Bernstein Analysis: 50GW Power Revaluation of Equipment Stocks, Is the AI Equipment Super Cycle Coming?

TL;DR
· Bernstein estimates that under the 50GW scenario, the cumulative WFE spending from 2027 to 2029 will be approximately $736 billion.
· Each additional 1GW/year of AI computing power requires about 46K-50K WSPM wafer capacity, with DRAM and HBM making up the majority.
· Applied Materials, Lam Research, and KLA have higher profit elasticity, but pipeline capacity does not equal real orders.

Bernstein's latest report translates the expansion of AI data centers into semiconductor manufacturing equipment orders: If AI data centers add 50GW of computing power annually by 2030, global related WFE spending from 2027 to 2029 could reach approximately $736 billion, with about $291 billion in 2029 alone.

WFE refers to wafer fab equipment spending, which is a key demand source for equipment companies like Applied Materials (AMAT), Lam Research (LRCX), KLA (KLAC), ASML, and Tokyo Electron. For investors, the most direct question from this estimate is: How much additional orders and profit elasticity can the equipment manufacturers gain from the continued expansion of AI computing power?

At the time of the report's release, semiconductor equipment stocks had already experienced a significant rise, followed by a noticeable pullback from their highs. Meanwhile, the pipeline for data center construction in the U.S. continues to expand. Public excerpts show that as of June 2026, the project pipeline capacity has risen to 338GW, an increase of 217GW over the past 12 months, far exceeding the current operational scale.

Changes in active capacity and project pipeline for U.S. data centers, with pipeline capacity increasing from 121GW to 338GW.

Every additional 1GW of computing power requires about 50,000 wafers per month

The core calculation of this report is that each additional 1GW/year of AI computing power requires approximately 46K-50K WSPM of additional wafer capacity. WSPM refers to the number of wafers processed per month, a common metric for measuring wafer fab capacity.

The capacity of data centers does not directly translate into equipment orders. Only when AI servers require more GPUs, HBM, DRAM, NAND, and advanced logic chips do wafer fabs need to expand production, and equipment companies will see increased WFE spending.

Among the approximately 46K WSPM/GW of additional demand, DRAM has the highest proportion at about 53%; NAND accounts for about 20%; HBM about 16%; and advanced logic about 11%. This means that the expansion of AI data centers not only drives advanced GPU processes but also increases demand for storage capacity, especially for DRAM and HBM.

This is also why Applied Materials is the most closely watched in this calculation. The incremental wafer demand mainly comes from DRAM, HBM, and NAND, and Applied Materials has a higher exposure in storage equipment, deposition, etching, and other segments, leading to more direct profit elasticity.

Each additional 1GW of computing power requires about 46K WSPM, with DRAM at 53%, NAND at 20%, HBM at 16%, and Logic at 11%.

WFE in 2029 could approach $291 billion

Under the baseline scenario, AI data centers are expected to achieve an additional 50GW of computing power annually by 2030 compared to the baseline of 2026. To support this goal, related WFE needs to be gradually in place from 2027 to 2029.

Scenario calculations show that AI-driven WFE spending alone will accumulate to about $376 billion from 2027 to 2029; if we add about $120 billion of non-AI baseline spending each year, the total WFE spending over three years will be approximately $736 billion. The annual path is estimated to be about $200 billion in 2027, $245 billion in 2028, and $291 billion in 2029.

These figures are higher than the spending assumptions in the current more conservative models. If the 50GW scenario materializes, WFE scale in 2029 will approach $300 billion; under a higher GW scenario, equipment spending has further upside potential.

The report also provides a more aggressive scenario. Under the 75GW scenario, the profit revision for equipment companies could exceed 100%; under the 100GW scenario, the potential for WFE spending in 2029 could further expand, with some companies' valuation multiples potentially dropping below 10 times.

However, these are still model estimates, not confirmed orders. They depend on whether the data center construction pipeline can convert into actual operations, whether AI server shipments can keep pace, whether wafer fabs are willing to expand production in advance, and whether the equipment supply chain has sufficient delivery capacity.

Under the 50GW scenario, cumulative WFE spending from 2027 to 2029 will be approximately $736 billion, with about $291 billion in 2029; under the 100GW scenario, it will be about $542 billion in 2029.

Applied Materials has the highest elasticity, while Lam and KLA also benefit

The impact on stocks is mainly concentrated in Applied Materials, Lam Research, and KLA.

Under the 50GW scenario, the EPS of these three companies could rise by about 37%-60% compared to the current Wall Street consensus by 2029. Corresponding forward P/E ratios could drop to the range of 15-20 times, while current equipment stocks are generally still trading at higher multiples.

Among them, Applied Materials has the highest elasticity. Under the 50GW scenario, its EPS in 2029 could rise nearly 60% compared to consensus, corresponding to a forward P/E ratio of about 14.9 times; Lam Research's EPS could rise about 54%, corresponding to about 18.4 times; KLA's EPS could rise about 37%, corresponding to about 21.2 times.

The reason lies in the composition of wafer demand. The new capacity brought by AI expansion is mainly concentrated in DRAM, HBM, and NAND, rather than solely in advanced logic. Applied Materials has a broader coverage in storage-related equipment, making it easier to capture incremental demand compared to companies that only benefit from certain segments.

According to the report, Bernstein maintains an outperform rating for several equipment stocks, including Applied Materials, Lam Research, KLA, ASML, and Tokyo Electron, with Applied Materials still being the top choice. Screen is rated neutral.

Under the 50GW scenario, AMAT/LRCX/KLAC's EPS in 2029 could rise by 59.5%/54.4%/37.1% compared to consensus, with P/E ratios dropping to 14.9x/18.4x/21.2x.

Pipeline capacity must also overcome power, financing, and delivery hurdles

The most easily misinterpreted aspect of this calculation is treating the data center pipeline capacity as future equipment orders.

The significant expansion of the U.S. data center pipeline over the past year indicates that the willingness to invest in AI infrastructure remains strong. However, there are multiple hurdles between the pipeline and actual operations: power access, land approvals, financing costs, GPU supply, customer demand, network and cooling infrastructure, all of which will affect the final implementation speed.

There are also constraints on the equipment side. If WFE scale is to reach the $300 billion level within a few years, equipment companies, parts suppliers, and wafer fabs need to expand production simultaneously. The semiconductor equipment industry is not one that can ramp up production indefinitely; advanced equipment, key components, installation, and customer certification will all extend delivery cycles.

The model assumptions themselves also have boundaries. The relevant calculations are based on specific GPU architectures, power consumption, chip area, and capital intensity assumptions, and assume that WFE needs to be in place by the end of 2029 to support the additional computing power in 2030. Changes in architecture, reductions in energy consumption per unit of computing power, improvements in chip yield, and adjustments in capital intensity could all alter the final equipment demand.

Another directional deviation also exists. If the demand for replacing outdated computing power before 2030 is not fully accounted for, equipment demand may still have upside potential; however, if the monetization speed of AI applications is slower than expected, or if large cloud providers slow down capital expenditures, the 50GW, 75GW, or even 100GW scenarios may also be overly optimistic.

This report does not provide a definitive order list but rather a clearer calculation: For every additional 1GW of AI data centers, it may correspond to about 50,000 wafers per month of capacity and about $8 billion in incremental WFE demand. Equipment stocks have already reflected part of the AI expectations in advance, and the divergence lies in whether data center construction can deliver enough to support an annual WFE level of $300 billion.