a16z's $1.1B Machine Age fund bets on AI's physical layer

Andreessen Horowitz announced a $1.1 billion Machine Age fund on Friday, aimed squarely at the physical layer AI runs on — chips, memory, interconnects, cooling, power-efficient edge devices, and the real estate around all of it. From the firm whose entire brand was "software is eating the world," that is a loud signal about where the constraint has moved.

What actually happened

Per TechCrunch, the fund targets "faster, more efficient systems," "cheaper and higher-bandwidth memory across the memory hierarchy," faster interconnects, power-efficient edge devices, and the cooling, materials, electrical, and real estate buildout underneath. TNW reports five senior partners signed the launch — Ben Horowitz, Martin Casado, Raghu Raghuram, David Ulevitch, and David George — and that hardware now accounts for more than 20% of the firm's deal flow. It sits alongside the $15B-plus a16z announced across new funds in January 2026, including a $1.7B Infrastructure Fund 2 and a $1.18B American Dynamism Fund 2.

Stated goal, in the firm's words: "open the throttle and accelerate the physical buildout of AI."

Why it matters for your business

You are not raising a fund. But you are downstream of one, and this is a useful read on the next 24 months of your vendor bills.

The signal is that the bottleneck for AI is no longer clever software — it is memory bandwidth, interconnect, power, and floor space. Those are capital-intensive, slow, and permit-bound. That means the current pricing you're planning around is a snapshot of a supply crunch, not a stable floor, and the vendors most exposed to it are the ones reselling somebody else's inference capacity at a fixed monthly price.

Two practical moves. First, do not architect around one inference provider's current price. Keep your prompts, your evals, and your business logic in your own repo, with the model behind an interface you can swap. When capacity loosens — or tightens — you want a config change, not a rewrite. Second, watch the funded-vendor lifecycle. A hardware-heavy VC cycle produces a lot of well-capitalized startups selling into small businesses at a subsidized price, and a lot of them will consolidate. Anything holding your data should have an export path you have actually tested, not a documented one you have read about.

The money moving to atoms is a bet that AI's cost curve gets fixed in the data center, not in the API. Build like the API price is a variable.

Sources: TechCrunch, The Next Web.