Jen Kha | The a16z Show Summary

1. Key Themes

Hardware Is Back: Venture Capital's Return to Its Roots

After decades of moving away from hardware toward software, AI is reversing the trend dramatically. a16z went from seeing essentially zero hardware pitches to hardware now representing over 20% of deal flow.

"We went from seeing basically nothing in terms of hardware pitches to over now 20% of our pitches are hardware kind of physical world related side." [00:19:56]

"What's old is new again... we're almost thrown back to where venture capital actually started and why Silicon Valley is actually called Silicon Valley." [00:00:59]

The Existing Infrastructure Stack Was Never Built for AI

The core thesis is that the entire physical layer of computing — chips, networking, memory, cooling, power — was designed for the prior era of the internet and SaaS, and needs to be rebuilt from scratch for AI's mathematical intensity.

"All the physical parts of enabling AI, from data centers to chips to custom silicon, networking, Rackspace, all the stuff in the physical world that was honestly largely an uninvestable category for the most part for the last 30 years, because we kind of built that infrastructure out for the last era of the internet... all that infrastructure, sort of the poor man's version that we're limping along with today, needs to be repurposed for the AI age going forward." [00:02:35]

AI Adoption as a National Priority — Governments Are Moving Faster Than Expected

Countries around the world are treating AI as a strategic national interest, and some smaller or emerging nations are moving faster than the U.S. The Korea presidential visit to Silicon Valley as the sole U.S. stop is emblematic of this shift.

"A month ago, the president of Korea came to the U.S. And their first and only stop in the U.S. was in Silicon Valley... countries are viewing this as a national priority and interest to get their citizens, to get their government onboarded into the AI age." [00:10:52]

"Countries of the future are going to be the ones that adopt AI first across defense, across public safety, across healthcare." [00:11:37]

Agents Are Driving a Parabolic Demand Shock

The token consumption of AI agents is already overwhelming existing infrastructure, and we are still at less than 5% of AI usage penetration — meaning the supply constraint is only going to intensify.

"Agents are using five times the amount of tokens as humans are. And we just crossed over to agents before. The volume of agents are now more on the internet than humans are... there's less than still 5% of AI usage today." [00:18:28]

The Founder Profile for Hardware Is Uniquely Experienced

Unlike typical software startups, hardware AI companies are being built by veterans with deep prior-era experience spinning out of incumbents — people who spent 20–30 years incubating the knowledge needed to rebuild this infrastructure.

"You are seeing a generation of talent that are from a prior era, sometimes spinning out from some of the incumbents to start companies... people who have the benefit of experience and prior errors of the buildout that had been effectively kind of incubating for the last 30 plus years, 20, 30 plus years and now saying, I'm going to go build the company around this." [00:20:50]

Private Markets Capture the AI Value, Not Public Markets

Institutional investors are starving for private AI exposure because most of the value creation is happening before companies go public. The coming wave of trillion-dollar IPOs (SpaceX, Anthropic, OpenAI) is accelerating LP urgency.

"If you look at most of the value accrual in AI, that's largely been on the private side, not the public side. And so people are just starving for private capital in general, because that's where all the growth is." [00:05:36]

"You're going to see multiple trillion dollar businesses go public. And so people will have access to that. But there's a whole swath of companies on the private side that are not yet public. And so people want exposure to that." [00:07:20]

Data Center Political Backlash Creates Opportunity for Next-Gen Builders

The narrative against data centers is creating an opening for technically sophisticated, next-generation data center companies that actually solve the environmental and power concerns the critics raise — while also being purpose-built for AI workloads.

"The narrative has gotten in the way of reality... they're one of the very few data centers that are actually capable of being able to manage around fluidics in the future for chips... the next set of chips are going to be DC-powered versus AC-powered. And most data centers aren't actually built for that infrastructure." [00:15:49]

a16z's Global LP Network as a Competitive Moat

Jen's role as head of global partnerships reframes LPs not just as capital sources but as go-to-market partners — countries and sovereign funds that can accelerate adoption of portfolio companies domestically.

"We want to be partners not just in the capital construct, but also in how a country and government is thinking about technology if we can help them accelerate that adoption. And then also, quite frankly, benefit from the economic development of it as well." [00:12:56]

2. Contrarian Perspectives

AI Has "Solved" Software — The Real Problem Is Now Physical

Most investors are still focused on software layer competition, but Jen argues AI has effectively commoditized software creation, shifting the real bottleneck — and therefore the real investment opportunity — entirely to the physical world.

"Mark had famously said software's eating the world. 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." [00:03:32]

Smaller and Non-Western Countries Will Lead AI Adoption

Counterintuitively, countries like El Salvador and Singapore are moving faster on AI adoption than the U.S., which is hamstrung by political headwinds around data centers and surveillance concerns.

