A projected surge in artificial intelligence data center construction could drive a major new wave of semiconductor manufacturing equipment spending, with one new estimate suggesting that AI-related wafer fabrication equipment demand could reach hundreds of billions of dollars by the end of the decade if power capacity expands at the pace now being discussed across the technology industry.
A new Bernstein analysis links future AI computing power demand to semiconductor production capacity, estimating that if AI data center capacity grows by 50 gigawatts annually by 2030, related wafer fabrication equipment spending could total $736 billion between 2027 and 2029. Under that base case, annual spending tied to AI semiconductor demand would rise to about $291 billion in 2029 alone.
In a more aggressive scenario, where annual AI computing capacity additions reach 100 gigawatts, the projected 2029 wafer fabrication equipment figure could climb to $542 billion. That would represent a much larger expansion cycle for the semiconductor equipment sector, with potential effects across memory, logic chips, lithography tools, etching systems, deposition equipment and process control technologies.
Wafer fabrication equipment, commonly referred to as WFE, covers the specialized machinery used to manufacture semiconductors on silicon wafers. It is a key spending category for chipmakers and a major revenue source for suppliers such as Applied Materials, Lam Research, KLA, ASML and Tokyo Electron. The category includes tools used across the chip production process, from patterning and deposition to etching, cleaning, inspection and metrology.
The analysis attempts to translate one of the most visible bottlenecks in the AI boom — electricity for data centers — into expected demand for semiconductor production tools. The core idea is straightforward: more AI computing capacity requires more chips, more memory and more advanced packaging, and that eventually requires more wafer manufacturing capacity.
But the scale of the estimate also highlights how dependent the AI buildout is on physical infrastructure. Data centers need land, grid connections, substations, cooling systems, transformers, chips, networking hardware and long-term power contracts. Semiconductor manufacturers, in turn, need multi-year capital programs, qualified suppliers and reliable demand before committing to the production expansions that would support such a boom.
AI power demand becomes a semiconductor spending signal
Bernstein estimates that each additional 1 gigawatt of AI computing capacity would require roughly 46,000 to 50,000 wafers per month in new semiconductor production capacity. That conversion is central to the firm’s framework, because it links data center power growth directly to chip manufacturing needs.
The estimate also shows that AI semiconductor demand is not limited to advanced graphics processors. According to the analysis, DRAM would account for about 53% of the required incremental wafer capacity, NAND would account for about 20%, high-bandwidth memory would represent about 16%, and advanced logic chips would make up about 11%.
That split suggests memory could be one of the largest beneficiaries of AI infrastructure growth. Advanced GPUs and accelerators receive most of the attention because they perform the core AI training and inference workloads. However, those systems also require substantial memory capacity, including high-bandwidth memory used near the processor and conventional DRAM and NAND used across servers and storage systems.
The result is a broader semiconductor demand profile than many market discussions imply. AI data centers require accelerators, but they also require networking chips, storage, CPUs, power management devices and large volumes of memory. If data center capacity growth remains strong, the spending cycle could extend across multiple segments of the chip supply chain rather than concentrating only in the most advanced logic nodes.
Under the 50 gigawatt base scenario, AI-related wafer fabrication equipment spending would rise gradually from about $200 billion in 2027 to $245 billion in 2028 before reaching $291 billion in 2029. When combined with baseline non-AI WFE spending of roughly $120 billion per year, total spending would reach the estimated three-year sum of $736 billion.
That would mark a significant expansion of the semiconductor capital equipment cycle. It would also create a test for suppliers that already face long production lead times and highly technical delivery requirements.
Equipment suppliers could see sharp earnings sensitivity
The projections imply meaningful earnings sensitivity for major semiconductor equipment companies if AI-related capacity demand translates into actual orders.
In the Bernstein framework, Applied Materials’ 2029 earnings per share could rise about 59.5% above current expectations, with a forward price-to-earnings multiple near 14.9. Lam Research could see earnings per share rise by about 54.4%, with a multiple near 18.4. KLA’s potential earnings increase is estimated at 37.1%, implying about 21.2 times forward earnings.
These estimates reflect the operating leverage embedded in semiconductor equipment companies. When large chipmakers increase capital spending, equipment suppliers can see revenue and margins move quickly, especially if factories are operating at high utilization and customers compete for delivery slots.
