AI Bottleneck Investment Strategy: 14 Targets Covering Every Layer from Power to Lithography

Original Author: George Kikvadze

Original Translation: TechFlow

Introduction: George Kikvadze, Vice Chairman of Bitfury Group, proposes a contrarian perspective: the most lucrative opportunities in the AI sector are not at the model layer, but in infrastructure bottlenecks such as power, cooling, memory, and networking. He breaks down 7 critical "chokepoints" in AI systems and publicly shares his portfolio of 14 stocks, which has yielded approximately 60% returns so far. This "bottleneck investment" framework is worth a close look for anyone focused on AI investing.

To understand where the money can be made in AI, don't look at the headline news. Look at where the system is under pressure.

The simplest analogy: today's AI is like a factory with infinite orders, but the power, cables, and cooling can't keep up.

This mismatch itself is the opportunity.

After detailed due diligence, we are betting on the following "AI Bottleneck" portfolio:

$CEG $GEV $VST $WMB $PWR $ETN $VRT $MU $ANET $ALAB $ASML $LRCX $CIFR $IREN

The Real Question to Ask

Most investors ask: "Who will win the AI race?" That's the wrong question.

The right question is: Where will the system break? Who gets paid to fix it?

In markets, dependencies are leverage.

AI's dependencies are not abstract; they are all physical:

• Megawatt-scale power
• Transformer lead times
• Per-rack cooling capacity
• Memory bandwidth

The economic center of gravity is shifting to these areas.

The Only Analytical Framework You Need

AI Expansion → Infrastructure Strain → Forced Investment → Bottlenecks → Pricing Power → Earnings Upgrades

When demand is inelastic and supply is constrained: prices move first, earnings follow, and stock valuations are re-rated last.

Why Now

A few numbers tell the whole story:

Nearly 50% of data center projects in the US are currently delayed, not due to a lack of demand or capital, but because they can't secure power. Transformer lead times have stretched from 24 months before 2020 to over 5 years now. Data center construction cycles are 18 months. The math simply doesn't add up.

Hyperscalers are set to spend $700 billion on AI infrastructure alone by 2026, nearly six times the amount in 2022. Amazon: $200B, Google: $175-185B, Meta: $115-135B. Not one of them is slowing down.

Semiconductors currently represent 42% of the total market cap of the S&P 500 IT sector, more than double their weight from the 2022 bear market bottom, and over four times their weight in 2013. Semiconductors also contribute 47% of the IT sector's forward EPS, nearly triple the figure from 2023.

The market is converging on the compute layer with unprecedented density.

But compute is no longer the bottleneck.

Capital is flooding into chips, but the real constraint has shifted elsewhere.

This discrepancy is the trading opportunity.

The Bottleneck Map: Where the Pressure Really Is

• Power: The Foundation

AI cannot scale without power. Period.

The US needs to add capacity equivalent to the entire current data center power base every two years to keep up with AI demand projections through 2030. Nuclear power is the only source capable of providing the scale and reliability hyperscalers require for baseload, but even the fastest reactor restarts take years.

Positions: $CEG $GEV $VST $WMB

These are not utility stocks; they are AI capacity providers. The market has not yet completed this reclassification. This mispricing is the opportunity.

Constellation Energy ($CEG) operates the largest fleet of nuclear power plants in the US, making it one of the few suppliers capable of providing massive, reliable, zero-carbon baseload power. Hyperscalers are accelerating long-term power purchase agreements with nuclear suppliers, and Constellation sits directly on this demand path.

GE Vernova ($GEV) is building the generation backbone for the next energy cycle, spanning gas turbines, renewables, and grid solutions. As AI demand accelerates, the ability to deploy power quickly and at scale becomes critical, and GE Vernova's gas turbines and electrification capabilities are at the heart of this.

Vistra Corp ($VST) has a diversified generation portfolio including nuclear, gas, and retail power, allowing it to handle both baseload and peak demand. The volatile power demands of AI workloads make this flexibility particularly valuable.

