Semiconductor equipment stocks have declined by 30% from their highs, yet are still up 80% year-to-date.

Bernstein provides an answer through a bottom-up projection: every additional 1GW of AI data center compute capacity requires $8 billion in equipment spending. At a projected annual addition of 50GW, cumulative equipment outlays from 2027 to 2029 will exceed $700 billion, driving annual WFE (Wafer Fabrication Equipment) spending toward $30 billion.

This figure carries substantial weight. Current market pricing for equipment stocks implies only $120–150 billion in annual WFE spending—half the projected level. If AI buildout momentum remains unchanged, today’s equipment stocks are not expensive; they may actually be undervalued.

Bernstein maintains an overall positive view on the equipment sector, with a particular overweight on Applied Materials (AMAT). Over half of incremental wafer demand comes from DRAM and HBM, where Applied Materials has the largest exposure in the space. Applied Materials, Lam Research, KLA, ASML, Tokyo Electron, Kokusai, Lasertec all outperform the benchmark, while Screen is flat.

$8 Billion in Equipment Spend Behind Every 1GW of Compute

Bernstein breaks it down meticulously. A single Vera Rubin rack consumes 65 wafers, covering logic, HBM, DRAM, and NAND. Outside the rack, servers, CPUs, companion memory, and storage represent additional demand.

Aggregating all components, each additional 1GW of annualized compute capacity requires 46,000 wafers per month of new wafer capacity. Half is driven by DRAM and HBM, 20% by NAND, and 10% by advanced logic. When converted into equipment investment, this equates precisely to $8 billion.

This estimate does not include replacement of legacy equipment. A wave of aging compute infrastructure will require upgrades between 2027 and 2029, meaning actual demand will likely exceed these projections.

50GW Scenario: Annual WFE Spending Soars to $30 Billion

If annual compute additions reach 50GW by 2030—adding 50GW above the 2026 baseline—cumulative WFE spending over three years would surpass $70 billion. Adding non-AI baseline demand of approximately $12 billion annually, annualized WFE spending would jump from $20 billion to nearly $30 billion.

Even faster growth scenarios—75GW or 100GW—would make $30 billion just a starting point.

The current market expectations fall far short. Consensus-implied annual WFE spending is only $120–150 billion, less than half of Bernstein’s base case. If this scenario materializes, significant valuation upside remains for equipment stocks.

Doing the Math: Are Equipment Stocks Expensive Now?

Bernstein runs three scenarios, directly comparing current valuations.

50GW scenario: Applied Materials’ 2028 EPS rises from the consensus $18.70 to $24.30 (+30%), and to $30.60 in 2029 (+60%). Corresponding P/E multiples drop from 23x to 15x in 2028, and further to 11x in 2029.

75GW scenario: 2028 EPS reaches $34.20 (+over 80%), and $46.70 in 2029 (+over 100%). P/E drops to 11x in 2028 and 8x in 2029.

100GW scenario: 2028 EPS hits $44.70 (+over 100%), and $62.40 in 2029 (+over 200%). P/E falls to 8x in 2028 and 6x in 2029.

Lam Research and KLA exhibit similar leverage. The conclusion is clear: as long as AI construction continues unabated, current equipment stocks are far cheaper than the market assumes.

Why Applied Materials?

Of incremental wafer demand, DRAM and HBM account for 55%. Applied Materials holds the highest exposure across both DRAM and HBM equipment segments—this is the core reason behind Bernstein’s strongest conviction in the stock.

Ratings on other names remain unchanged. Lam Research, KLA, ASML, Tokyo Electron, Kokusai, Lasertec all outperform the benchmark; Screen maintains market performance.

Tide-Forward Perspective

The market recognizes that AI demands data centers, data centers need chips, and chips require equipment. But few have rigorously quantified exactly how much equipment is needed, what revenue it generates, and what valuation it supports.

Bernstein has done the math: 1GW corresponds to $8 billion in equipment, 50GW translates to $30 billion in annual WFE, implying a 15x P/E for equipment stocks. Yet current market pricing reflects only $150 billion in implied annual WFE spending and a 20x+ P/E.

This gap represents a one-time expectation gap.

Risks remain tangible: could construction pace slow? Can equipment capacity keep up? Will customers cut orders during cyclical downturns? Each is a variable. Equipment stock volatility has always been greater than semiconductor industry-wide swings.

But one thing is clear: if the AI compute expansion narrative persists, equipment stock valuations are far from peak. Whether today’s pullback represents risk or opportunity ultimately hinges on investor belief in the speed of AI infrastructure deployment.

Source: TechFlow Column