"You see these different examples around the world where they're accelerating their AI development and adoption way faster than in the U.S... we might also see the influx of a lot of this data center supply chain buildup happen overseas because of this sentiment in the U.S. as well." [00:00:09]

Data Centers Are Being Judged by the Worst Actors, Not the Best

The political narrative paints all data centers as bad environmental actors, but Jen argues the modern, tech-company-built data centers actually contribute power back to the grid and use minimal water — the broad brushstroke is factually wrong and creates mispriced opportunity.

"There's a very small percentage of bad actors with data centers. Vast majority have actually configured for the new form factor. And where we are investing is the next generation of those data centers... that have optimized for the future." [00:16:44]

Hardware Was Uninvestable for 30 Years — That Itself Is the Opportunity Signal

The very fact that hardware was ignored for so long means the infrastructure is deeply mismatched to current demand, and the category is relatively uncrowded despite being massive.

"All the stuff in the physical world that was honestly largely an uninvestable category for the most part for the last 30 years." [00:02:35]

3. Companies Identified

Next Hop AI-first high-performance networking company, founded by former Arista veterans. Mentioned as a portfolio investment at the Series B growth stage, representing the kind of experienced-founder spinout a16z is backing.

"We invested in a company called Next Hop, which is building kind of AI first high performance networking. And this was a company that was at the series B inflection point, but we put in 65 million at the growth stage." [00:09:24] "Next Hop, were former folks from Arista, for example, that had recognized this problem and decided to start a company to go actually after the new version and form of this." [00:20:50]

Unconventional Chip design company founded by Naveen Rao (formerly of Databricks), rebuilding chip architecture from first principles for AI workloads. Backed at seed by a16z with a significant up-round shortly after.

"This is a company that's actually rebuilding the chip from the design perspective of if you were to design a chip for AI. And so what they're trying to do is target much simpler structure to enhance performance in the chips themselves... we invested at the seed. And subsequently, they raised a huge up right after that." [00:09:49]

Switch Modern data center company that contributes power back to the grid, uses minimal water, is built for liquid/fluidics cooling, and is architected for DC power — a portfolio company cited as the model for next-generation data center investment.

"Switch data center, which is one of our portfolio companies... they actually contribute power back to the grid. They don't take away. And they've built their infrastructure to enable that. They use very little water. And they're one of the very few data centers that are actually capable of being able to manage around fluidics in the future for chips." [00:15:49]

NVIDIA Named as the canonical example of tech companies reaching GDP-scale, used to illustrate why AI infrastructure is now a geopolitical and national priority matter.

"NVIDIA, you know, at 5.5 trillion is the same size as countries in terms of their GDP. So NVIDIA is larger than all of the G7 countries except the U.S." [00:10:52]

SK Hynix / Samsung Named as public market examples of data center supply chain companies whose stocks have been outperforming due to compute supply constraints.

"Companies also like SK Hynix, like Samsung. These are all kind of companies along the data center supply chain that have gotten so much interest because we're so constrained on the supply side." [00:06:05]

Anthropic / OpenAI / SpaceX Cited as the vanguard of the shift from private-staying-private to trillion-dollar IPO candidates, driving LP urgency into private AI exposure.

"We've seen in the last couple of months, so obviously with SpaceX going public, Anthropic coming shortly, you know, OpenAI just on the tails. I think people are now seeing this shift from what has historically been private staying longer and longer, no liquidity coming out. And now you're going to see multiple trillion dollar businesses go public." [00:06:52]

Nicera (acquired by VMware) Martin Casado's former company, cited to establish the deep data center credibility of the Machine Age Fund's investment team.

"Martin Casado, you know, formerly CEO and founder of a company called Nicera, which sold into VMware." [00:21:43]

4. People Identified

Jen Kha Managing Partner and Head of Global Partnerships at a16z. Leads capital raising strategy and LP relationships globally. Orchestrated raising the $1.1B Machine Age Fund in approximately two months over summer, capturing over 23% of all venture capital raised this year for a16z.

"We raised, I think the numbers close to over 23% of all venture capital at this year's. We've been on a tour." [00:05:36]

Martin Casado General Partner at a16z, leading the Machine Age Fund. Former CEO and co-founder of Nicera (sold to VMware). Deep data center and networking background.

"Martin Casado, you know, formerly CEO and founder of a company called Nicera, which sold into VMware... both of them come from a very deep kind of selling into the data center background." [00:21:43]

Raghu Raguram General Partner at a16z on the Machine Age Fund. Former CEO of VMware, which acquired Nicera. Martin Casado's former boss at VMware.

"Raghu Raguram, who was a former CEO of VMware, who actually acquired Nicera and then was Martin's boss while at VMware." [00:21:43]

Guido Appenzeller General Partner at a16z on the Machine Age Fund. Former CTO of Intel, ran Intel's data center business. Described as very technical with deep hardware expertise.

