Investors should shift focus from chipmakers to AI infrastructure, specifically targeting Bloom Energy (BE) for its ability to bypass five-year grid delays with modular fuel cells deployable in 90 days. To address the data transmission bottleneck, Lumentum Holdings (LITE) is a high-conviction play as AI clusters transition from copper wiring to high-velocity optical fiber. Marvell Technology (MRVL) offers essential exposure to the power dispersal market, serving as a critical "picks and shovels" provider for the entire custom semiconductor industry. For those seeking cloud infrastructure exposure, Nebius (NBIS) is a strategic pick due to its massive contracts with Meta and Microsoft and direct backing from NVIDIA. Monitor the "energy cost per bit" as a key metric, favoring companies that specialize in photonics and energy efficiency to lead the next phase of the AI cycle.

The investment landscape for AI is shifting from the "crowded" GPU trade (NVIDIA) toward the physical infrastructure required to keep those chips running. The current bottleneck is no longer just chip supply, but the energy, cooling, and data transmission infrastructure surrounding them.

Bloom Energy (BE)

• The Problem: Traditional electrical grid connections for new data centers have lead times of up to five years. • The Solution: Bloom provides modular fuel cells and gas turbines that can be deployed on-site in as little as 90 days. • Key Context: • Oracle is using Bloom to power a 2.4 gigawatt data center. • Noted investor Leopold Ashenbrenner reportedly holds a significant position (approx. 12.7% of his portfolio). • Revenue hit $750 million in Q1 2026, up 130% year-over-year.

Takeaways

• Speed as a Moat: In the AI race, getting power today is worth a massive premium over waiting five years for the grid. Bloom acts as the "intermediary step" for immediate power needs. • Execution Risk: The company still needs to prove it can manufacture and deliver these modular units at a massive global scale.

Lumentum Holdings (LITE)

• The Problem: As data center clusters grow to hundreds of thousands of GPUs, the distance data must travel increases. Traditional copper wiring melts or loses efficiency due to heat when handling petabytes of data over long distances. • The Solution: Photonics/Optical Fibers. Using light (lasers) to transmit data between chips. Light has the lowest friction and highest velocity for information transfer. • Key Context: • NVIDIA CEO Jensen Huang has reportedly invested significantly in the company. • Sales volume grew from 20 million units last year to 60 million units this year.

Takeaways

• The Copper-to-Fiber Shift: As AI models grow to tens of trillions of parameters, the "energy cost per bit" makes optical fiber a necessity rather than an option. • Competition: Corning (GLW) is a major player providing the actual glass fiber, while Lumentum focuses on the transmitters and receivers.

Marvell Technology (MRVL)

• The Problem: Distributing power within a chip and between chips is becoming incredibly complex as power demands per server rack increase 10x (from 10kW to 100kW+). • The Solution: Marvell specializes in the design of power dispersal within silicon. They ensure the right amount of electricity reaches the right part of the chip at the right time. • Key Context: • Jensen Huang recently referred to Marvell as a potential "next trillion-dollar company." • The stock surged significantly (approx. 76-80%) following these mentions and its inclusion in the S&P 500.

Takeaways

• Diversification: Unlike other vendors, Marvell is used by Broadcom and other firms trying to build their own custom AI chips to compete with NVIDIA. They are a "picks and shovels" play for the entire semiconductor design industry.

Nebius (NBIS)

• The Problem: Large companies (Hyperscalers) like Meta and Microsoft often prefer to rent "ready-to-use" compute rather than building and maintaining every data center themselves. • The Solution: A "NeoCloud" provider. They build the data centers, handle the complex wiring, cooling, and software optimization, and then rent the capacity back to big tech. • Key Context: • Formed from the remains of the Russian tech giant Yandex. • Holds massive contracts: $20 billion with Microsoft and $27 billion with Meta. • Backed by NVIDIA with direct equity investments.

Takeaways

• Regulatory Arbitrage: Companies like Nebius, CoreWeave, and IREN win because they already have the permits and land-use rights that take years for others to acquire. • Financial Engineering: Renting from NeoClouds allows big tech companies to spread their massive AI spending over years as an "expense" rather than a one-time "capital expenditure" that might scare shareholders.

Investment Themes & Sector Insights

The "Energy Bottleneck"

• Lead Times: The primary constraint on AI growth is now the 3-to-5-year wait for transformers, wiring, and grid access. • The 10% Irony: Currently, only 10% of an AI lab's budget goes to power/infrastructure (90% goes to chips), yet that 10% is causing 100% of the delays.

Emerging Metrics

• Joules per Bit: Investors should watch the energy efficiency of moving data. As this becomes the primary cost, companies specializing in Photonics and Liquid Cooling will likely see increased capital inflows.

The "Jensen Huang" Effect

• The transcript notes a "copy trade" phenomenon where companies mentioned or invested in by NVIDIA’s CEO (Marvell, Lumentum, Nebius) see immediate and massive market revaluations.