Author: George Kikvadze

Translated by: Deep Tide TechFlow

Deep Tide Guide: George Kikvadze, Vice Chairman of Bitfury Group, presents a contrarian viewpoint: the most profitable opportunities in the AI field are not in the model layer, but in the infrastructure bottlenecks such as power, cooling, memory, and networks. He identifies 7 "choke points" in AI systems and reveals his portfolio of 14 targets, with a current return of approximately 60%. This "bottleneck investment" framework is worth every AI investor's serious consideration.

To understand where to make money in AI, don’t look at the headlines; look at where the system is under pressure.

The simplest analogy: today’s AI is like a factory with unlimited orders, but power, cables, and cooling cannot keep up.

This mismatch itself is an opportunity.

After conducting detailed due diligence, we have bet 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 are asking: "Who will win in AI?" This question is fundamentally wrong.

The question should be: Where will the system break? Who will profit from the repairs?

In the market, dependencies are leverage.

AI dependencies are not abstract; they are all tangible:

• Megawatt-level power
• Transformer delivery cycles
• Heat dissipation capacity per cabinet
• Memory bandwidth

The economic focus is shifting to these areas.

The Only Required Analytical Framework

AI Expansion → Infrastructure Under Pressure → Forced Investment → Bottleneck → Pricing Power → Earnings Upgrade

When demand is rigid and supply is constrained: prices move first, profits follow, and stock prices are re-evaluated last.

Why Now

A few numbers explain the entire issue:

Nearly 50% of data center projects in the U.S. are currently delayed, not due to a lack of demand or funds, but because power cannot be secured. Transformer delivery cycles have stretched from 24 months before 2020 to over 5 years now. The construction cycle for data centers is 18 months. This equation doesn’t add up.

Hyperscale vendors are projected to spend $700 billion on AI infrastructure alone by 2026, close to six times the amount in 2022. Amazon: $200 billion, Google: $175-185 billion, Meta: $115-135 billion. None of these companies are slowing down.

Semiconductors currently account for 42% of the total market value of the S&P 500 IT sector, more than doubling since the bear market bottom in 2022 and more than quadrupling its weight from 2013. Semiconductors also contribute 47% of the IT sector's forward EPS, nearly tripling compared to 2023.

The market is flowing to the compute layer with unprecedented density.

But compute is no longer the bottleneck.

Capital is flooding into chips, but the real constraints have shifted elsewhere.

This gap presents a trading opportunity.

Bottleneck Map: Where is the Pressure?

• Power: The Foundation

AI cannot expand without power. Period.

The U.S. needs to add the equivalent of the current entire power capacity of data centers every two years to keep up with AI demand forecasts before 2030. Nuclear power is the only source that can provide the scale and reliability needed by hyperscale vendors, but even the fastest nuclear restart takes years.

Targets: $CEG $GEV $VST $WMB

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

Constellation Energy ($CEG) operates the largest fleet of nuclear power plants in the U.S., being one of the few suppliers that can provide large-scale, reliable, zero-carbon baseload power. Hyperscale vendors are accelerating long-term power purchase agreements with nuclear suppliers, with Constellation right in the middle of this demand pathway.

GE Vernova ($GEV) is building the framework for the next energy cycle, covering gas turbines, renewable energy, and grid solutions. As AI demand accelerates, the ability to quickly and massively deploy power becomes crucial, positioning GE Vernova’s gas turbine and electrification capabilities at the core of this transformation.

Vistra Corp ($VST) has a diversified power generation portfolio that includes nuclear, gas, and retail power, capable of addressing both baseload and peak demand simultaneously. AI workloads create highly volatile power demands, making such flexibility particularly valuable.

Williams Companies ($WMB) operates one of the largest natural gas pipelines in the U.S., providing fuel to bridge the gap between current demand and future nuclear power scale. In the expansion of AI infrastructure, natural gas is the fastest way to bring online incremental power. Williams is effectively the raw energy supplier for AI growth.

The Grid and Electrification: Constraints Behind Power

Generating power is one thing; delivering it is much harder.

The interconnected U.S. grid is now queuing for connectivity until after 2030. Meeting existing commitments in the next decade will require more than $50 billion in transmission investment, not accounting for the onboarding of a new AI data center.

Targets: $PWR $ETN

The timetable is slipping here, and profit margins are widening. Companies solving the "last mile" delivery problem possess lasting long-cycle pricing power.

Quanta Services ($PWR) is a leading contractor for constructing and upgrading transmission infrastructure, connecting energy generation and consumption. As grid congestion becomes a major bottleneck for AI expansion, Quanta is directly on the path of many multi-year, non-discretionary capital expenditures. Its backlog of orders serves as a forward indicator of grid pressure.

Eaton Corporation ($ETN) provides distribution systems, switchgear, and power management technologies that enable the safe and efficient distribution of power at scale. As data centers advance to higher power densities and more complex energy flows, Eaton’s components shift from standardized hardware to critical infrastructure.

Cooling: The Silent Ceiling

Heat throttles performance. Thermodynamics does not have software patches.

The goal for the next generation of AI facilities is 250 kW per cabinet, while a standard enterprise data center only had 10-15 kW ten years ago. Liquid cooling is no longer optional; it is a necessary infrastructure. Each GPU sold necessitates corresponding cooling capacity; this ratio will not change.

Target: $VRT

Vertiv is close to a monopoly in the cooling sector for hyperscale data centers. This is one of the most severely undervalued links in the entire AI stack, as no one cares 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 cabinets transition from air cooling to liquid cooling, Vertiv is at the center of this structural upgrade cycle, expanding directly in synchronization with AI compute deployments. This is not optional expenditure; it is a prerequisite for normal operation.

