Andreessen Horowitz said on 28 August 2026 that it had raised what it calls the Machine Age Fund, a 1.1 billion dollar pool dedicated to the physical layer of artificial intelligence. In the firm’s own words, the fund “will invest into all of the computer infrastructure on which AI runs, including chips, memory, networking, and storage,” and “full systems for running AI: from data centers to robotics to home AI appliances.” The announcement is signed by Ben Horowitz, Martin Casado, Raghu Raghuram, David Ulevitch and David George.

The figure is disclosed by the fund’s manager, which is the party in a position to know it, and we report it as such rather than as an audited or independently verified number. What makes the move notable is less the size than the direction. A firm whose founding argument was that software would eat the world is now committing a dedicated fund to the hardware that software has started to choke on.

This slots directly into a thread we have been tracking: capital moving out of software subscriptions and into the physical constraints that now govern how much AI can actually be run. It sits alongside Nvidia’s reported server price rises, where memory rather than the GPU is the cost that moves the bill.

Why it matters

When the largest software-native venture firm raises a fund specifically for chips, memory and data centres, it is pricing a bottleneck. The bet only makes sense if the returns are in the constraint itself, not in the applications sitting on top of it, and that is a statement about where margin is expected to accrue over the next few years.

For anyone buying or building on AI, the read-through is about cost and availability, not venture returns. If the people closest to the deal flow are putting a billion dollars behind the premise that hardware cannot scale fast enough, then the assumption that inference gets cheaper and more abundant every quarter is exactly the assumption under pressure.

What a16z committed, and how it framed it

The hard facts come straight from a16z’s own page, which is a primary source and was retrievable to us. The fund is 1.1 billion dollars, announced 28 August 2026, and its scope spans chips, memory, networking, storage, data centres, robotics and home AI appliances. The firm describes the fund as making hardware “an official motion” for a16z; it does not, on the page we read, claim this is its first hardware fund, and we do not assert that.

The reasoning a16z gives is a supply-and-demand mismatch stated in plain numbers. “The hardware industry supply side is used to growing 20% to 30% per year at most,” the firm writes, “not the triple-digit growth that’s needed to catch up with demand.” It describes every layer of the stack as “all hitting the wall of today’s supply chain capability, and the limits of physics and computer science.”

Follow-on reporting adds that hardware has grown from a sliver of a16z’s deal flow to more than a fifth of it; that figure comes from named secondary coverage rather than the announcement page itself, so we mark it as reported. It is not load-bearing for the story, which stands on the fund size and the firm’s own stated thesis.

The bet is that physics, not code, is now the binding constraint

The Machine Age Fund is a wager that the scarce, valuable thing in AI has moved down the stack. For a decade the a16z reflex was to fund the application and let someone else worry about the metal. The premise now is the reverse: that the metal is the constraint, and whoever supplies more of it, faster, captures the value.

That premise has company. Nvidia’s own most recent quarter, where the firm sold 96.2 billion dollars of chips in three months while purchase commitments grew faster than revenue, is the same story told from the supply side. So is the wider shift in August’s capital flows, where data centres, chip supply chains and power absorbed the largest cheques while software rounds shrank. A dedicated hardware fund from a16z is the venture-side confirmation of a pattern the operating companies have been signalling for months.

The risk in the thesis is timing rather than direction. Supply bottlenecks are precisely the thing capital rushes to relieve, and a fund raised at the peak of a constraint can find that constraint easing by the time its money is deployed. Memory, power and packaging capacity all expand on multi-year lead times, but they do expand. The bet pays only if demand keeps outrunning that expansion for long enough.

What this means for buyers

Do not budget as though inference prices only fall. The dominant experience of the past year has been rapid price cuts, but a billion-dollar fund built on the opposite premise is a signal that the input costs underneath those prices are under strain. Plan for the possibility that the cheapest tier you rely on today gets repriced upward as memory and power costs feed through, and keep a note of what you would do if it did.

For teams making multi-year commitments, weight supply security alongside price. A provider’s access to compute, its power contracts and its position in the memory supply chain are now legitimate procurement questions, not background trivia. Our company signals tracking exists partly to make those movements legible before they show up on an invoice.

For most buyers, though, the practical action is modest: treat this as confirmation to diversify suppliers and avoid designing a workflow that only survives at today’s floor price. The fund is a forecast, not an invoice, and nothing changes for you this quarter because of it.

What would change our reading

The load-bearing facts here are first-party and were verified against a16z’s own page, so the main risk is not that the announcement is wrong but that the thesis is. If AI inference prices keep falling steeply through the next few quarters with no supply-driven reversal, the fund’s premise looks early or mistaken, and we would read it as a contrarian position rather than a confirmed trend.

We would sharpen the story in the other direction if a16z discloses specific portfolio commitments, a Form D confirming the raise, or co-investors, which would move parts of this from first-party assertion to independently checkable record. And if hardware-focused funds from other large software investors follow in the coming weeks, the single data point becomes a category shift, and we would say so.

Sources

  • Andreessen Horowitz, “The Machine Age Fund” (fund announcement, 28 August 2026) — a16z.com
  • TechCrunch, “a16z creates a $1.1B ‘Machine Age’ fund to ‘accelerate the physical buildout of AI'” — techcrunch.com
  • Bloomberg, “Andreessen Horowitz Raises $1.1 Billion for AI Infrastructure Fund” — bloomberg.com