Maintain core exposure to NVIDIA (NVDA) alongside cloud hyperscalers Microsoft (MSFT), Alphabet (GOOGL), Meta (META), and Amazon (AMZN), with enterprise GPU orders booked out through 2028 and Big Tech infrastructure spending scaling to $1 trillion next year.

Expand semiconductor investments into high-bandwidth memory (HBM) and custom ASIC designers to capture high-margin growth amid a three-year memory supply backlog.

Buy power utilities, energy infrastructure, and electrical equipment makers supplying turbines and transformers to exploit a projected 19-gigawatt power shortfall by 2028.

Target advanced liquid cooling providers and physical data center developers required to support the massive heat and structural demands of high-density AI clusters.

Take strategic positions in copper to profit from acute commodity shortages driven by global high-voltage grid upgrades and electrical wiring needs.

NVIDIA Corporation (NVDA)

• NVIDIA continues to hold a dominant incumbent position in AI compute, acting as the premier system architect for AI acceleration hardware.
• High-performance GPUs are effectively booked out through 2028, with instances of secondary market hardware reselling for up to 4x the original purchase price.
• Massive market demand allows the company to focus primarily on high-volume, mainstream enterprise computing workloads, leaving room for niche startups to capture specialized subsectors at the margins.

Takeaways

• Hardware demand and pricing power remain robust with multi-year revenue visibility, though long-term margin optimization could open the door for specialized alternative chips.

Hyperscalers & Big Tech (MSFT, GOOGL, META, AMZN)

• Combined capital expenditures (CapEx) among the major cloud hyperscalers are expected to jump from approximately $700 billion this year to $1 trillion next year.
• Unlike traditional software development, scaling AI performance is no longer purely an engineering bottleneck; capital is being directly converted into compute capacity, which directly drives intelligence capabilities.
• Hyperscalers are seeing broad-based, non-speculative demand across frontier AI research labs, enterprise clients, and native AI software applications.
• Frontier AI models now cost $3 billion to $5 billion to train, requiring massive downstream inference revenue (estimated at $10 billion+) to justify the investment.

Takeaways

• Big tech infrastructure spending is accelerating at historic levels, ensuring sustained capital inflows to hardware, facility, and component suppliers.

AI Memory & Custom Silicon (ASICs)

• A severe bottleneck has emerged in memory and chip supply chains, with leading memory manufacturers reporting that current demand will require 3 years of capacity to fulfill.
• The cost to train modern frontier models makes it economically viable to build bespoke ASICs (Application-Specific Integrated Circuits) costing $1 billion per model if it yields a 20% efficiency gain during inference.
• GPU architectures and system designs from the previous compute era are reaching physical limits, requiring full-system redesigns across chip interconnects, matrix-multiplication accelerators, and memory hierarchies.

Takeaways

• High-bandwidth memory (HBM) and custom ASIC silicon designers have substantial growth runway as the industry shifts focus from general-purpose GPUs to cost-effective inference hardware.

Energy, Utilities & Power Infrastructure

• By 2028, new data centers will require approximately 44 gigawatts (GW) of power capacity, outstripping the expected 25 GW of grid additions.
• Rack power requirements are increasing from standard 5–10 kW loads to 100–250 kW per rack, forcing data centers to move to high-voltage 800V DC power.
• The power shift is constrained by severe supply shortages of critical equipment like turbines and transformers, as well as a severe labor shortage (only 2% of US electricians are certified for high-voltage DC power).
• Data centers increasingly need to build independent power-generation solutions that operate symbiotically with local municipal grids.

Takeaways

• Power generation, electrical equipment manufacturers (turbines, transformers), and specialized grid contractors represent critical bottleneck plays essential to continued AI expansion.

Data Center Materials & Physical Facilities

• Compute density increases of roughly 70x have made air cooling obsolete, turning eco-friendly liquid cooling into an industry requirement for new state-of-the-art facilities.
• The physical weight and acoustics of modern high-density hardware have led to price inflation for industrial materials like reinforced concrete to support floor loads and dampen noise.
• Fundamental raw materials such as copper are facing structural demand pressures from electrical wiring and interconnect needs, pushing resource constraints down to the mining level.
• Regulatory friction, permitting delays, and power limitations in the US are prompting operators to construct new data centers internationally, such as in Mexico and Australia.

Takeaways

• Investment opportunities extend beyond pure technology into industrial hardware, advanced liquid-cooling systems, physical data center construction, and core commodities like copper.