considerably increased the demand for baseboard management controllers (BMC).

Based on strong expectations for AI and general server markets, Nomura has comprehensively raised profit projections and target prices for the Asian semiconductor hardware supply chain. In the foundry and chip design segments, TSMC, as the core enabler of AI chips, has had its profit forecast for 2026–2028 raised; MediaTek will become the biggest beneficiary of Google TPU share growth.

~~~~~~~~~~~~~~~~~~~~~~~~

The above highlights are from Chase Wind Trading Desk.

For more detailed interpretation, including real-time insights and frontline research, please join the [Chase Wind Trading Desk ▪ Annual Membership]

Risk Disclaimer and Exemption ClauseThe market has risks; investment requires caution. This article does not constitute personal investment advice and does not take into account individual users' particular investment goals, financial status, or needs. Users should consider whether any opinions, viewpoints, or conclusions in this article are suitable for their specific circumstances. Invest at your own risk. ```