Williams Companies ($WMB) operates one of the largest natural gas pipeline networks in the US, providing the fuel to bridge the gap between current demand and future nuclear capacity. Natural gas is the fastest way to bring incremental power online for AI infrastructure expansion. Williams is, in effect, a fuel supplier for AI growth.

The Grid & Electrification: The Constraint Behind the Power

Generating power is one thing; delivering it is harder.

The interconnection queue for the US power grid is now booked beyond 2030. Meeting existing commitments alone will require over $50 billion in transmission investment over the next decade, and that's without adding a single new AI data center.

Positions: $PWR $ETN

This is where timelines slip and margins expand. Companies solving the "last mile" delivery problem have durable, long-cycle pricing power.

Quanta Services ($PWR) is a leading contractor for building and upgrading transmission infrastructure, connecting power generation to consumption. As grid congestion becomes a primary bottleneck for AI expansion, Quanta sits directly on a multi-year, non-discretionary capital expenditure path. Its backlog is a leading indicator of grid stress.

Eaton Corporation ($ETN) provides power distribution systems, switchgear, and power management technologies that enable safe and efficient power delivery at scale. As data centers push toward higher power densities and more complex energy flows, Eaton's components transition from standardized hardware to critical infrastructure.

Cooling: The Silent Ceiling

Heat kills performance. There is no software patch for thermodynamics.

Next-generation AI facilities target 250 kW per rack, compared to the 10-15 kW standard for enterprise data centers a decade ago. Liquid cooling is no longer an option; it's essential infrastructure. Every GPU sold requires corresponding cooling capacity, and this ratio is fixed.

Position: $VRT

Vertiv holds a near-monopoly position in hyperscale data center cooling. It is one of the most undervalued parts of the entire AI stack because no one thinks about cooling until the cluster goes down.

Vertiv Holdings ($VRT) designs and deploys thermal management systems that keep high-density AI clusters operational under extreme power loads. As racks shift from air cooling to liquid cooling, Vertiv sits at the center of this structural upgrade cycle, expanding in direct sync with AI compute deployment. This is not optional spending; it's a prerequisite for uptime.

Memory: The Next Bottleneck

AI is shifting from being compute-bound to being memory-bound.

As models grow larger and inference volumes explode, memory bandwidth and capacity become the binding constraints, not raw processing power. HBM (High Bandwidth Memory) supply is already tight. The top three global AI memory suppliers control over 90% of global HBM output. Micron is the primary Western beneficiary.

Core Position: $MU

This is the next wave of earnings upgrades. Most portfolios are not positioned for it yet. When the market catches on, they will be.

Micron Technology ($MU) is one of the few companies globally capable of mass-producing advanced HBM, a critical component for AI training and inference workloads. As memory becomes the limiting factor for system performance, Micron transitions from a historically cyclical supplier into a structural beneficiary of AI demand. This shift is not yet fully reflected in valuations, leaving room for sustained earnings upgrades and multiple expansion.

Networking: The Throughput Layer

An AI cluster is only as fast as its slowest connection.

A single network bottleneck can stall an entire cluster of thousands of GPUs, wasting hundreds of millions of dollars in capital per facility. As cluster scales extend to 100,000 GPU configurations, interconnection issues grow exponentially. One choke point, and everything stops.

Positions: $ANET $ALAB

Quiet, critical, under-owned. No one talks about the network until the network breaks.

Arista Networks ($ANET) builds high-performance networking infrastructure that allows data to flow seamlessly within massive AI clusters. When workloads demand ultra-low latency and high throughput, Arista's software-defined networking becomes key to maintaining cluster efficiency. The cost of downtime or inefficiency is enormous, and Arista captures value by ensuring the system runs at full speed.

