Why Top Founders Are Racing Into AI Infrastructure

Fund Overview and Thesis

• The "Machine Age Fund" is launched to address the unprecedented infrastructure bottlenecks required to support the AI revolution, characterized by the thesis that hardware constraints, rather than software engineering, are now the primary limiting factor.
• The fund defines the current era as "Machine Intelligence" rather than just AI, emphasizing that the "machine" (hardware, power, cooling, supply chain) is the critical differentiator for future value creation.
• The investment scope covers the entire physical stack of computing infrastructure: chips, memory, networking, power systems, cooling, and facilities, down to raw material sourcing like copper mines.
• Founders of AI-driven hardware companies represent a growing segment of the market, shifting from ~3-5% of top founder deals historically to over 20-30% currently.

Supply and Demand Dynamics

• Hyperscalers project collective capital expenditure (CapEx) to reach $1 trillion next year, up from approximately $700 billion currently, driven by infinite demand signals from frontier labs, enterprise, and global markets.
• GPU supply is fully booked through 2028, with some components facing multi-day auctions where units are resold at four times the purchase price.
• Memory capacity is critically constrained, with leading providers stating that current demand requires three years of existing capacity to satisfy.
• The unit of work in AI is becoming exponentially more consumptive; token usage is expanding from single-chat interactions to multi-agent systems requiring thousands of tokens per task.
• Data center power requirements are shifting dramatically, with rack power density moving from 5–10 kW to 100–500 kW, necessitating a transition from air cooling to liquid cooling.
• By 2028, the industry will need an additional 44 gigawatts of power, significantly outpacing the expected grid addition of 25 gigawatts.

Infrastructure Bottlenecks and Constraints

• Traditional data center designs are obsolete for AI workloads; the industry faces shortages in transformers, turbines, and electrical contractors certified to handle high-voltage DC power (800+ volts).
• Only 2% of US electrical contractors possess certification for DC power systems, creating a severe labor bottleneck for new facility construction.
• Construction lead times for new power generation and data centers are 4–5 years, creating a structural mismatch with AI demand growth which is increasing 10x annually.
• Political and regulatory headwinds, including environmental concerns over water usage and noise, are delaying or preventing the construction of data centers in the US, pushing some operations to Mexico and Australia.
• The "engineering problem" of the past has been replaced by a "resource limitation" problem where pouring more money into systems yields diminishing returns until physical constraints are resolved.

Economic and Operational Shifts

• The industry is moving toward per-model ASICs (Application-Specific Integrated Circuits); with model training costs at $3–5 billion, efficiency gains of 20% justify the $2 billion investment to build a dedicated chip for a single model.
• Unlike traditional software, where margins were historically stable, AI hardware efficiency directly dictates business viability, as the cost of inference must pay back the massive capital investment in training.
• The "Mythical Man Month" concept no longer applies; adding capital (e.g., $3 billion to light up a cluster) can now accelerate capability, whereas adding engineers previously yielded diminishing returns.
• Companies are adopting "employee" AI agents that operate autonomously with their own compute environments, requiring organizations to manage these entities for security, behavior, and productivity alongside human staff.

Investment Strategy and Market Structure

• Incumbents like NVIDIA face a natural market constraint where they focus on "gold bricks" (mass market, incremental improvements), leaving significant "silver brick" opportunities for new entrants to innovate at the margins.
• The fund targets "systems founders" who possess expertise in manufacturing, supply chains, and full-stack architecture, distinguishing them from pure software researchers.
• Founders in this space tend to be more experienced than typical software entrepreneurs due to the complexity of the supply chain and regulatory environment required to build hardware.
• Investment rounds are significantly larger than traditional VC norms, often reaching hundreds of millions at the seed stage due to the high capital requirements for hardware development.
• The fragmentation of the market is expected as use cases multiply, creating niche opportunities for specialized hardware providers even as major players consolidate the core silicon market.

Forward-Looking Outlook

• The fund projects a 5–10 year horizon where success depends on securing supply chains for power, cooling, and materials to prevent the US from losing its infrastructure leadership to other nations.
• The industry expects a "Machine Age" where physical constraints drive innovation, requiring a complete rebuild of the internet infrastructure from copper mines to data center power grids.
• A shift toward eco-friendly, self-generating power systems is predicted, where data centers contribute to the grid by absorbing off-peak power and selling surplus during peak hours.
• The market is expected to mature into a cycle of expansion and consolidation, where new entrants rise to solve specific bottlenecks before eventual industry consolidation occurs.
• The long-term trajectory suggests decades of sustained growth in compute demand as AI applications expand from coding and chatbots to complex reasoning, autonomous agents, and embodied robotics.