Jen Kha, the a16z managing partner who runs the firm's fundraising and its relationships with its limited partners, explains why the firm put $1.1 billion into a fund for chips, networking, memory, cooling and data centers — the layer venture capital spent three decades walking away from. Theo Jaffee and Sophia Dew press her on what a fund that size can do against a build-out a thousand times larger, on the public backlash against data centers, and on why governments now treat AI adoption as a national priority.
👤 Guest: Jen Kha, managing partner at a16z
🎙️ Hosts: Theo Jaffee and Sophia Dew
📰 Published: 30 August 2026
🔴 YouTube | 🟢 Spotify | 🟣 Apple Podcasts | 🔗 Episode page | ⏱️ 25 min | ✅ Time saved: 11 min
Key Takeaways
AI solved software, and the constraint moved into the physical world
"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."
The physical layer was uninvestable for a generation and is now being rebuilt
"largely an uninvestable category for the most part for the last 30 years" — built for the internet era, then for SaaS, and not for AI's compute load
A hardware company loses every like-for-like comparison inside a generalist fund
Against an infra or apps deal already doing "a billion, two billion in revenue", the generalist bias goes to the sure thing
$1.1 billion is small against the build-out because the fund buys ownership, not capacity
Seed and Series A cheques of "$25, $30, $35 million" for substantive ownership, before the inflection
The value in AI accrued privately, and LPs are repositioning for the unwind
"with SpaceX going public, Anthropic coming shortly, OpenAI just on the tails"
Nvidia is now the size of a country, which is why AI adoption is a head-of-state issue
"Nvidia is larger than all of the G7 countries except the U.S."
Korea, El Salvador and the city-states are adopting AI faster than the United States
Premium AI for every citizen in Korea; "They implemented Grok in their schools, for example, for free." in El Salvador
The US data center backlash is a narrative problem, and the firm treats it as an opportunity
"the narrative has gotten in the way of reality" — and the supply chain build may move overseas because of it
The next bottleneck is electrical and it is a labor problem
"there's less than 2% of electricians in the U.S. that are actually trained on DC power because it's very dangerous"
Agents are the demand curve
"agents are using five times the amount of tokens as humans are"
Hardware went from almost nothing to more than 20% of the pitches a16z sees
"we went from seeing basically nothing in terms of hardware pitches to over now 20% of our pitches are hardware"
The founders are throwbacks, and so is the team diligencing them
Former Arista, Nicira, VMware and Intel operators, on both sides of the table
Venture Capital Spent Thirty Years Walking Away From Hardware; a16z Just Put $1.1 Billion Back In
Dew opened by asking why this fund and why now. Kha started with the calendar: "the classic adage is sell in May and go away", and the firm was announcing a fund in the last week of August. "This has been the most prolific summer and all the fact that we're announcing a fund on August 28th when typically Wall Street is dead is like a classic sign of where we are in the cycle", she said.
The mandate is everything below the software stack. The fund invests in "all of the physical constructs of the world that is now so bottlenecked given all of the demands and AI" — data centers, chips, custom silicon, networking and racksIt sits underneath a16z's existing vehicles: the Infra Fund, which invests in products for developers, and the Apps Fund, which sells into business-to-business and business-to-consumer
The category was written off for a generation. It was "largely an uninvestable category for the most part for the last 30 years", because that infrastructure had already been built out for the internet era and then for SaaS
All of it is now being rebuilt, she said, because AI is far more mathematically and compute intensive than what came beforeWhat the industry is running on in the meantime is "the poor man's version that we're limping along with today"
Why a Separate Vehicle: the Signal to Founders and the Portfolio Math
Jaffee asked why this needed its own vehicle rather than being done through the main growth and infra funds. Kha gave three answers — the firm's read on where founders are going, the signal a dedicated fund sends, and portfolio construction.
The job is to follow the entrepreneurs. "My partner, Chris Dixon, calls it following the nerd energy.""what is the nerd doing on nights and weekends is probably what us normies will be doing in the future"
The thesis in one line: "Marc had famously said software's eating the world. Turns out AI has solved software""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."
A separate fund is a signal to the founders building in the category to come and spend time with a16z first
Judged like for like inside a generalist portfolio, a hardware company loses. These companies raise a lot of capital just to get off the ground, and set against an infra or apps deal, "you're almost always going to bias towards that sure thing with infra, what it's doing, a billion, two billion in revenue"
Separating it stakes out the firm's commitment to the space and, in portfolio-construction terms, is built to capture the winners at the earliest stages and maximize ownership in them
What Changed on the LP Side: the Value Accrued Privately and the Supply Stack Went Vertical
Dew asked about Kha's own role — she leads the partner relationships and the capital-raising strategy — and whether conversations with institutional investors have changed. Kha said the shift comes from three places.
