Context

This is a companion to a16z's broader "Infrastructure Behind the Machine Age" discussion, but told from a different seat. Jen Kha, managing partner and head of global partnerships at a16z, explains the mechanics behind the firm's new $1.1B Machine Age Fund, dedicated to the physical layer beneath AI (chips, custom silicon, memory, networking, cooling, data centers, and in-data-center robotics). The infrastructure thesis itself (the model is no longer the bottleneck, the layer beneath it is) is covered more fully elsewhere; the distinct value here is the how and the why-now: why this needed its own fund rather than the existing ones, how she reads investor and national demand, and why AI adoption is becoming a matter of competitive and even geopolitical priority. The lessons about structuring capital, spotting early signal, and building for objections transfer well beyond venture.

The Big Idea

How you structure your investment vehicle determines what you can actually back. By carving out a dedicated fund for capital-heavy hardware, a16z avoids the trap where a like-for-like comparison against capital-light software always wins, and positions to own the emerging physical layer of AI at its earliest, cheapest stage.

Kha is explicit that a hardware deal needs large upfront investment before there is any product, so if you judge it head-to-head against a software deal in the same pool, you will "almost always bias" toward the software one. A separate vehicle removes that bias.

Key Insights

Structure the vehicle so unlike bets don't compete

The sharpest transferable lesson. Hardware companies raise a lot of capital just to get off the ground, long before revenue, while a software deal can be doing a billion or two in revenue off the cuff.

• The problem: if you evaluate a capital-heavy hardware bet like-for-like against a capital-light software bet in the same fund, you almost always pick the software one, because the near-term numbers look better. The structure quietly makes the decision for you.
• The fix: a dedicated pool with its own portfolio-construction rules lets those bets be judged on their own terms, and signals commitment to founders in the space.
• Why it matters for PMs and leaders: the same dynamic governs any shared resource pool (a roadmap, a budget, a headcount allocation). If long-payback, capital-heavy bets sit in the same bucket as quick, cheap wins, the cheap wins win by default. Sometimes the only way to fund the important-but-slow thing is to give it its own protected pool.

Follow the nerd energy

a16z's early-signal heuristic, credited to Chris Dixon: watch what "the nerd" is doing on nights and weekends, because that is likely what everyone else will be doing later. Entrepreneurs are the leading indicator. The concrete signal here: hardware went from almost nothing to more than 20% of the pitches a16z sees. The transferable point is to treat the earliest, most obsessive builders in a space as your forward-looking demand signal, rather than waiting for a trend to be obvious in the numbers.

Get ownership before the inflection

A timing principle. a16z wants to enter at seed or Series A, before a company inflects, when a $25-35M check buys substantive ownership, rather than at the late stage where you need a much larger check for the same stake.

• The contrast: they cite entering NextHop (AI-first high-performance networking) at a $65M growth-stage check versus getting into a company like Unconventional (rebuilding the chip from an AI-first design) at seed, before it inflected.
• The general lesson: the cost of a given level of ownership or influence rises sharply once something is obviously working. Whether it is equity, a market position, or a strategic bet, getting in before the inflection is far cheaper than buying in after it is proven.

Adoption speed is a national advantage

Kha frames AI adoption as the new equivalent of industrializing first. The countries that adopted industrialization first led militarily, culturally, and economically; she argues the countries that adopt AI first across defense, public safety, and healthcare will lead next.

• The examples: South Korea announcing premium AI for every citizen as a public utility, El Salvador putting Grok in schools for free and using AI doctors, and city-states like Singapore and the UAE subsidizing adoption because their smaller populations make rollout easier. Several are adopting faster than the US.
• The partnership angle: a16z's global LP relationships are increasingly about helping countries adopt (often US-based) technology, not just about capital. Distribution and adoption, not just invention, decide who benefits.
• Why it matters: the lesson generalizes to companies. Being first to adopt a transformative technology across your operations, not just first to build something, is itself a durable advantage, and it can shift geographically toward whoever removes the friction to adopt.

