The Machine Age Fund: Rebuilding Infrastructure for AI

a16z’s new Machine Age Fund is introduced as a response to the “biggest technological revolution” of our time, AI, which necessitates an entirely new, foundational infrastructure. Ben Horowitz, Martin Casado, Raghu Raghuram, and Erik Torenberg discuss why this fund is critical now, emphasizing that AI’s impact extends beyond traditional computing infrastructure, reaching “all the way down to the mines of copper.”

The Shifting AI Bottleneck and Unprecedented Demand

The discussion highlights a critical shift: the bottleneck in AI has moved from the models themselves, which are rapidly advancing, to the underlying infrastructure, or “south of the model.” This has led to an “infinite” demand for AI, placing “every part of the supply chain under duress.”

Several indicators confirm this isn’t a hype cycle:

• Surging Hyperscaler CapEx: Major hyperscalers, with their comprehensive view of demand from frontier labs, AI-native companies, and enterprises, are increasing their capital expenditure dramatically, projected to reach $1 trillion next year.
• Insane Growth: Application companies and frontier labs are experiencing unprecedented growth rates.
• Rising Prices: Unlike previous tech cycles where component prices typically fell, prices for critical AI components like chips are increasing, indicating severe supply constraints.
• Booked Supply: Key components are “booked out to 2028,” a situation previously unheard of, unlike the speculative “dark fiber” buildout of the internet era. Every GPU produced is already pre-sold.
• Exponential Token Consumption: The progression from chatbots to reasoning and multi-agents exponentially increases the number of tokens required per task. “Nobody likes to use AI more than AI,” creating an autocatalytic effect where AI’s improvement drives its own compute consumption.
• Resource Limitation, Not Engineering Problem: Unlike past engineering-constrained problems, AI’s advancement can now be fueled directly by capital and compute, limited only by the ability to create supply. This means “any problem that you have can be solved with enough infrastructure,” as long as there’s money, GPUs, and data.
• Expanding Use Cases: AI’s application is rapidly expanding from casual apps and coding for professionals to knowledge workers and, with the advent of “GrokBot,” agents acting as employees. This unlocks orders of magnitude more demand, with embodied AI and robotics yet to fully emerge.

Redesigning Infrastructure for the Machine Age

Current infrastructure was not designed with AI’s unique demands in mind, leading to a need for fundamental redesign:

“AI-Designed” Architecture

• Optimization is crucial for inference engines that consume vast memory and generate new tokens alongside compute. This requires rethinking memory, compute, inter-connectivity, power, and cooling at every level.
• The massive capital investment in training frontier models (e.g., $3-5 billion) creates an opportunity for bespoke hardware. Even a 20% efficiency gain justifies building per-model ASICs, a new paradigm where fixed model weights enable highly specialized chips.

Power, Cooling, and Data Centers

• Power Requirements: Rack power is escalating from 5-10 kilowatts to 100-150 kilowatts. This mandates a shift from AC to inherently dangerous DC power.
• Cooling: Air cooling is obsolete for state-of-the-art data centers, which now require liquid cooling, ideally “eco-friendly liquid cooling.”
• Data Center Challenges:
  • Scale: New data centers will need 44 gigawatts of additional power by 2028 (a gigawatt can power 50,000 homes), far exceeding projected grid additions.
  • Construction & Regulation: Building faster is hampered by permits, limited access to power grids, and shortages of critical components like transformers and turbines.
  • Skilled Labor: The transition to high-voltage DC power has created a severe shortage of certified electricians (only 2% in the US are certified), leading to initiatives to train new workforces.
  • Environmental Impact: Future data centers must be designed to be good community citizens, minimizing noise and water waste, and even contributing power back to the grid during peak demand.
  • Robotics: Hyperscalers are actively experimenting with robots for assembling and managing servers within data centers.
• Fund Focus: The Machine Age Fund focuses on “computer science infrastructure” (chips, network interconnect, storage, electricity) and computer science platforms that enable AI’s interaction with the physical world, including embodied AI and robotics.

Investment Opportunities and the “Systems Founder”

The discussion addresses why new companies can still thrive despite the dominance of incumbents like Nvidia:

• Law of Markets: As markets expand, they fragment. Incumbents focus on their “gold bricks” (major existing opportunities), leaving “silver bricks” (emerging, niche but still massive markets) for startups.
• Innovation Gap: Achieving 10x improvements in metrics like “tokens per second per dollar” requires new innovation, which traditionally comes from founders solving problems from first principles.
• Investment Subsectors: Opportunities exist across all layers: chip design (requiring full system builds), memory, networking, power chips, and the crucial software layer for automation and fleet management.
• Capital-Intensive: These ventures require “massive” initial funding rounds (hundreds of millions) before product delivery, reflecting the complexity and scale of hardware development.
• The “Systems Founder”: This new era demands experienced founders—not just researchers, but individuals capable of architecting and designing systems, managing complex supply chains, and considering manufacturing and the entire ecosystem from inception (e.g., Jensen Huang). This explains why “older” founders are increasingly prevalent in this space, as hardware and manufacturing were historically de-emphasized in academia and industry.
• Favorable Climate: Labs are desperate for solutions and engaging with startups early, and there’s ample capital available for follow-on rounds, creating a conducive environment for these new ventures.

The Vision: “Machine Age” and America’s Future

The name “Machine Age” is deliberately chosen to emphasize “Machine Intelligence” over “Artificial Intelligence,” reflecting that current AI learns from vast human knowledge rather than independently forming it. It acknowledges the deep irony that software’s future now hinges on the physical machines below it. The quality of these machines will drive the “next breakthroughs in intelligence.”

The long-term vision (5-10 years) for the fund’s success includes America leading in infrastructure, characterized by abundant, super eco-friendly, efficient data centers, chips, memory, and power. This leadership is seen as vital for maintaining America’s position as a place where profound technological contributions can be made.

• Tags:
• AI Agents
• Venture Capital
• Supply Chain
• AI Models
• AI Infrastructure
• Data Centers
• Founders
• System Architecture
• Liquid Cooling
• Semiconductors
• Power Consumption
• Machine Intelligence
• Machine Age Fund
• Hardware Investment
• Compute Shortage
• GPU Supply