Andreessen Horowitz (a16z) has closed a $1.1bn investment vehicle dedicated exclusively to hardware, a notable departure for a firm historically associated with software and internet companies. The Machine Age Fund, announced on Friday, targets what the venture capital firm calls the "physical layer" of artificial intelligence — the tangible infrastructure that AI systems rely on, from silicon chips to the power-hungry data centers that train and run them.

The fund's scope is deliberately broad. Its mandate covers chips, memory, networking, and storage, and extends to complete systems including data centers, robotics, and AI appliances for the home. This reflects a growing recognition within the investment community that the next phase of the AI revolution will be constrained less by software breakthroughs and more by physical realities such as energy consumption, heat dissipation, and supply chain capacity.

Five a16z partners have put their names to the launch: Ben Horowitz, Martin Casado, Raghu Raghuram, David Ulevitch, and David George. The involvement of five senior partners in a single vehicle underscores how central the firm considers this thesis to its future. It also signals that a16z is willing to commit substantial human capital to a sector that many venture firms have historically avoided due to its long development cycles and capital-intensive nature.

The physics of AI infrastructure

The investment case for the Machine Age Fund rests not on conventional market sizing but on the physical constraints of computing. The firm points to a 28-fold increase in compute density between Nvidia's H100 generation and its upcoming Rubin racks. A single server rack that once drew between 5 and 10 kilowatts of power now draws between 100 and 250 kilowatts. Within three years, the firm expects that figure to reach one megawatt per rack.

These numbers have profound implications for data center design and energy infrastructure. Individual data centers are already shifting from tens of megawatts of power capacity to hundreds, with some campuses approaching gigawatt scale. This trend is forcing builders and operators to seek financing on an entirely different order of magnitude, and it is putting unprecedented pressure on electrical grids, cooling systems, and real estate markets.

According to a16z, every layer of the AI stack is running into the limits of what the supply chain can produce, and in some cases into the limits of physics and computer science. The bottleneck, on this reading, has moved out of the models and into the world. Model architecture and algorithmic efficiency remain important, but the ability to deploy AI at scale now depends on access to transformers, substations, advanced cooling systems, and the materials needed to build them.

A shifting deal flow

a16z's own deal flow reflects this shift. Hardware has gone from a marginal share of the deals the firm sees to more than 20% of its total pipeline. For two decades, hardware has been considered a harder sell in venture capital because it takes longer to develop, requires more capital, and scales less efficiently than software. But the firm suggests that when a fifth of the pipeline consists of hardware opportunities, that traditional objection stops being decisive.

The machine age portfolio that the fund builds on spans a great deal more than semiconductors. Unconventional AI, Nexthop, Volta, Atoms, and Mind Robotics sit alongside Skydio, SpaceX, Anduril, and Waymo. That list illustrates how broadly "systems" is being interpreted here. Drones, launch vehicles, defense hardware, and autonomous vehicles all qualify, placing the fund some distance from a conventional deeptech mandate focused solely on foundational research.

Token intensity and demand growth

On the demand side, the argument turns on token intensity. Both the volume of AI work and the compute required for each unit of that work are rising by orders of magnitude. a16z describes this growth in triple digits, suggesting that even efficiency improvements in algorithms are unlikely to offset the sheer scale of compute demand in the near term.

The specific areas of interest outlined by the fund are narrower and more revealing. a16z says it is particularly focused on memory and interconnect improvements, power-efficient edge devices, and the cooling, materials, electrical, and real estate infrastructure that surrounds a modern AI facility. These are areas where the constraints on AI growth have quietly settled, often away from the spotlight that follows chip announcements.

Memory and interconnect are illustrative. A rack full of accelerators that cannot be fed data fast enough is simply an expensive way to generate heat. The industry has spent the past two years discovering how often this is the actual limit on performance, rather than raw compute capacity. Similarly, power-efficient edge devices are critical for bringing AI inference closer to the data source, reducing the need for massive centralized compute.

The European dimension

The last category — real estate, power distribution, and cooling — is particularly worth watching in Europe. These have not historically been venture capital categories, but they are becoming binding constraints on AI deployment. Siting is already a critical issue, with 63% of new data center capacity now going somewhere other than the five established hubs. This dispersal creates opportunities for new players in regions with available power and land, as well as for technologies that make data centers more energy-efficient and easier to locate.

Europe's energy transition adds another layer of complexity. AI data centers are increasingly competing with households and industry for electricity, and the push for renewable energy sources is creating both challenges and opportunities. Investors who understand the interplay between AI infrastructure and energy policy may find particularly attractive opportunities in this space.

Context within a16z's broader strategy

The Machine Age Fund arrives on top of an already unusual year for a16z. The firm announced more than $15bn across new funds in January, including a $1.7bn Infrastructure Fund 2 and a $1.18bn American Dynamism Fund 2. The relationship between these vehicles and the Machine Age Fund has not been made clear, and there is no word yet on limited partners, cheque sizes, or stage focus.

Recent cheques point in the same direction regardless. Notable among them is a Series A investment in Netris, a company that automates the networking layer that often slows down GPU clouds. This focus on removing bottlenecks in AI infrastructure is consistent with the fund's broader thesis, even if the specific parameters of the fund remain opaque.

The pitch, stripped of the language around it, is that the scarce thing has changed. For most of the past decade, the scarce resources in technology were talent and distribution. a16z is now betting $1.1bn that the scarce resources are transformers, substations, and thermal design. It is a bet on the physical world after a generation of software dominance, and it may reshape how venture capital thinks about infrastructure.

Source: TNW | Investors-funding News