A16z announces $1.1B Machine Age Fund for AI infrastructure

Andreessen Horowitz just put $1.1 billion behind the idea that the future of AI won’t be written in code alone. It’ll be forged in silicon, copper, and concrete.

The venture firm’s new Machine Age Fund, which closed on August 28, is a16z’s first vehicle exclusively targeting AI hardware and physical infrastructure. We’re talking chips, memory, networking, storage, interconnects, data centers, and robotics. The full stack of atoms that make the bits possible.

Why hardware, why now

Consider the compute density trajectory. From NVIDIA’s H100 to its Rubin architecture, compute density in AI racks has jumped by 28 times. That’s not an incremental improvement. That’s a fundamentally different engineering problem at every layer of the stack.

Power requirements tell a similar story. Standard data center racks historically drew 5 to 10 kilowatts. Current AI-optimized racks already pull 100 to 250 kW. Within three years, projections suggest individual racks could hit 1 megawatt.

A16z’s pivot from pure software

The Machine Age Fund sits alongside, not within, a16z’s existing fund family. The firm has raised over $15 billion across its various vehicles, including a $1.7 billion Infrastructure Fund 2 that covered a broader set of investments. This new fund is purpose-built and narrowly focused.

Hardware investments already account for more than 20% of a16z’s recent deal flow. Prior hardware-adjacent investments include positions in Unconventional AI, Mind Robotics, and SpaceX.

The original source material hints at the competitive dynamics at play: incumbent companies like Nvidia and CoreWeave are reportedly willing to leave margin on the table to support innovation from new founders approaching these problems from different angles.

What the fund will actually back

The investment mandate spans the full physical infrastructure stack. Chip design is the obvious starting point, but the fund’s scope extends well beyond processors.

Memory architecture is one area where bottlenecks are becoming acute. Modern AI workloads are increasingly memory-bound rather than compute-bound, meaning the speed at which data can be fed to processors matters as much as the processors themselves.

Networking and interconnects represent another critical layer. Training large models requires thousands of GPUs to communicate with each other at extraordinary speeds.

Storage infrastructure, data center design, and cooling technologies round out the hardware side. Robotics, included in the fund’s mandate, reflects a16z’s view that physical AI systems will eventually need to interact with the real world, not just process data about it.