a16z Launches $1.1 Billion Machine Age Fund Focused on Physical Layer for AI Operations
a16z announced the completion of a new $1.1 billion Machine Age Fund, dedicated to investing in the physical layer required for artificial intelligence operations. The official statement describes it as "opening the throttle and accelerating the physical infrastructure for AI," framing the progress as a social and national mission. Notable partners include Ben Horowitz, Martin Casado, Raghu Raghuram, David Ulevitch, and David George. Raghuram told Bloomberg that this is new capital outside of the approximately $15 billion series from the new year, not a transfer from the original fund.
The investment scope covers chips, memory, networking and storage, as well as data centers, robotics, and home AI appliances. Partners highlighted the demand for faster and more energy-efficient systems, cheaper and higher bandwidth memory across all levels, cabinet interconnects that break the limits of copper cabling, low-power edge devices, and support for cooling, materials, power, and real estate. The article outlines physical constraints: cabinet computing density increasing 28 times from H100 to Rubin, cabinet power consumption rising from about 5 to 10 kilowatts to 100 to 250 kilowatts, aiming for 1 megawatt within three years; data centers moving from tens of megawatts to hundreds of megawatts and even gigawatt-level parks, with more power coming from behind-the-meter or dedicated sources.
Hardware has accounted for over 20% of the firm's recent deal flow. Recent funding recipients include Unconventional AI, Nexthop, Volta, Atoms, Heron Power, and Mind Robotics; earlier investments included Skydio, SpaceX, Anduril, and Waymo. Ulevitch and Erin Price-Wright are working on U.S. manufacturing through the American Dynamism line. The team background is noted as coming from Intel's data center division and data center systems software. In January, there was already about $1.7 billion in phase two infrastructure and about $1.12 to $1.18 billion in phase two U.S. power, but the division of this fund from the two has not been detailed. Limited partners, check sizes, and stages have not been disclosed.
Jensen Huang's concurrent financial report indicated that demand growth could exceed 70% in fiscal year 2028, but is constrained by supply. Buyers want to expand models from dialogue to reasoning and knowledge work, but are hitting the limits of cabinets and power grids in labs and cloud providers; sellers are hardware startups that can change memory hierarchies, interconnects, and cooling. Funding is shifting from software multiple valuations to watts, copper, and real estate. Beneficiaries are component suppliers and park developers outside of NVIDIA's supply curve; those under pressure are data center owners who still price according to software rhythms but must digest megawatt-level cabinets. The event is driven by the closure of the new fund, not by the IPO of any hardware company.
Source: Public Information
ABAB AI Insight
Andreesen Horowitz formalizes hardware as a business line, acknowledging that the software scaling law has hit the physical limits of cabinets. The 28-fold density increase from H100 to Rubin is not just product launch rhetoric; it is a diagnosis of simultaneous failures in copper cabling, heat, and electricity. The $1.1 billion relative to the total $15 billion in January is not large, but its function is to separately account for the hardware that already occupies 20% of the deal flow, avoiding the holding period and governance terms of the software fund. Casado and Raghuram's data center backgrounds are highlighted in the signature line, signaling to limited partners that this is not a cultural transformation, but a change of track for the original team.
The capital path integrates manufacturing and defense hardware from "U.S. Power" with cloud-side chip memory interconnects into a single procurement list. SpaceX, Anduril, and Waymo provide complete narratives, while Unconventional AI and Nexthop supply components within cabinets. Funds are directed towards changing memory hierarchies and behind-the-meter power sources, as Huang has already framed supply shortages as growth limits. Home appliances are included in the scope, indicating that the edge side is also considered part of the infrastructure, preventing the fund from only investing in gigawatt parks. The lack of disclosed stages means the same fund can write checks for seed chips and growth-stage cooling facilities.
Benchmarking against Sequoia and Lightspeed's standalone infrastructure funds in the cloud era, and SoftBank's Vision Fund's aggressive semiconductor investments, as well as Intel Capital's long-term ecosystem binding. The current phase is expansion: model companies are still burning cash on reasoning, while venture capital is pre-purchasing the next bottleneck. Control remains with TSMC and NVIDIA's capacity calendar, with risk funds buying alternative parts in the gaps of the calendar.
Structurally, this represents a reconstruction of the supply chain. Pricing power shifts from model APIs to cabinet kilowatts and memory bandwidth. The mechanism is: after software marginal costs approach zero, incremental profits are locked within irreplicable physical constraints; whoever first prices these constraints will collect tolls in the next round of capital expenditures. The Machine Age Fund is a pre-sale ticket for tolls.
ABAB News · Cognitive Laws
After software hits a wall, the fund will pivot to hardware
Cabinet power consumption will be the next pricing power
National missions are easiest to write as fundraising memos