Astera Labs ($ALAB) operates inside the data path, ensuring high-speed connections between GPUs, CPUs, and memory within AI systems. As cluster density increases, bottlenecks move from the network edge to chip-to-chip communication, which is exactly where Astera sits. In high-performance AI environments, if components can't communicate fast enough, the entire system slows down.

Manufacturing: The Long-Cycle Constraint

You cannot scale AI without the ability to manufacture chips. You cannot manufacture advanced chips without manufacturing tools.

ASML's EUV lithography machines have a production cycle over a year, cost over $200 million each, and have no credible substitute. Every advanced chip on earth, from NVIDIA's H100 to Apple's M-series, requires their equipment. Lam Research's etching and deposition tools are embedded in the production lines of every major fab in the world.

Positions: $ASML $LRCX

Long-cycle constraints. Structurally harder to disrupt than any software moat. Their discussion volume is far lower than it should be.

ASML Holding ($ASML) is the sole supplier of EUV lithography systems, the most advanced chip-making tools in existence and a prerequisite for producing cutting-edge semiconductors. With a multi-year order backlog and no viable competitor, ASML controls a critical chokepoint in the global chip supply chain.

Lam Research ($LRCX) supplies the etching and deposition equipment that forms the backbone of semiconductor manufacturing. Its tools are deeply embedded in all major fabs, making it a recurring and indispensable partner in chip capacity expansion. As AI demand drives ongoing capacity additions, Lam captures long-cycle revenue directly tied to global semiconductor manufacturing growth.

Misclassification: The Source of Alpha

This is the part most investors miss. It's the most asymmetric opportunity on the map.

There is a class of companies that the market prices as 'A', but their operational and financial reality is already 'B'.

Take $CIFR (Cipher Digital) and $IREN (IREN Limited).

The market still sees Bitcoin miners.

What they are becoming is far more valuable: AI Power Infrastructure and HPC Data Center Platforms.

These companies locked in low-cost power when nobody was watching and built out infrastructure before the demand arrived. Today, those are exactly the two things hyperscalers are scrambling to acquire.

Cipher Digital has already begun executing its transition, signing 15-year leases with investment-grade hyperscale tenants (its third AI/HPC campus) and securing a $200 million revolving credit facility from top-tier global banks. These are not speculative moves; they are long-cycle revenue commitments.

IREN is executing the same strategy across multiple sites, combining energy access with scalable data center construction. Its advantage is speed: it already controls the land, power, and infrastructure needed to pivot to AI workloads.

The market still sees miners. The balance sheets already look like infrastructure companies.

This gap will converge. And when it does, it won't be slow.

Portfolio at a Glance

This isn't a collection of stocks. It's a system.

Each position targets a specific constraint in the AI stack. Each constraint must be resolved for the system to function. That's the discipline.

• Power: $CEG $GEV $VST $WMB
• Grid: $PWR $ETN
• Cooling: $VRT
• Memory: $MU
• Networking: $ANET $ALAB
• Manufacturing: $ASML $LRCX
• Misclassification: $CIFR $IREN

The Cognitive Shift Most Investors Haven't Made

We are moving from compute scarcity to infrastructure scarcity.

This means:

• GPUs are no longer the only narrative
• Power, grid, memory, and cooling become dominant earnings drivers
• Returns follow constraints, not hype

Most portfolios are still positioned for the old world.

Risks: Discipline Matters Too

This framework breaks down under certain conditions. They deserve honest acknowledgment.

Hyperscaler CapEx Slowdown. If Amazon, Google, and Meta pull back on infrastructure spending due to margin pressure or weaker-than-expected demand, the assumption of inelastic demand weakens. This is the primary risk to monitor. Watch quarterly CapEx guidance as a leading indicator.

Bottlenecks Resolve Faster Than Expected. Government intervention in transformer manufacturing, accelerated nuclear approvals, or streamlining the grid interconnection queue could compress the premium on constrained infrastructure. These changes are slow but real.

Regulatory Friction. Power