The world is too loud. Read what matters.

The a16z Show

Venture capital is returning to hardware: AI is pulling capital back into the physical world

For thirty years venture capital moved steadily away from hardware; AI has pulled it back. a16z's new $1.1 billion Machine Age Fund is a bet on the layer of physical infrastructure beneath the software stack that has been left fallow for three decades.

AI infrastructureventure capitaldata centershardwarecompute

The video won't play here. Listen to the audio instead:

Medium information density, but it offers a transferable judgment framework: why hardware has become investable again, and how a16z used a standalone fund to solve an organizational-level investment bias.

The argument · tap a timestamp to hear it

2:07

The bottleneck sits beneath the software stack

Jen Kha defines the Machine Age Fund as "everything beneath the software stack": data centers, chips, custom silicon, networking, racks. Her argument is that Mark Andreessen once said software is eating the world, and now AI has solved software — any software need, AI can do it — but to actually solve software you first have to solve physical constraints. This layer of infrastructure has been essentially an uninvestable category for the past thirty years, because it served the previous generation of the internet and SaaS; AI is far more intensive in math and compute, so that "poor man's" infrastructure has to be rebuilt.

— Jen Kha
3:07

A standalone fund is meant to fight investment bias

Why not do this inside the existing infra or apps funds? Jen gives two reasons. First, signal: a separate fund is a commitment to founders that "we are serious about this." Second, portfolio construction at the organizational level — hardware companies need far more money to get started than software, and if you compare them on the same sheet as infra and apps deals, you will almost always lean toward the "sure thing" that already has one or two billion in revenue. Carving out a separate pool of capital is what lets them maximize ownership at the earliest stage and then chase the category with $75 million-class checks.

— Jen Kha
5:12

What LPs want is private-market growth

Jen says a16z raised more than 23% of all venture funding industry-wide this year. She explains the shift in LP mentality: AI's value accrual is happening mainly in private markets rather than public ones, and the public companies in the data center supply chain — SK Hynix, Samsung — have already run up hard, so LPs are "hungry" on the private side. The other change is liquidity: private companies used to stay private longer and longer with no exits, but now multiple trillion-dollar companies will go public, alongside a large set of private companies that have not yet listed. LPs want that exposure and no longer want to put money into asset classes built on the previous technology cycle.

— Jen Kha
8:21

$1.1 billion against $1 trillion plays a different position

Facing the challenge that "US companies will spend $1 trillion building compute this year, and your $1.1 billion is one-thousandth of that," Jen's response is: these are two completely different investments. The Machine Age Fund is an early-stage investor, entering at seed and Series A, putting in $25 million to $35 million for meaningful ownership. She gives two contrasting cases: NextHop is an AI-first high-performance networking company that has reached a Series B inflection point, and a16z invested $65 million through its growth fund; while Unconventional (founded by Naveen Rao) is redoing AI chips from a chip design angle, and a16z entered at seed. The ideal is the latter — getting in before the inflection point.

— Jen Kha
10:28

Hardware companies want more than money

Jen says this has risen to the national level: a month ago the South Korean president visited the US, and his only stop was Silicon Valley. She quotes Mark Andreessen's line that "technology is the dog that caught the bus," and gives a sense of scale — NVIDIA's size is equivalent to the GDP of all G7 countries except the US. Her inference: for the past hundred years the most successful countries were the first to industrialize; in the future the most successful countries will be the first to adopt AI. So a16z's positioning with global LPs is not just capital partnership, but helping them adopt American technology and often co-invest directly in these companies, thereby accelerating local adoption.

— Jen Kha
13:33

The fastest adopters are not the US

Jen lists a string of examples: South Korea announced it would provide premium AI as a public utility to every citizen; El Salvador deployed Grok free in schools and is also using AI doctors; city-states like Singapore and the UAE find it easier to subsidize or provide it free because of their population size. Her judgment is that these countries are accelerating AI development and adoption far faster than the US. The US, meanwhile, faces political headwinds — opposition around data centers and surveillance, some of which she calls "falsehoods." From this she draws a possible outcome: because of this sentiment inside the US, a large share of data center supply chain construction may move overseas.

— Jen Kha
15:35

Data centers are being painted with one brush

Jen concedes the fund would be much easier to run if everyone were data-center friendly, but that is not reality. Her rebuttal: modern data centers — AWS, Meta, and a16z portfolio company Switch — are built by technologists rather than real estate people, and are designed around people's sensitivities. Switch gives power back to the grid rather than taking it, uses very little water, and is one of the few data centers prepared for liquid cooling of future chips. She acknowledges a very small percentage of bad actors, but the vast majority have configured for the new form. The fund is investing precisely in the next generation of data centers.

— Jen Kha
16:39

The real bottleneck is electricians

This passage gives the most concrete mechanism in the whole piece: next-generation chips will be powered by DC rather than AC, and most data centers were not built for that. More critically, fewer than 2% of American electricians are trained in DC power, because DC is extremely dangerous and highly unstable, and mishandling it can be fatal. Jen uses this example to show that for AI to spread, these infrastructure bottlenecks have to be fixed first — and that is where most of the fund's energy goes. She also draws a boundary: the fund does not invest in regulated industries like power (that belongs to the American Dynamism team), only in everything inside the data center that is driven by computer science.

— Jen Kha
18:44

Agents use five times the tokens of humans

The demand-side argument: agents consume five times the tokens of human usage, and the number of agents on the internet has just surpassed the number of humans. Jen believes this will keep growing parabolically, especially once consumers start having personal use cases (she mentions Crock-Bot, Instinct and others). At the same time, AI usage today is still under 5%. The supply side is this fund's thesis: invest in everything driven by computer science — accelerators, CPUs, custom silicon, memory, storage, liquid cooling, robotics inside the data center, systems software.

— Jen Kha
19:46

Hardware founders come from the previous generation

Jen says founders in this category are in some ways "retro": because they need relationships with hyperscalers and customers, and need to genuinely understand physical dynamics, the founders often come from a previous era, frequently spinning out of incumbent companies. NextHop was started by former Arista people. She calls this experience incubated over twenty or thirty years during the last build-out, and now this cohort is leaving to start companies. Diligence is different as a result: on the team, Martin Casado (founder of Nisera, which sold to VMware), Raghu Raghuram (former VMware CEO, who acquired Nisera back then) and Guido Appenizer (former Intel CTO) all come from backgrounds of deep selling into data centers, so the diligence targets are often people and customers they have known for years.

— Jen Kha

In their own words · checked verbatim

we announced this $1.1 billion machine age fund to invest into all of the physical constructs of the world that is now so bottlenecked, given all of the demands in AI.

Jen Kha2:07

Turns out AI has solved software, right? Any software need you have today, AI can actually do it. But we have to solve for all the physical constraints in order for AI to actually solve for software.

Jen Kha3:07

if you do the like for like, you're almost always going to bias towards that sure thing

Jen Kha4:12

technology is the dog that caught the bus

Jen Kha10:28

We think countries of the future are going to be the ones that adopt AI first across fence, across public safety, across healthcare

Jen Kha11:30

there's less than 2% of electricians in the US that are actually trained on DC power because it's very dangerous.

Jen Kha16:39

agents are using five times the amount of tokens as humans are.

Jen Kha18:44

What's old is new again.

Jen Kha22:54

Figures

Machine Age Fund size$1.1 billion2:07
Share of industry-wide venture funding a16z raised this yearmore than 23%5:12
NextHop growth round investment$65 million9:24
NVIDIA market cap$5.5 trillion10:28
Share of American electricians trained in DC powerunder 2%16:39
Agent token consumption relative to humans5x18:44
Share of hardware pitches among all a16z pitchesmore than 20%19:46
AI usage rate todayunder 5%18:44

Glossary

Machine Age Fund
a16z's new $1.1 billion fund investing in the physical infrastructure beneath the AI software stack.
custom silicon
Chips customized for specific AI workloads rather than general-purpose GPUs.
liquid cooling
A data center cooling method that uses liquid rather than air to dissipate heat from high-power chips.
DC powered
Next-generation chips will be powered by DC rather than AC, and most existing data centers were not built for this.
hyperscaler
Cloud providers like AWS and Meta that build their own hyperscale data centers.
American Dynamism
a16z's team investing in regulated national-priority areas such as defense and energy.

How to listen

Who it's for

Investors watching AI infrastructure, hard-tech investing and LP allocation logic; founders who want to understand why venture capital is returning to hardware and where the data center supply chain bottlenecks are.

Skip

The first 0:00-1:04 is a show trailer and can be skipped.