Still, the earnings impact would depend on the timing and composition of orders. Different equipment suppliers have different exposure across deposition, etching, inspection, process control, lithography and services. Memory capacity additions may benefit one group of suppliers more heavily, while advanced logic expansions may support another. High-bandwidth memory growth may also require additional capacity in specialized process steps and advanced packaging.
The largest potential gains would come only if semiconductor manufacturers decide that AI demand is durable enough to justify new fabrication capacity. Chipmakers typically avoid overbuilding when demand visibility is uncertain, because new fabs require very large capital commitments and can take years to construct, equip and qualify.
Data center pipeline grows rapidly in the United States
The United States remains one of the main centers of AI data center expansion. As of June 2026, project pipeline capacity had reached 338 gigawatts, an increase of 217 gigawatts over the previous 12 months, according to the figures cited in the analysis. That pipeline is significantly larger than the capacity of facilities already in operation.
The scale of the pipeline shows how aggressively cloud operators, AI developers, infrastructure companies and power-intensive technology firms are planning for future computing demand. It also reflects growing competition for suitable sites near transmission infrastructure, water resources and reliable electricity supply.
However, planned capacity does not automatically become operating capacity. Many projects must still secure grid interconnections, power purchase agreements, permits, construction financing, cooling systems, transformers and server equipment. Local approval processes can also slow development, especially in regions where residents and officials are concerned about water use, land use, noise, emissions or pressure on electricity prices.
A large pipeline can therefore signal future demand without guaranteeing that all planned capacity will be built. For the semiconductor industry, this distinction is critical. Wafer fabrication equipment spending depends not only on announcements from data center developers, but also on confidence that those sites will become operational and consume chips at scale.
If data center projects are delayed by power constraints or regulatory approvals, semiconductor manufacturers may postpone equipment orders. If AI workloads become more efficient and require less hardware for the same output, the required wafer capacity could also be lower than projected.
Power supply is becoming the central constraint
The semiconductor demand outlook is increasingly tied to the availability of electricity. AI data centers are among the most power-intensive facilities in the modern technology economy, and their expansion is creating competition across utilities, cloud providers, manufacturers, digital asset miners and industrial users.
The International Energy Agency expects global data center electricity consumption to more than double from about 415 terawatt-hours in 2024 to 945 terawatt-hours by the end of 2030. That increase would place data centers among the fastest-growing sources of electricity demand in many markets.
Power availability is already influencing where new facilities are built. Developers are prioritizing sites with access to large grid connections, stable power prices and the ability to scale capacity over time. In some regions, data center operators are also exploring direct arrangements with power producers, including renewable energy developers, natural gas plants and nuclear power operators.
The search for power is also affecting companies that previously used large amounts of electricity for digital asset mining. Some mining operators have begun converting facilities into high-performance computing centers because AI workloads can generate far higher revenue per unit of electricity than traditional hash-based operations.
Hosting AI workloads can generate roughly $25 in steady revenue per kilowatt-hour, according to industry estimates cited in the source material, while standard hashing operations may produce closer to $1 per kilowatt-hour under current conditions. That gap is changing strategic decisions across power-heavy computing businesses.
Miners shift toward high-performance computing
Publicly traded digital asset mining companies secured more than $65 billion in cloud computing contracts during 2025, according to the figures provided. These firms often already control assets that hyperscale data center operators need, including land, electrical infrastructure, cooling systems, grid approvals and relationships with utilities.
That existing infrastructure gives some mining companies a path to reposition themselves as providers of high-performance computing capacity. Instead of using electricity mainly to secure proof-of-work blockchain networks, they can lease power and facilities to AI and cloud customers under longer-term commercial contracts.
The shift is not simple. AI data centers require different hardware, higher reliability standards, more advanced networking, stronger redundancy and more demanding customer service agreements than many traditional mining facilities. Facilities may need substantial upgrades before they can host AI workloads at scale.
Even so, the economic incentive is powerful. If AI customers are willing to sign long-term contracts at higher margins, mining companies with attractive power assets may have reason to redirect future capacity. Some operators may maintain a mixed model, using part of their energy allocation for digital asset mining and part for machine learning or cloud computing workloads.