Memory: The Next Bottleneck

AI is transitioning from being compute-constrained to memory-constrained.

As models grow larger and inference volumes explode, memory bandwidth and capacity become the constraints, rather than 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 a major beneficiary in the West.

Core Target: $MU

This is the next wave for earnings upgrades. Most portfolios have not yet positioned for this. When the market catches on, they will.

Micron Technology ($MU) is one of the few manufacturers capable of mass-producing advanced HBM, which is a key component for AI training and inference workloads. As memory becomes the limiting factor for system performance, Micron is transitioning from a historically cyclical supplier to a structural beneficiary of AI demand. This transition has not yet been fully reflected in valuations, leaving room for sustained earnings upgrades and multiple expansion.

Networking: The Throughput Layer

The speed of an AI cluster depends on the slowest connection.

A network bottleneck can cause a cluster of thousands of GPUs to stagnate, wasting hundreds of millions of dollars in capital for each facility. As cluster sizes expand to 100,000 GPU configurations, interconnect issues become exponentially magnified. One bottleneck, and everything stops.

Targets: $ANET $ALAB

Quiet, critical, and under-owned. No one talks about networking until problems arise.

Arista Networks ($ANET) builds high-performance networking infrastructure that allows data to flow seamlessly in large-scale AI clusters. When workloads demand ultra-low latency and high throughput, Arista’s software-defined networks become essential for maintaining cluster efficiency. The costs of downtime or inefficiency are extremely high, and Arista captures value by ensuring systems run at full speed.

Astera Labs ($ALAB) operates within the data path, ensuring high-speed connections among GPUs, CPUs, and memory in AI systems. As cluster density increases, bottlenecks shift from the network edge to chip-to-chip communication, which is exactly where Astera operates. In a high-performance AI environment, if communication between components is not fast enough, the whole system slows down.

Manufacturing: Long-Cycle Constraints

AI cannot scale without chip manufacturing capabilities. Without manufacturing tools, advanced chips cannot be produced.

ASML's EUV lithography machines have a production cycle of more than one year, with a single unit costing more than $200 million and no credible alternatives available. 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 wafer fab in the world.

Targets: $ASML $LRCX

Long-cycle constraints. Structurally harder to disrupt than any software moat. The discussion's intensity is far below what it should be.

ASML Holding ($ASML) is the sole supplier of EUV lithography systems, the most advanced chip manufacturing tools currently available, and a prerequisite for producing cutting-edge semiconductors. With a backlog of multi-year orders and no viable competitors, ASML controls crucial chokepoints in the global chip supply chain.

Lam Research ($LRCX) supplies the etching and deposition equipment that form the backbone of semiconductor manufacturing. Its tools are deeply integrated into all major wafer fabs, making it a cyclical yet indispensable partner in chip capacity expansion. As AI demand drives sustained capacity expansion, Lam secures long-cycle revenues closely linked to global semiconductor manufacturing growth.

Misclassification: A Source of Alpha

This is a part most investors overlook and is the most asymmetrical opportunity on the entire map.

There is a category of companies that the market prices as A, but the operational and financial realities have already become B.

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

The market still sees them as 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 no one was paying attention and built the infrastructure ahead of demand. Today, hyperscale vendors are frantically vying for exactly these two items.

Cipher Digital has already begun executing its transformation by signing a 15-year lease with investment-grade hyperscale tenants (for the third AI/HPC park) and securing a $200 million revolving credit facility from top global banks. These are not speculative moves; they are commitments to long-cycle revenues.

IREN is executing the same strategy across multiple sites, combining energy procurement with scalable data center construction. Its advantage lies in speed: it has already secured the land, power, and infrastructure needed to transition to AI workloads.

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

This gap will converge. When it does, it won’t be slow.

Portfolio Overview

This is not a bunch of stocks; it is a system.

Each position corresponds to a specific constraint in the AI stack, and every constraint must be resolved for the system to operate. This is 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 Have Yet to Complete

We are transitioning from compute scarcity to infrastructure scarcity.

This means:

• GPUs are no longer the only narrative
• Power, the grid, memory, and cooling have become the dominant profit drivers
• Returns follow constraints rather than hype

Most portfolios are still stuck in the old world.

Risks: Discipline is Also Important

This framework may fail under certain conditions. These should be addressed honestly.

If hyperscale vendors slow down capital expenditures. If Amazon, Google, and Meta slow their infrastructure spending due to margin pressure or lower-than-expected demand, the assumption of rigid demand will weaken. This is the primary risk to monitor; keep an eye on quarterly capital expenditure guidance as a leading indicator.

Bottleneck resolution speeds up faster than anticipated. Government interventions in transformer manufacturing, acceleration of nuclear power approvals, or restructuring of grid interconnect queues could compress the premium on constrained infrastructure. These changes may be slow but are real.

Regulatory friction. Power and grid infrastructures intersect with utility regulation, environmental reviews, and rate-setting agencies. When regulation turns adverse in this field, it will structurally and durably limit return ceilings.

The key distinction is: this is not a bet on product cycles. Product cycles can reverse in a quarter. Industrial constraints take years to build and also require years to resolve. This asymmetry is the point.

In Conclusion

In every industrial era, wealth is not created by the companies making trains.

But by the companies that own the tracks, coal, and rights of way.

The tracks of AI are measured in megawatts, transformer delivery cycles, and cooling capacity per cabinet.

Most investors are chasing AI. The real opportunity lies in the essentials that AI cannot do without.

In every system, headlines follow innovation, profits follow constraints. We focus on constraints rather than narratives, and our current return is about 60%. As AI infrastructure accelerates, this is not the end of the trade; it is still early. We believe we are still in the third inning.