Andreessen Horowitz Raises $1.1 Billion Machine Age Fund as AI Money Shifts From Software to Silicon

Fifteen years after Marc Andreessen declared that software was eating the world, his firm has decided the world needs a lot more machines to eat it with.

Andreessen Horowitz said on Aug. 28 that it has closed $1.1 billion for a new vehicle called the Machine Age Fund, dedicated entirely to the physical layer of artificial intelligence — chips, memory, networking, storage, data centers, robots, and what the firm calls home AI appliances. For a partnership whose brand was built on the infinite margins of code, it is the most explicit hardware pivot in its 17-year history.

“It’s time to open the throttle and accelerate the physical buildout of AI: the strongest tool ever developed for solving problems and bestowing abundance,” five general partners — Ben Horowitz, Martin Casado, Raghu Raghuram, David Ulevitch and David George — wrote in the announcement post. “It is our social and national imperative.”

The thesis is scarcity. Every rung of the AI hardware ladder, the firm argues, is now pinned against a supply ceiling that the industry was never built to raise quickly. Compute density has jumped 28-fold from an Nvidia H100 rack to a Rubin rack. Rack power has climbed from roughly 5 to 10 kilowatts a few years ago to 100 to 250 kilowatts today, and a16z expects it to reach one megawatt within three years. Data center campuses have moved from tens of megawatts to hundreds, and in some cases to gigawatt scale. Networking inside a single rack has run up against the physical limits of copper cabling.

“The hardware industry supply side is used to growing 20% to 30% per year at most; not the triple-digit growth that’s needed to catch up with demand,” the partners wrote. “This will change, quickly.”

A software firm rediscovers atoms

a16z is not arriving cold. The firm led Skydio’s Series A in 2016, invested in SpaceX, wrote its first check into Anduril in 2019 and was among the earliest venture backers of Waymo’s 2020 round. More recently it has funded Unconventional AI, networking startup Nexthop, data center builder Volta, Atoms and Mind Robotics, plus Heron Power, which is building solid-state transformers that swap the traditional oil-immersed metal coil for silicon carbide power chips and a built-in battery for outage ride-through.

What changed is the deal flow. Hardware pitches went from a rounding error to more than 20% of what the firm sees, the partners said — a shift that made a dedicated fund, and a dedicated team, defensible to limited partners.

That team leans on operators rather than pure financiers. Guido Appenzeller was chief technology officer of Intel’s Data Center Group. Casado and Raghuram spent decades in data center systems software at VMware, work that required deep hardware partnership. Shangda Xu and David George have invested across silicon, networking and large-scale compute platforms, while Ulevitch and Erin Price-Wright drive hardware and U.S. manufacturing bets through the firm’s American Dynamism practice.

“Hardware is in our team DNA, and it’s great to get back to our roots,” the partners wrote, framing the moment as a platform reset on the scale of the shift from mainframe to client-server, or from on-premise to cloud — but faster, and broader. The opportunity, they argue, is “a once-in-a-generation opportunity to rearchitect them as platforms, all the way down to the electricity.”

Analysis

The Machine Age Fund is a small number attached to a large signal. At $1.1 billion, it is a fraction of what hyperscalers spend on capital expenditure in a single quarter, and it cannot meaningfully finance a gigawatt campus. What it can do is fund the component layer — the memory architectures, interconnects, power electronics and cooling systems that the hyperscalers will need to buy from someone.

That is the real bet: that the AI value chain is bifurcating. The model layer is consolidating into a handful of extraordinarily capitalized labs whose valuations already price in near-perfect execution. The application layer is crowded and increasingly commoditized by the models themselves. The infrastructure layer, by contrast, is capacity-constrained, capital-intensive and structurally hard to copy — exactly the conditions under which venture returns tend to be durable rather than fashionable.

It also reflects a broader repricing of what counts as venture-scale. Hardware was long considered uninvestable by classic VC math: long development cycles, thin gross margins, punishing capital requirements. Demand certainty changes that calculus. When customers are queuing years in advance for transformers, high-bandwidth memory and power distribution gear, the traditional hardware risk profile starts to look more like a supply problem than a market-adoption problem.

The competitive context is telling. Kleiner Perkins raised $3.5 billion across two funds in March, and Thrive Capital pulled in $10 billion in February. a16z is not out-raising anyone here. It is out-positioning them, staking a claim to a category most software-native firms are still not equipped to underwrite — and doing it under an explicitly nationalist framing that ties the physical AI buildout to American industrial capacity.

The risk is timing. If the AI buildout decelerates, or if hyperscalers vertically integrate the very components a16z is funding, portfolio companies could find themselves with beautiful silicon and no buyer.

What to watch

Three things will show whether this fund is prescient or premature. First, deployment pace: a16z says hardware is now more than 20% of its deal flow, so the first several Machine Age checks should land within months, and their targets will reveal whether the firm is chasing power and cooling or chips and memory. Second, whether rivals follow with dedicated hardware vehicles — one fund is a hypothesis, five is a category. Third, the physics itself: if rack power really does approach one megawatt on a three-year horizon, the grid, not the fab, becomes the binding constraint, and the most valuable company in this portfolio may end up being an electrical one.