"Guido Appenzeller, who is actually the former CTO of Intel, is also someone who is very, very technical, sold into and ran kind of the data center business for Intel as well." [00:22:05]

Naveen Rao Founder and CEO of Unconventional, a16z's seed-stage chip design portfolio company. Formerly at Databricks. Long-standing relationship with a16z.

"The Naveen Rao company who was formerly with Databricks that we've known for a very long time. And this is a company that's actually rebuilding the chip from the design perspective of if you were to design a chip for AI." [00:09:49]

Sanjay Mehrotra (referenced as "Sanjay, the CEO") CEO of Micron Technology. Praised internally at a16z for grinding through difficult years to become the "Taylor Swift of the industry" as memory demand has exploded.

"Sanjay, the CEO, has become the Taylor Swift of the industry, right? Like he has made much credit to him. He's been grinding the grind for a very long time." [00:06:05]

Chris Dixon General Partner at a16z. Credited with the "follow the nerd energy" framework used to justify the Machine Age Fund's creation.

"My partner, Chris Dixon, calls it following the nerd energy. Like, what is the nerd doing on nights and weekends is probably what us normies will be doing in the future." [00:03:32]

Mark Andreessen Co-founder of a16z. Credited with two key framings: "software is eating the world" (now superseded by AI solving software) and "technology is the dog that caught the bus."

"Mark had famously said software's eating the world. Turns out AI has solved software, right?" [00:03:32] "Our partner Mark Andreessen calls it, you know, technology is the dog that caught the bus." [00:11:17]

Nayib Bukele President of El Salvador. Cited as an aggressive AI adopter — implemented Grok in schools for free and deployed AI doctors nationally.

"It's El Salvador who, you know, President Bukele has been super aggressive around utilizing AI. They implemented Grok in their schools, for example, for free. And they're utilizing AI doctors, for example." [00:13:58]

5. Operating Insights

Separate the Pool of Capital to Unlock a Different Risk Appetite

When hardware deals compete head-to-head with software deals in the same fund, the high upfront capital requirements of hardware almost always lose to proven software revenue. Creating a dedicated vehicle removes that structural bias and enables genuine early-stage ownership in capital-heavy categories.

"If you do the like for like, you're almost always going to bias towards that sure thing with Infra... by separating it into a fund, we're kind of staking in the ground, A, our commitment to the space, and then more importantly, B, from an organizational portfolio construction perspective, that the intention of this is to capture those winners at the earliest stages, maximize ownership." [00:04:27]

Use LP Relationships as Distribution Channels, Not Just Capital Sources

a16z is using its global LP base — sovereign funds, government-linked investors — as active go-to-market partners who can accelerate adoption of portfolio companies in their home countries. This is a non-obvious competitive advantage that compresses the sales cycle for portfolio companies internationally.

"How can we help them utilize the latest and greatest technologies... potentially utilizing these technologies by adopting these technologies and then also oftentimes coming alongside of us in investing directly in some of these companies as well. And then by doing so, you could actually accelerate the adoption process locally." [00:12:06]

Signal Fund Formation as a Recruiting Tool for Founders

The act of launching a dedicated fund is itself a signal to the entrepreneurial community. It changes what kinds of founders show up and creates a self-fulfilling pipeline of the right deal flow.

"By creating a separate fund, we're putting out the back signal to the world, two entrepreneurs who are building to spend time with us, first and foremost." [00:04:02]

6. Overlooked Insights

The DC Power Transition Is a Massive, Under-Discussed Bottleneck

Jen mentions almost in passing that the next generation of AI chips will be DC-powered rather than AC-powered, and that less than 2% of electricians in the U.S. are trained on DC power — which is also extremely dangerous to work with. This is not a software problem or even a chip design problem; it's a skilled trades infrastructure problem that will constrain AI buildout in ways that almost no one is talking about publicly.

"The next set of chips are going to be DC-powered versus AC-powered. And most data centers aren't actually built for that infrastructure. And by the way, there's less than 2% of electricians in the U.S. that are actually trained on DC power because it's very dangerous and you could potentially kill yourself. And it's very volatile to work with." [00:17:06]

This creates a potential investment thesis entirely separate from the chip or data center layer: companies that train, certify, or provide DC-power-capable electrical labor, or companies that design data center electrical infrastructure specifically for DC conversion.

Political Headwinds in the U.S. May Accelerate Global Data Center Buildout — and That's an Opportunity

Jen briefly notes that Elon is building data centers in space, and that U.S. political resistance to domestic data center construction may push the entire supply chain offshore. This is framed almost as a throwaway observation, but it implies that non-U.S. jurisdictions — particularly LP countries that a16z already has relationships with — could become the primary build sites for next-generation AI infrastructure, giving a16z a structural first-mover advantage in directing that capital flow.

"We might also see the influx of a lot of this data center supply chain buildup happen overseas because of this sentiment in the U.S. as well." [00:00:09]