Build for the objection, then fight narrative with reality

On the backlash against data centers, Kha argues "the narrative has gotten in the way of reality." Modern data centers built by tech people rather than real estate people (she cites portfolio company Switch) are designed for the actual objections: contributing power back to the grid rather than only drawing from it, using very little water, and being ready for future needs like liquid cooling and DC power. A small percentage of bad actors get the whole category painted with a broad brush. The transferable move: when your category faces reputational backlash, the answer is to engineer directly for the stated objections and then point to the reality, rather than letting the worst actors define the narrative for everyone.

Hard domains need experienced founders

Because these companies require deep relationships with hyperscalers and customers and a real grasp of physical build-out, the founders are "a throwback": experienced operators spinning out of incumbents rather than 22-year-olds (NextHop came from Arista veterans; the leadership brings backgrounds from Nicira, VMware, and Intel's data-center business). The diligence reflects it: a finite world of people who can credibly attack these problems, assessed largely through existing relationships and hyperscaler customers. The lesson: for domains that are technically and operationally elaborate, relevant experience and an existing network are closer to prerequisites than in pure software.

Mental Models & Frameworks

Don't judge unlike bets in one bucket

A resource-allocation model. When bets with very different capital needs and payback horizons compete in the same pool under the same near-term criteria, the fast, cheap ones win structurally, regardless of long-term value. The remedy is to separate them into pools with their own rules so the slow, heavy, important bet is evaluated on its own merits. Use it whenever a strategically important initiative keeps losing funding to quicker wins.

What's old is new again

Kha's framing of hardware's return to venture: for 30 years capital moved away from hardware toward software, and AI is pulling it back, almost to where venture capital and Silicon Valley began (the silicon). The model is category rotation: what was "uninvestable" or unfashionable for a cycle can become the most active area when the underlying constraints change. It is a prompt to periodically re-examine categories everyone wrote off, because a shift in constraints can revive them.

Practical Application

Protect the slow, heavy bet in its own pool

If an important initiative with a long payback keeps losing to quick wins in your roadmap or budget, stop making them compete in the same bucket. Give the strategic bet a protected allocation judged on its own criteria, the way a16z gave hardware its own fund, so it is not biased out of existence by nearer-term comparisons.

Track the earliest builders as demand signal

Watch where the most obsessive early builders are concentrating (the "nerd energy") as a forward indicator of where demand is heading, rather than waiting for a trend to show up in mainstream metrics. A sharp rise in a previously ignored category (hardware going from near-zero to 20% of pitches) is the kind of early signal to act on.

Engineer for your category's objections

If your product or category faces reputational backlash, identify the specific stated objections and build directly against them (as modern data centers did with power contribution and water use), then lead with that reality. Do not let the worst actors in your category define how everyone is perceived.

Questions to Consider

• In our roadmap or budget, is an important but capital-heavy, slow-payback initiative quietly losing to quick, cheap wins because they share the same pool and criteria, and should it have its own protected allocation?
• Where is the "nerd energy" in our space right now (what are the earliest, most obsessive builders working on), and are we treating that as a forward demand signal or waiting for it to be obvious?
• Are we trying to get ownership or a strategic position in something after it has visibly inflected, when getting in earlier would have been far cheaper?
• For a transformative capability, are we first to actually adopt it across our operations, or only watching others build, given that adoption speed itself is becoming the advantage?
• What are the specific objections our product or category faces, and have we engineered directly against them rather than letting the worst examples define our narrative?

Bottom Line

The Machine Age Fund is a lesson in structure as strategy: capital-heavy hardware bets need their own vehicle because they lose a like-for-like comparison to capital-light software every time, so isolating them is what makes backing them possible. Around that sit broadly useful ideas: follow the earliest builders as your demand signal, secure ownership before the inflection, treat adoption speed as a competitive and national advantage, and engineer directly against your category's objections rather than ceding the narrative.

Notable Quotes

"Follow the nerd energy. What is the nerd doing on nights and weekends is probably what us normies will be doing in the future." (Jen Kha, citing Chris Dixon)

"The narrative has gotten in the way of reality." (Jen Kha, on data centers)