admin The fund, called the Machine Age Fund, will invest across processors, memory, networking, storage and other technologies required to sustain the rapidly increasing computational demands of AI. It will also target complete systems, including data centres, robots and AI-enabled devices designed for homes and workplaces.
The initiative reflects a growing conviction among technology investors that constraints on AI development are shifting from software towards physical infrastructure. More capable reasoning, coding and autonomous systems consume substantially more computing resources, creating pressure on semiconductor supply, electricity networks, cooling systems and data-centre capacity.
Andreessen Horowitz said the objective was to accelerate the physical buildout required to support AI. The firm believes existing hardware supply chains, typically accustomed to annual growth of roughly 20% to 30%, may have to accommodate demand expanding at triple-digit rates in some segments.
Power requirements illustrate the scale of the challenge. AI server racks that once required about 5 to 10 kilowatts are moving towards systems consuming between 100 and 250 kilowatts. The firm expects rack-level requirements to approach one megawatt within three years as processors become denser and AI models demand greater computational performance.
Compute density has also increased sharply. Andreessen Horowitz estimates that density per rack has risen about 28-fold between systems based on Nvidia’s H100 accelerators and those designed around its newer Rubin architecture. Networking capacity must increase alongside processing power, pushing traditional copper connections towards their technical limits and encouraging investment in faster interconnect technologies.
Data centres themselves are becoming dramatically larger. Facilities traditionally measured in tens of megawatts are being supplemented by campuses requiring hundreds of megawatts, while some planned AI developments are approaching gigawatt scale. Operators are increasingly examining dedicated power generation alongside electricity supplied through conventional grids.
The Machine Age Fund will consequently look beyond processors. Opportunities identified by the firm include high-bandwidth memory, networking components, efficient storage, cooling technologies, electrical infrastructure, advanced materials and specialised real estate.
Robotics represents another important part of the strategy. AI systems are increasingly moving from text and digital applications into machines capable of interacting with physical environments. Investment has accelerated in humanoid robots, autonomous industrial equipment and general-purpose robotic systems as companies seek to combine increasingly capable AI models with machines that can perform real-world tasks.
Andreessen Horowitz has already backed companies including Mind Robotics, Unconventional AI, Nexthop, Volta, Atoms and Heron Power. Its longer record of hardware-related investments includes autonomous drone manufacturer Skydio, aerospace company SpaceX and defence technology company Anduril, while the firm also participated in Waymo’s 2020 financing.
The dedicated fund formalises a shift that has been building across the firm’s investment pipeline. Hardware companies have grown from representing a relatively small proportion of opportunities examined by Andreessen Horowitz to more than 20% of its deal flow during the past two years.
The strategy is being led by executives with backgrounds spanning data-centre technology, networking and semiconductor infrastructure. Managing partner Raghu Raghuram previously led VMware, while general partner Martin Casado has spent much of his career working on networking and data-centre technologies. Other members of the team bring experience from Intel, infrastructure investing and manufacturing.
The fund arrives as venture capital increasingly follows the enormous capital spending required by the AI industry. Technology groups and specialised infrastructure providers are committing tens of billions of dollars to processors, server farms, power supplies and networking equipment as competition intensifies to train and operate increasingly sophisticated models.
Hardware investment carries risks that differ substantially from software. Semiconductor companies and robotics manufacturers generally require greater upfront capital, longer product-development cycles and complex manufacturing relationships. Supply chains can also be exposed to geopolitical restrictions, shortages of specialised components and fluctuations in energy availability.
Those obstacles are simultaneously creating opportunities for start-ups attempting to redesign parts of the computing stack. Greater power density requires new cooling approaches, expanding AI clusters need faster networking, and pressure on memory bandwidth is encouraging alternatives to established architectures.
