Andreessen Horowitz raised $1.1 billion to bet AI's bottleneck moved from talent to power substations and memory chips

Andreessen Horowitz raised $1.1 billion for a new vehicle dedicated entirely to the physical hardware underneath artificial intelligence -- chips, memory, networking gear, data-center systems, robotics and AI-enabled home appliances -- the firm announced August 28. Five general partners put their names on the launch: Ben Horowitz, Martin Casado, Raghu Raghuram, David Ulevitch and David George. a16z frames the goal as accelerating what it calls a "social and national imperative": the physical buildout of AI, and says the firm is "opening the throttle" on hardware after a decade spent mostly financing software.

The case a16z makes for why now, in its own numbers: compute density has jumped 28 times from an H100 rack to a Rubin rack, and power draw per rack has climbed from 5-10 kilowatts to 100-250 kilowatts today -- a figure the firm projects could reach 1 megawatt within three years. Data centers themselves are scaling from tens or hundreds of megawatts to gigawatt-scale campuses, and networking inside the rack is, in the firm's words, "hitting the limits of copper cabling." None of that is independently verified here -- it's a16z's own scoping of the constraint it's betting on, though it tracks with power and density figures reported elsewhere across the data-center buildout this year.

It's the rack-power figures, more than the fund size itself, that a16z leans on to justify a dedicated vehicle rather than folding hardware bets into its existing funds.

a16z points to a roster of existing bets as evidence the thesis already has traction -- Unconventional AI, Nexthop, Volta, Atoms, Heron Power and Mind Robotics among them -- though the firm's announcement does not say which, if any, of those positions were funded out of the new $1.1 billion vehicle specifically versus its broader portfolio. Team members named alongside the five GPs include Guido Appenzeller, a former Intel Data Center Group CTO, and Shangda Xu. The firm says hardware deals now account for more than 20% of its overall deal flow, up from a share it describes as historically negligible.

The scarce thing has changed. For most of the past decade, it was talent and distribution. a16z is now betting $1.1 billion that it's transformers, substations, and thermal design.

What the Machine Age Fund isn't is a16z's first move into infrastructure this year. On January 9, the firm disclosed over $15 billion raised across six vehicles -- its largest fundraise ever, equal to roughly 18% of all US venture capital allocated in 2025 -- including a $1.7 billion Infrastructure Fund 2 and a $1.176 billion American Dynamism Fund 2, both already adjacent to physical AI buildout and defense hardware. (American Dynamism is a16z's standing label for its bets on manufacturing, defense and "hard tech" more broadly -- it predates this specific hardware-vs-AI framing by several years.) The Machine Age Fund is a narrower, later, hardware-specific carve-out layered on top of that January raise, not a replacement for it.

That context cuts against reading the Machine Age Fund as a wholesale pivot. The same January raise included a $6.75 billion growth fund and a $3 billion "other venture strategies" bucket -- both several times the size of the new hardware vehicle, and neither restricted to physical infrastructure. $1.1 billion is real money and a real, named strategic bet, but it's a small slice of a firm now managing over $90 billion, deployed by partners who have also spent the past eight months raising for cloud software, defense manufacturing, biotech and straightforward growth equity. The framing -- "the scarce thing has changed" -- describes where a16z wants attention pointed more than it describes where most of the firm's capital is actually going.

The bet also arrives alongside far larger sums already moving through the AI buildout from other directions -- Nvidia's own financing alliance with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR is structured to mobilize over $500 billion in third-party capital, roughly 450 times the size of a16z's new fund. Venture capital of this kind isn't trying to compete on scale with hyperscaler balance sheets or bank credit lines; it's trying to pick the winners inside the supply chain -- the memory, cooling and power-management startups -- before that larger money decides which ones matter.

Whether the bet pays off depends on questions the announcement doesn't answer: which specific companies the $1.1 billion actually reaches, how much of it competes for the same handful of promising power-and-memory startups every other infrastructure-focused fund is also chasing, and whether a16z's hardware deal flow keeps climbing past 20% once the rush of Hot Chips-season announcements settles. The firm's own history offers one more variable no press release addresses: general partnerships built for evaluating software cap tables don't automatically transfer to hardware's longer development cycles, capital intensity, and physical supply chains.

That gap is exactly why the fund's own scoping matters more than the round number. "Chips, memory, networking, storage, data centers, robotics and AI home appliances" spans businesses with wildly different risk profiles inside one $1.1 billion vehicle -- a fabless chip startup burns cash for years before first revenue and competes directly with companies backed by Nvidia, Intel and the hyperscalers' own custom-silicon programs; a home-appliance company selling an AI-enabled device is a consumer hardware bet with retail margins and a completely different failure mode. a16z hasn't said how the fund allocates across that spread, only that all of it now qualifies as "physical AI" under one thesis.