For traders in digital asset markets, the trend creates a new set of physical constraints to monitor. Hashprice, or daily revenue per unit of computing power, has recently been near cyclical lows around $34 per petahash. If mining economics remain weak while AI hosting economics remain strong, more operators could reduce their exposure to pure mining.
Implications for digital asset networks
A large-scale shift away from mining and toward AI computing may affect proof-of-work networks over time. These networks depend on miners committing physical equipment and electricity to secure transactions and produce new blocks. If industrial operators allocate more future capacity to machine learning workloads, the balance of dedicated computing resources may change.
The effect would depend on the scale and speed of the transition. If only marginal or less efficient mining capacity exits, stronger operators may remain and network security may adjust through difficulty changes. If a larger share of industrial operators redirects power and sites toward AI, proof-of-work networks could face a more meaningful reduction in physical backing.
Some estimates suggest that operators could dedicate up to 70% of future capacity to machine learning workloads if AI contracts continue to offer stronger economics. That would reshape business models for companies that once depended primarily on token issuance and transaction fees.
At the same time, shortages in traditional cloud capacity could increase interest in decentralized computing networks. Utility tokens linked to distributed computing may benefit when centralized data centers face power bottlenecks, long construction timelines or capacity shortages. These networks attempt to aggregate unused or underused computing resources and make them available for public workloads, though reliability, verification and enterprise-grade service standards remain important hurdles.
Low-energy blockchain ecosystems may also draw more attention as electricity becomes a more strategic resource. Proof-of-stake and other less energy-intensive consensus models are not directly exposed to the same power competition as proof-of-work mining. That difference may become more relevant if AI data centers continue bidding aggressively for electricity and infrastructure.
Supply chain limits remain a major risk
Even if data center demand continues to grow, the semiconductor equipment supply chain may struggle to expand quickly enough to match the most aggressive scenarios.
Advanced semiconductor manufacturing tools are among the most complex industrial products in the world. Lithography systems, deposition tools, etching equipment, inspection machines and metrology systems involve specialized components, precision engineering and extensive customer qualification. Production capacity cannot be expanded instantly.
Lead times can be long, especially for the most advanced tools. Supplier networks are limited, and some parts require highly specialized manufacturing. Equipment must also be installed, calibrated and qualified inside fabs before it can support volume chip production.
These constraints mean that higher demand may not translate smoothly into completed capacity. Semiconductor manufacturers could place more orders, but delivery schedules, installation bottlenecks and technical validation could delay the actual output of chips.
The projections also rely on assumptions about GPU architecture, chip power density, fabrication timelines, capital intensity, memory requirements and manufacturing yields through 2029. Any major change in chip design could alter wafer demand. More efficient AI accelerators could reduce the number of chips needed per gigawatt. Better yields could reduce the amount of capacity required. Conversely, larger models, heavier inference workloads or more memory-intensive architectures could increase demand beyond current estimates.
Budget constraints could also slow the buildout. AI infrastructure is expensive, and returns depend on how quickly companies can monetize AI services. If revenue growth fails to justify the pace of spending, data center operators may slow expansion plans, reducing the need for new semiconductor capacity.
The next phase depends on execution
The analysis presents a clear quantitative link between AI power expansion and semiconductor equipment demand: each new gigawatt of AI computing capacity could correspond to roughly 50,000 monthly wafer starts and nearly $8 billion in additional WFE spending.
That conversion, however, remains conditional. It depends on data center projects moving from pipeline to operation, power supplies being secured, chipmakers committing to new capacity and equipment suppliers delivering tools on schedule.
For semiconductor equipment companies, the AI buildout represents one of the largest potential growth drivers of the decade. For data center operators, the biggest challenge is no longer just chip access, but power access. For digital asset mining companies, the AI boom is turning electricity rights and grid connections into assets that may be worth more than the mining activity they were originally built to support.
The central question is whether the physical infrastructure can keep pace with the financial ambition behind AI. If it can, wafer fabrication equipment spending could enter a powerful new cycle by 2027. If power, permitting, financing or supply chain limits slow construction, the spending surge may arrive later, be smaller, or concentrate among only the best-positioned companies.
Explore how on-chain assets reshape capital markets—read our latest insights on tokenized private credit migration and future investment flows.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