Most of the value creation in AI has happened out of public view: "if you look at most of the value accrual in AI, that's largely been on the private side, not the public side", so allocators are moving toward private capital because that is where the growth has been
The public part of the data center supply chain has been repriced, and its executives have become celebrities. "those stocks have been ripping", she said, and on an internal all-hands the firm joked that "Sanjay, the CEO, has become the Taylor Swift of the industry"She named SK Hynix and Samsung alongside him as companies along that supply chain drawing interest because supply is so constrainedFounders look at that stack and see inefficient ways to build for AI; from a blank sheet of paper, she said, it would look very different from shoehorning today's infrastructure
The liquidity picture is turning. "with SpaceX going public, Anthropic coming shortly, OpenAI just on the tails", the long stretch of companies staying private with no liquidity is ending, and "now you're going to see multiple trillion-dollar businesses go public"LPs still want the swath of private companies that have not listed, and she said they do not want to keep allocating to asset classes such as private equity that were built off the prior technology cycle
The speed of the raise was itself the signal: when she started her career "it would have been impossible to close a fund effectively in two months over the summer"
$1.1 Billion Is Small Against the Build-Out Because the Fund Buys Ownership, Not Capacity
Jaffee put the scale problem to her. The fund is large, but he said he thinks "American companies are going to be spending something like a trillion dollars this year, a thousand times as much" — so what is the specific niche?
Kha said the sizing was genuinely argued over internally: "There's a partner who I will name unnamed, who said, gosh, we're going to deploy this fund in like six weeks"
The answer is entry point rather than size. "we want to come in at the seed, Series A and then kind of invest at the inflection point", where "we can put $25, $30, $35 million and have a very substantive ownership"
That is a different business from a late-stage or public-market investor, who has to come in later to get any ownership at all — and then needs enough of it to move fund returns
NextHop is the example of a deal that belongs to the growth fund instead. The company builds AI-first high-performance networking, and "but we put in 65 million at the growth stage", which she called a great opportunity for the growth fund as the business was taking off
Unconventional is the shape the Machine Age Fund actually wants. "this is the Naveen Rao company who was formerly with Databricks", rebuilding the chip from the design up around a much simpler structure meant to raise performance"they're kind of going off in the cave and they're going to go and do this endeavor project, but we invested at the seed" — and the company subsequently raised a large up-round
Nvidia Is the Size of a Country, So AI Adoption Is Now a Head-of-State Issue
Dew asked what hardware companies need from partners beyond capital — market access, geopolitical relationships — and whether that is different from what a16z's earlier companies needed. Kha said the question has moved to the national level.
A month ago the president of Korea came to the United States, and "their first and only stop in the U.S. was in Silicon Valley"
The comparison she used for why: "Nvidia, at 5.5 trillion is the same size as countries in terms of their GDP", and "Nvidia is larger than all of the G7 countries except the U.S."
She borrowed her partner Marc Andreessen's line for the shift: "technology is the dog that caught the bus" — technology is a bigger and bigger part of the economy, and countries are treating getting their citizens and their governments onboarded into the AI age as a national priority
The historical analogy is industrialization: "if you look at countries of the last 100 years that were the most militarily, culturally, economically successful, they were the ones that industrialized first"She expects the countries that win the AI era to be the ones that adopt it first across public safety, health care and government generally, with citizens pulling it through from the bottom up
What global partnerships means in practice: helping LPs, who are often global, adopt the latest technologies, which are usually American-built, and in many cases co-invest directly alongside the firm — "How can we help America and her allies accelerate into the future"The pitch to a government is that the firm wants to be a partner in how the country thinks about technology, not only in the capital contract, and that it benefits from the economic development that follows
Korea, El Salvador and the City-States Are Moving Faster Than the United States
South Korea has just announced premium AI for every citizen, as a public utility. Kha called the country "the most extreme form of democracy" and said that cuts both ways
She calls Korea the country of the future, and not only for AI: "they've got the number one boy band in the world with BTS. They have the number one girl band with Blackpink", plus Squid Games and KPop Demon Hunters"they're like the new Hollywood of the future from a content perspective, but they're also full board into AI", which is why the AI rollout did not surprise her
El Salvador is the other example. "President Bukele has been super aggressive around utilizing AI. They implemented Grok in their schools, for example, for free." — and the country is using AI doctors
City-states can move fastest. "sometimes there's city-state countries like Singapore or UAE, for example, where it's easier, just given the population to adopt technologies", and to subsidize them or make them free
The United States is going the other way, and she thinks much of the objection is false. She pointed to political headwinds around data centers and around surveillance "that are just falsehoods"The firm reads that as an opportunity as much as a risk: with "Elon's building data centers in space", she said a lot of the data center supply chain build could land overseas because of US sentiment
The Data Center Backlash Is a Narrative Problem; DC Power Is the Real Bottleneck
Jaffee asked how public backlash against data centers affects the fund — whether it becomes much harder to build, or whether data centers simply get built wherever they are allowed and everything is basically fine.
Kha said the process would obviously be easier if everyone were positive on data centers, but that is not the world, and "the narrative has gotten in the way of reality". The modern ones — she named AWS, Meta's data centers and Switch Data Center, an a16z portfolio company — are "modern versions of the data center that are built by tech people, not real estate people", built around the sensitivities people have.
