Maintain a core long-term position in NVIDIA (NVDA), as its pricing power and massive demand from hyperscalers could eventually drive its market cap toward $10 trillion. Diversify into the "connectivity layer" by investing in Astera Labs (ALAB), which acts as essential plumbing for the massive data center clusters required for AI scaling. Capitalize on the critical shortage of high-speed memory by holding Micron Technology (MU), a primary beneficiary of the "memory problem" that currently constrains AI performance. For exposure to the next phase of AI, look to Apple (AAPL) for on-device "Edge AI" and Unity Software (U) for the simulation tools needed to train humanoid robots. Given the high energy demands of data centers, consider energy solution providers like Bloom Energy (BE) while using NASDAQ 100 (QQQ) puts to hedge against broader market volatility.

Based on the discussion with veteran investor Gavin Baker, the following investment insights highlight a shift in focus from AI software to the physical infrastructure—the "watts and wafers"—that powers the industry.

NVIDIA (NVDA)

• Baker has held a position in NVIDIA for over 20 years, demonstrating extreme long-term conviction. • He believes the company can maintain its high profit margins and meet massive demand, potentially pushing its market cap toward $10 trillion (it is currently roughly halfway there). • The transcript suggests that NVIDIA could sell $2 to $3 trillion worth of GPUs annually if supply chain constraints (specifically from TSMC) were removed.

Takeaways

• Long-term Hold: Baker views NVIDIA not as a bubble stock, but as a generational opportunity that still has significant room to double in value. • Margin Resilience: Investors should look for NVIDIA's ability to maintain pricing power even as competitors enter the space.

Astera Labs (ALAB)

• This is one of the largest positions in Baker’s fund (approximately 7.4% to 9%). • Astera Labs acts as the "connectivity layer" or "plumbing" between GPUs. • As AI clusters scale to hundreds of thousands of chips, the bottleneck shifts from the raw power of the GPU to the speed at which data can be transferred between them. Astera Labs solves this connectivity bottleneck.

Takeaways

• Infrastructure Play: Move beyond the chip makers to the companies that enable those chips to talk to each other. • Scalability: The company is a direct beneficiary of the trend toward larger and more complex data centers.

Micron Technology (MU)

• Micron is a leader in AI memory (HBM - High Bandwidth Memory). • The transcript notes that Micron grew nearly 10x in a single year, highlighting the critical nature of the "memory problem" in AI. • Demand is so high that competitors like SK Hynix are seeing companies like Google and Microsoft offer $50-$100 billion just to secure future supply.

Takeaways

• Memory is Essential: AI models cannot function without massive amounts of high-speed memory; Micron is a primary "picks and shovels" play for this requirement.

Unity Software (U)

• While known as a gaming engine, Baker views Unity as a "world model builder." • Unity’s software understands physics, lighting, and textures, which is essential for simulating virtual environments to train humanoid robots and AGI (Artificial General Intelligence).

Takeaways

• Robotics/AGI Proxy: Unity offers a unique way to invest in the "physical" training of AI without buying hardware. • Diversification: It represents a "forward-looking" bet on where AI training will go after the current text-based LLM phase.

Cerebras Systems & Positron

• Cerebras recently IPO’d and has seen significant gains (up 40% post-IPO). • Both companies focus on inference chips—the hardware used when an AI model is actually "thinking" or answering a prompt, rather than just being trained. • Baker believes the revenue opportunity for inference is 5x to 10x larger than the opportunity for pre-training.

Takeaways

• Shift to Inference: The "pre-training" phase (building the model) is maturing; the "inference" phase (using the model) is where the long-term volume and money reside.

Apple (AAPL)

• Although Baker’s specific investment interest wasn't detailed, he is "hugely bullish" on Apple as the dominant player for Small Language Models (SLMs). • Apple is positioned to run AI locally on-device, which is critical for privacy-sensitive data like medical or financial records.

Takeaways

• Edge AI: Apple is the primary play for "Local AI" where the processing happens on your phone rather than in a giant data center.

Investment Themes & Sector Insights

The "Watts and Wafers" Thesis

• Baker argues that AI is not in a bubble because it is constrained by physical reality: Electricity (Watts) and Semiconductor Fabrication (Wafers). • Unlike the Dot-com bubble, which was fueled by debt, the current AI cycle is funded by the massive free cash flow of the "Hyperscalers" (Google, Microsoft, Amazon, Meta).

The QQQ Hedge

• Interestingly, Baker holds a large Put position (a bet that the price will go down) on the NASDAQ 100 (QQQ). • Insight: While he is extremely bullish on specific AI infrastructure, he is bearish on the broader market. This suggests a "stock picker's market" where only the winners of the infrastructure race will thrive while the rest of the tech sector may struggle.

Sovereign Infrastructure & Energy

• Energy Constraints: 40% of new data centers face protests or grid limitations. • Opportunities: Companies solving the energy crisis (like Bloom Energy) or those looking toward Orbital Compute (Space-based data centers via SpaceX) are key areas for future growth.

Key Risk Factors

• TSMC Bottleneck: If TSMC (Taiwan Semiconductor) suddenly increased capacity, it might actually trigger a bubble by causing companies to over-leverage themselves to buy chips. Currently, TSMC's inability to keep up with demand is actually keeping the market "sustainable."