Shift your investments toward the physical layer of the AI compute stack and hardware infrastructure rather than pure software layers. Capitalize on the transition to custom silicon and optical data center designs by accumulating shares of Broadcom (AVGO) and Lumentum (LITE). Exploit the severe structural supply-demand mismatch in memory by targeting dominant players like SK Hynix and database innovators like MongoDB (MDB). Position portfolios for the multi-year infrastructure boom surrounding next-generation power solutions and computing bottlenecks leading up to 2030. Diversify long-term holdings into robotics and autonomous systems by monitoring emerging opportunities in healthcare and social automation to capture global aging trends.

Semiconductors, Hardware & Compute Infrastructure

• Transitioning from a software-centric world to a compute-centric world, with the base layer of compute increasing by roughly 10,000x (moving from a $10,000 server to a $1,000,000 server). • Global market capitalization has shifted significantly: • $10+ trillion in software, services, and internet • $22–$23 trillion in the "Magnificent 7" (Mag7) • $30+ trillion in chips and hardware • Data centers are undergoing a massive redesign driven by the physical limits of power density, bandwidth, heat, and grid capacity: • Transitioning from megawatts to gigawatts • Moving data transmission distances from kilometers down to meters, centimeters, and millimeters • Shifting infrastructure from a regime of copper to a regime of light (optics) • Rising power architectures moving toward 800-volt systems and solid-state transformers • The chip industry features duopolistic power in key categories, very high profit margins, and strong pricing power rather than operating as a low-margin commodity. • Specific companies mentioned in hardware, chips, data center design, and next-gen computer architectures: • Broadcom (AVGO) – Mentioned regarding XPUs, AI co-design for chips, and the rollout of new custom chips (like the "jalapeno chip"). • Lumentum (LITE) – Mentioned regarding the introduction of optics into next-gen data center designs. • D-Matrix – Mentioned as a next-gen computer architecture accelerator. • Cerebras – Mentioned in the broader context of AI chip infrastructure. • SambaNova – Mentioned as an AI compute/architecture company. • PsyQuantum – Mentioned regarding utility-scale quantum computing developments targeting 2030.

Takeaways

• Investment thesis heavily favors the physical layer of AI (compute stacks, power, grid infrastructure, and silicon co-design) over the pure software layer, which is viewed as temporarily sleepy. • Look for companies positioned in the "token flow"—either creating foundational compute, serving tokens, or packaging them with enterprise context layers. • Long-term structural tailwinds exist in physical infrastructure, material science, custom silicon design, and next-generation power solutions (such as Small Modular Reactors/SMRs and nuclear fusion) with major milestones anticipated around 2030.

Memory & The "Rampocalypse"

• A major structural bottleneck is occurring around memory intensity (Rampocalypse), driven by models requiring massive high-bandwidth memory (HBM), stack DRAM, and high-bandwidth flash to emulate human brain architectures. • There is a structural supply-demand mismatch in duration: building a memory fab takes 3 to 4 years, whereas AI demand is immediate and surging. • Mentioned memory and storage players: • SK Hynix – Highlighted as one of the few dominant global memory players with an upcoming public offering, high demand premiums, and massive supply constraints. • MongoDB (MDB) – Mentioned as a key database platform supporting real-time vector search and embeddings for modern AI agents.

Takeaways

• Investors must navigate the duration mismatch between immediate memory shortages and the multi-year timeline required to build new manufacturing capacity. • Companies solving memory-to-compute arbitration and custom memory co-design are critical bottlenecks in the current AI infrastructure boom.

Robotics & Autonomous Systems

• A massive physical-layer revolution is underway in robotics, combining foundational LLM "brains" with motor-control world models and hardware bodies. • Geographic divergence in the robotics market: • China has roughly 130 to 140 robotics companies and expects 30 to 40 potential IPOs, largely utilizing public markets as a funding mechanism. • The United States has very few robotics IPOs on the horizon, though Western labs maintain an edge in foundational model and brain development. • Specific robotics and hardware companies referenced: • Figure AI – Humanoid robotics company focused on commercial use cases, package sorting, and manufacturing. • Boston Dynamics (owned by Hyundai) – Creator of the hydraulic humanoid robot Atlas. • Emerging non-industrial use cases include social robotics addressing elderly care, demographic population cliffs, and the loneliness epidemic.

Takeaways

• The future of robotics may involve a mix-and-match approach—pairing Western AI brains with Asian manufacturing complexes for physical bodies. • Look beyond industrial automation toward social, healthcare, and consumer robotics as high-growth long-term markets driven by global aging trends.