On Switch specifically: "they actually contribute power back to the grid. They don't take away.", they use very little water, and they are among the very few able to manage fluidics for future chips
"everyone paints data centers with a broad brushstroke, but there's a very small percentage of bad actors with data centers" — the vast majority, she said, have configured for the new form factor
The next constraint is electrical, and it is a labor problem. The next set of chips will be DC powered rather than AC powered, and most data centers are not built for that"there's less than 2% of electricians in the U.S. that are actually trained on DC power because it's very dangerous"
The fund's focus sits on exactly those bottlenecks in the infrastructure
The Demand Side: Agents Use Five Times the Tokens Humans Do
Dew pointed out that the mandate runs past data centers into robotics and home AI hardware, and asked for the broader thesis.
Everything inside the data center is in scope: accelerators, CPUs, custom silicon, memory, storage and liquid cooling, plus robotics within the data center and system software"all things that were forgotten language, if you will, from an investable category perspective" for the last several decades
On demand: "agents are using five times the amount of tokens as humans are", and she said the crossover has already happened — there are now more agents than humans on the internet, and the trend goes parabolic from here as consumers pick up personal use cases
The supply side is the fund. It backs what is guided by computer science inside the data center
Regulated industries are out of scope: "we don't want to invest into regulated industries, like power, for example. That's more of our American Dynamism team."
Hardware Went From Almost Nothing to More Than 20% of the Pitches, and the Founders Are Throwbacks
Jaffee came back to the six-weeks line and asked how long it actually takes to identify and diligence deals of this scale, and whether it differs from software venture deals.
The deal flow is why the joke exists. Entrepreneurs have come out of the woodwork, she said, telling the firm the process is incredibly inefficient for the demand they can see, and that they want to build a company against it
"I think Martin was quoted saying, we went from seeing basically nothing in terms of hardware pitches to over now 20% of our pitches are hardware"
"The founders in this category are sort of like a throwback in some respects." The work needs relationships with the hyperscalers and with customers, and a real grasp of the physical dynamics, so the talent is coming from a prior era and often spinning out of the incumbentsNextHop's founders "were former folks from Arista, for example, that had recognized this problem"These are people who have been incubating the problem for twenty or thirty years and are only now building around it
The team doing the diligence sold into the data center themselves. "Martin Casado, formerly CEO and founder of a company called Nicira, which sold into VMware"; Raghu Raghuram, VMware's former CEO, who acquired Nicira and was Casado's boss there; and "Guido Appenzeller, who is actually the former CTO of Intel", who ran Intel's data center business"you're seeing folks who, for maybe the last 10, 15 years, became more pulled into software, come back into the hardware side of the world"
Because the set of people who can build at this scale is finite, diligence usually runs through people the firm already knows, and through customers it already knows — its own portfolio companies, or the hyperscalers
Dew closed the interview on the same note the fund's pitch ends on — "What's old is new again." Kha agreed: "we're almost thrown back to where venture capital actually started and why Silicon Valley is actually called Silicon Valley".
Kha's bottom line is that AI has taken software off the list of constraints and moved the constraint into the physical world, and that a dedicated $1.1 billion fund exists because a company building chips, networking or data centers can never win a like-for-like comparison against a software deal inside a generalist portfolio.
Products, Companies & Tools Mentioned
The Machine Age Fund (a16z's new $1.1 billion vehicle for everything below the software stack — chips, custom silicon, networking, racks, memory, storage, liquid cooling, data centers and robotics inside them)
a16z's Infra Fund, Apps Fund and American Dynamism team (The vehicles the new fund sits beside: developer products, B2B and B2C software, and the regulated industries such as power that the Machine Age Fund deliberately leaves alone)
Nvidia (Her measure of how far technology has outgrown the economy — "Nvidia is larger than all of the G7 countries except the U.S.")
SK Hynix and Samsung (Named with the CEO she calls the Taylor Swift of the industry as data center supply chain companies whose stocks "have been ripping" because supply is so constrained)
SpaceX, Anthropic and OpenAI (The listings pipeline behind her argument that the private-markets logjam is about to clear)
NextHop (AI-first high-performance networking, founded by former Arista people; a16z "put in 65 million at the growth stage")
Unconventional (Naveen Rao's chip company, formerly of Databricks, redesigning the chip around a simpler structure for AI; a16z invested at the seed and it raised a large up-round after)
Switch Data Center (a16z portfolio company and her counter-example to the backlash: contributes power back to the grid, uses very little water, and can manage fluidics for future chips)
AWS and Meta (Cited alongside Switch as modern data centers "built by tech people, not real estate people")
Grok (Deployed free in El Salvador's schools, in her example of a government adopting AI faster than the US)
Arista, Nicira, VMware and Intel (Where the founders and the a16z partners in this category came from — NextHop's founders from Arista, Martin Casado from Nicira, Raghu Raghuram from VMware, Guido Appenzeller from Intel)
Books & Resources Mentioned
The Machine Age Fund – a16z (The firm's own announcement of the fund and the thesis behind it) — https://a16z.com/the-machine-age-fund/
Get the latest market chatter as it happens: