Defense tech is a good example. Shape a regulation, get language into the NDAA, land a mandate, and demand becomes certain overnight. The revenue is real and sometimes fast, but the benefits rarely accrue to you alone. A ban on Chinese drones lifts every American drone maker at once. A new acquisition pathway is open to anyone who can fill out the paperwork. And rules change: the next administration, appropriations fight or program office reorg can take it all back. If your only edge is a rule in Washington, you don’t own your moat. You’re renting it.

Here’s how we think about the trade-offs:

We focus on the bottom right: an unproven market and a brutal build. Here uncertainty works in your favor. By the time the market agrees the thing is real, you’ve spent three years learning how to make it, sell it and support it. When certainty finally shows up, you’re the only one who can deliver. A moat built that way is very hard to take back (Falcon 9, anyone?).

Uncertainty alone isn’t a strategy, though. The bottom left is uncertain too. If an easy bet works, the answer is public and the clones show up the next week. You need uncertainty and difficulty together.

And pick uncertainty you can resolve. Adoption, cost curves and manufacturing at scale are questions that building answers. Physics that may not work, or the next election, are things you just wait out. Bet on the first kind.

Take the regulatory tailwind when it comes, because it can be a great catalyst. Just don’t build the company on it. Rent the tailwind and earn the moat.

In Hard Tech, a leaky bucket is your base state. Most customers aren’t on long-term contracts, and there’s no real lock-in early on. You have to deliver product consistently and at scale. If you can’t, customers churn.

The fix is your backlog. The rate at which you turn that backlog into revenue is your throughput.

Revenue = Backlog × Throughput Rate

To grow revenue, you either grow the backlog or raise the throughput rate. Your operating profitability comes down to how well and how consistently you convert backlog into revenue while operating as close to your theoretical max throughput rate as possible.

Fast revenue growth is great, but it burns down your backlog. To keep growing, your backlog has to grow at least as fast as your revenue.

The same goes for throughput. If you invest to push orders through faster, you have to refill the backlog just as fast to keep up your growth rate. This is also where capital is most often destroyed.

We often hear “suppliers take forever to fill orders.” In our experience, that’s on purpose. They’re managing their backlog to keep several months of revenue visibility at the throughput they can actually deliver.

You can invest in capacity. But if you don’t have the backlog to fill it, you take a massive capital charge and destroy equity value.

To win at scale:

- Keep refilling the backlog

- Grow production capacity carefully

- Balance betting on future demand with new capacity against protecting margins by extending the backlog

The biggest legacy players got that way from managing these intersecting variables thoughtfully.

Don’t be so quick to dismiss them, learn from what they do well then make it an order of magnitude better.

“…when we believe our actions do not affect outcomes, we become passive and lose hope. Subjects who did not feel helpless in the face of challenging circumstances viewed difficult moments as temporary, did not blame themselves for mistakes, and believed they had agency over their circumstances. To cultivate true agency, and ultimately optimism, kids must experience the direct connection between their choices and the outcomes that follow.”

In short, optimism is a learned belief that your choices affect outcomes. It’s a trait we spot early in high-potential founders. It isn’t blind optimism, burying your head in the sand and hoping for the best. It’s a belief, backed by evidence, that you can take on hard problems and work through them to good outcomes. Doing that consistently, over time, is where meaningful change comes from: in technology, in startups, and in life.

Neosuppliers leverage the latest technology to improve customer experience and product margins, and earn higher customer LTV as a result. They can scale fast, often getting to 9-figure revenue in just a few years if executed and capitalized well.

A Neosupplier is a supplier first and a software company second. It wins OEM programs because it’s faster, more reliable, and easier to work with than a legacy Tier 1, and AI is the reason it can be.

Neosuppliers that are viable as venture investments share a few core characteristics:

• Speed as the product. AI-driven quoting, DFM feedback, and process planning compress weeks into hours. An OEM gets a quote, a manufacturability critique, and first articles faster than a legacy shop can return an email.

• Engineering partner, not build-to-print. It co-designs subsystems with the OEM and owns integrated assemblies. That’s what makes it a true Tier 1 rather than a job shop.

• Selective vertical integration. It owns the bottleneck processes where quality, IP, or lead time is decided, and orchestrates a vetted network for everything else. Its edge comes from knowing which is which.

• High-mix, fast-ramp economics. Flexible automation and low changeover costs let it make money at prototype volumes and scale to rate production without requalifying.

• Data as the moat. Every part generates process and quality data, which feeds a closed loop that improves yield, pricing accuracy, and lead times. Traceability and compliance (AS9100, ITAR, CMMC) are built in from day one.

• Labor leverage. It needs fewer, more skilled people per dollar of revenue, often multiples of a legacy supplier’s revenue per employee.

• Better margins than the category. It structurally beats the thin margins of contract manufacturing, through pricing power on speed and lower cost to serve.

• Domestic and allied by design. It’s positioned for reshoring and defense demand, where resilience is worth a premium.

Importantly, if it doesn’t take ownership of the part and the outcome, it’s not a Neosupplier. Marketplaces that broker other shops’ capacity, or a SaaS tool sold to factories are not Neosuppliers.

A simple diligence test: would an OEM sole-source a critical subsystem to this company, and is that decision getting easier with each program it delivers? If yes, that’s an interesting opportunity!

Over time, I’ve shifted my definition.

A risk is what you get paid for!

Within that, there’s “good” and “bad” risk.

Good risk is something like “we believe the application layer is where AI value will accrue. We could be wrong.” It’s something that can be influenced, but is largely out of your control.

Bad risk is “we don’t know who this founder truly is and what motivates them.” These risks are largely in your control. You can kill them with diligence, or direct company support

Knowing the risks the market’s actually paying you to take is 90% of the investing game.

Doing it well is a real art.

I’ll keep the specific responses private, but to summarize, the answers were decidedly human. Beautiful vistas. A milestone for a family member. The achievement of a friend.

The answers each shared two elements: they were exceptional moments, and extremely personal. A challenge overcome. A place so beautiful it defies reality. A creation someone brought into the world that made them rethink what’s possible.

Not all of these things were grandiose. In fact, what struck me was how seemingly small many of these inspirational moments were, but important to the individual nonetheless.

There’s many ways we can choose to spend our time, but I personally feel there’s few better ways to do it than living your life in a way that inspires others. This is even more important as technology eats more and more of our daily lives.

Pursue things that are hard. Be a source of inspiration that helps others feel something. In the age of the machines, this is a big part of what it means to be human.

What I took from this was that automation alone isn’t enough to win. At its core, automation just shifts costs from opex (e.g. labor) to capex (e.g. equipment/robots). What that automation enables (e.g. lower prices, more uptime, etc.) is far more important than the fact that something is, in fact, automated.

As the market gets more infatuated with startups doing “automation”, a nuanced understanding of the technology behind it and what it enables are critical. Is a proposed innovation a true automation breakthrough or something a legacy system integrator could design and implement? What is the measurable payback on NRE (non-recurring engineering cost)? Does the return from incurring that NRE cost allow a manufacturer to (durably) win on price? As a general rule, the more specific the value prop, the more valuable the company equity.

Go see more factories if you can. I think you’ll be surprised at what you’ll learn!

I got this question earlier on a call. It’s a good one. The instinct is to do a lot of diligence. Technical, market, financial. I don’t think this is right.

My advice? Find the five smartest people you know, and try to convince them to join your quest.

In the process of doing that, yes, you’ll be forced to answer the technical, market, and financial questions you face. You’ll also learn whether you have a natural ability to sell.

But most importantly, that group choosing to spend their scarcest resource (time...) with you should give you all the conviction you need to move forward.

Worst case, you’ll have a lot of fun with people you respect.

Best case, you’ll build something generational.

“Nothing is a line. Everything, everywhere is always moving. Forever.”

- Logan Roy, in Succession

To be successful, you have to embrace the dynamism and uncertainty of building and operating. You’re measured on outcomes, not inputs.

Said differently, you have to play to win.

“In some cases, NVIDIA may provide a residual-value support mechanism for up to 25% of an opportunity, assessed carefully on a project-by-project basis. That support is limited, residual-value based and designed to complement — not replace — independent underwriting.”

Why is this important? To keep it simple, it comes down to the operating expense of running these assets, specifically depreciation. Let’s use an example:

If a GPU costs $100, with a 5-year useful life and is depreciated to zero, the annual depreciation expense is $20. If I generate $100 of revenue per year with those GPUs against $30 of additional COGS (beyond depreciation), then my gross margin is $50, or 50%.

Now take an example where the GPU has $25 of residual value (e.g. I assume I can sell the GPU to someone for $25 after 5 years of use). In this case, my annual depreciation is $15 and total annual COGS are $45. All else equal, establishing a clear residual value increased my gross margins from 50% to 55%. Multiply this effect across trillions in projected AI infrastructure Capex, and residual value has a very meaningful role to play in future industry profitability.

We can apply the same logic to robotics. Right now, the dominant use case for robotics is in the automotive industry. This mass adoption has established clear and demonstrable residual value (e.g. a used market for 6-DOF robotic arms), which reduces depreciation expense for the asset owners. As a result, we’ve seen the industry drive mass adoption of these systems.

When we look at frontier robotics (humanoids, vertical solutions, etc.), we are still far from the ability to accurately forecast residual value. Do we know how much a farmer will pay for a used weed-killing robot? What about a humanoid that restocks inventory? Because of this lack of agreed-upon residual value, companies must either expense these items in-year, or depreciate them to zero.

For these reasons, simply reducing the Capex required for robotics adoption is insufficient to spur adoption. Because of depreciation, the (non-cash) operating cost of these assets remains high, even when accounting for labor replacement and increased utilization rates. Raising the residual value is likely a much longer lever and a far simpler fix than an industry-wide cost down or performance enhancement exercise. We think there’s a big opportunity for any company that can unblock this!

We lead ~10 Formation-stage investments per fund.

Since 2020, we’ve backed founders building bigger satellites when everyone thought we should go smaller. We’ve backed companies making drugs in space when consensus was that comms was the only business case. And we’ve backed companies commercializing nuclear power before Executive Orders were even a thought.

We’re doing it again with Atlas Motion, which today announced their $11.5M seed, with Greycroft, Sunflower, Ravelin, and angels Jai Malik (AMCA) and Scott Sanders (Forterra) participating.

Atlas convinced us of two things we historically hadn’t been willing to believe:

1. You can generate venture returns building at the sub-system level; and

2. The ability to scale manufacturing is a defensible moat with potential for high returns on capital over time.

The authenticity, credibility, and clarity of thought from Christian, Tom, and Carlo forced us to evolve our thinking. And from what I’ve watched them build and deliver over the past 6 months, I’m glad we did.

They’ve gone from emptying their bank accounts to fund the initial days of Atlas to production capacity of 10,000 motors/mo, scaling to 40,000/mo by year end, representing revenue capacity in the tens of millions, and with real customer contracts already in hand.

They’ve built a world class team of nearly 20 across two continents, with more expansion on the horizon.

Most importantly, they’ve articulated a clear view of the true bottleneck in Physical AI:

“Intelligence is improving monthly. The motion systems are not. Physical intelligence is a manufacturing problem.”

What a fun journey ahead.

Visit them to learn more at https://atlasmotion.com/

This has been 7+ years in the making, dating back to Doug doing (self-funded) power modeling in his garage. It’s a special thing to see something that was a spreadsheet when we first invested in summer 2020 become a real, tangible thing over 6 years later, with the built version matching the 2020 specs nearly perfectly. While the hardware is impressive, it’s what the hardware represents that is most moving. You don’t build hard things without the right blend of mission alignment, seriousness, and fun. Radiant has all of them in spades. Kaleidos is the physical manifestation of a team with complete mission alignment that takes the promises they make seriously, and has a lot of fun along the way.

This idea that physical hardware represents company cultural strength is something that runs through many of our portfolio companies. K2’s massively successful recent Gravitas mission, Varda’s 6th (!!) capsule re-entry, and Northwood’s globally deployed ground sites are just a few examples of physical manifestations of strong underlying company culture. ARR numbers can be faked, but it’s harder to do that with hardware.

If you’re trying to figure out which companies are likely to make it and which won’t, focus on the ones that Build The Thing They Promised. That’s the leading indicator!

Some of these things could be learned in a textbook, but when you see it in person it’s clear that they’re more likely from experience and instinct possessed by someone that’s been obsessed with their craft for decades. They’ve made the mistakes, earned insight from long feedback loops, and have an intuitive sense of when something will just work. This is what makes them an artist.

We see this in the best entrepreneurs. Often they do things that defy logic and physics, but as a result create things that are only possible with enough experience and instinct. It’s this very specific (but intangible) thing that makes what they build defensible. There’s no set of steps someone can take to replicate the output, just a series of judgement calls that compound (positively or negatively) over a long period of time.

Understanding this is one of the main drivers of why the best early stage investors index so heavily on team and track record. What the “film” is about matters less than the unique experience the “director” delivers to viewers. A great movie isn’t made by a great script, or a great lead actor alone. It’s the output of the million small decisions made every day on set.

All that to say…if anyone has a 70mm IMAX Odyssey ticket they’d be willing to spare, please reach out :)

Though companies still make their best attempt to call their contracted revenue ARR, what they really have is backlog. Backlog is revenue that is contracted but hasn’t yet made its way through the production queue into realized revenue. This can be AI implementation service revenue, hardware sales, or government testing data collection, as a few examples. In this world, pipeline is king.

For investors, this new normal also changes how we diligence companies as they grow. Instead of doing SaaS cohort analysis and applying revenue multiples to companies, investors now need to prioritize a more rigorous pipeline analysis. They need to do customer references, analyze contract termination clauses, evaluate cash conversion cycles, and really understand where the marginal revenue will come from when they can’t safely assume that every year there will be a 10% price increase per seat license.

So if we’re headed into a world where backlog is the key metric, what should we look for at each stage of a company’s lifecycle to determine company progress? We look to answer four (often qualitative) questions:

1. Why are customers choosing this company (defensibility, product differentiation, GTM effectiveness)?

2. How big is the pipeline (TAM, “surface area for luck”)?

3. What is the near-term and long-run margin structure for the products and services sold to this pipeline? (e.g. what are the substitutes, how urgent is the need, how unique is this product, etc.)

4. How repeatable is the current sales motion? Has the team distilled GTM down into a core capability or is it scatter shot?

Depending on the category a company is operating in, the answers to these questions vary.

In hardware, product differentiation and defensibility are two of the easiest things to see. Relative to the digital world, building in the physical world is slower and has a more limited universe of alternatives to assess. The risk in hardware land more often tends to be TAM, margin structure, and sales repeatability.

In software today, long-run defensibility is the challenge. Short-run numbers may be great, but avoiding margin erosion and slowing top-line growth feel increasingly difficult to avoid. Without a hardware or network effect moat, unique GTM motions/unfair advantages and extreme talent advantages on the path to economies of scale are more important here than ever.

TLDR; Fun times and lots of opportunity for those willing to forego heuristic-based investing and go back to business valuation basics!

One of the most common are money problems.

There’s two types of money problems:

1. “We don’t have enough money”; and

2. Problems you can throw money at to make go away.

Number 1 is a separate blog post. Today I’m focused on number two. Call them “Type 2” Money Problems. Here’s some examples:

• Your supplier can’t ship for 4 weeks, but your customer needs the product in 3. Solution? Pay 2x to expedite it.

• An engineer you needed yesterday doesn’t want to leave their stock options on the table? Give them a signing bonus to exercise.

• Going to be late to a key meeting? Take an Uber. Unless you’re in NYC, then always just take the subway. It’s faster.

For startups that raise large rounds, it’s almost implied that an increasing proportion of the problems they face are Type 2 Money Problems. The assumption is that speed to scale and market position win, and that money solves that.

The challenge is being sure that’s the case and you’re not papering over another core deficiency of the company. No amount of money can paper over bad culture or leadership. No amount of money will fix hiring if the mission is uninspiring. And no amount of money can will materially accelerate the delivery of truly innovative technical solutions. Only talent and a culture of high iteration speed can do that.

• The largest IPO ever, funding a mission to explore the stars (and minting over 4,000 millionaires…)

• Father-son duos winning NBA championships after 50+ year droughts; and

• The beginning of the World Cup, an event that has historically brought the world together and created moments many citizens cherish forever.

For the romantics, it’s hard to experience these events and not feel something. Feeling something (inspiration, opportunity, potential) is what motivates people to take big risks. And people taking big risks is what creates the potential for change.

Contrast this romantic worldview with the more data-driven and measured view, the “classical.” In “Zen and The Art of Motorcycle Maintenance”, author Robert Pirsig explains the distinction well:

“To a romantic, classic mode often appears dull, awkward, and ugly, like mechanical maintenance itself. Everything is in terms of pieces and parts and components and relationships. Nothing is figured out until it’s run through the computer a dozen times. Everything’s got to be measured and proved. Oppressive. Heavy. Endlessly grey. The death force.”

Both views, romantic and classic, have a role to play in effecting change. But I’m excited that it feels like we’re entering a world where people are beginning to believe ambitious things are possible again.

As unapologetic romantics, we’re always looking for the reasons something might just work, not just the 10,000 reasons it won’t. It’s also a way more fun way to live!

Anyway!

There’s one metric I watch to predict what will happen in venture markets.

It’s not dollars deployed.

It’s not average valuation.

It’s the NUMBER and YoY GROWTH RATE of startups funded in a given year.

Case in point…from ‘95-‘99, number of startups funded nearly 3x’d.

We know how that ended…

Why is this so highly predictive? Because a greater absolute numbers of startups creates an unsustainable dispersion of talent and growth capital that bottlenecks companies getting to large outcomes.

So where are we today? Round count is actually DOWN from ‘24-25. Below even 2019 levels.

Makes you think…🤔

This could be true. But the reality is that any new technology deployment cycle will always oscillate between the euphoria that drives massive amounts of capital behind it and the disappointment when it fails to achieve the massive expectations behind it.

The remedy, however, is not to ban bubbles. In fact, there’s good evidence that bubbles are a feature of the innovation economy. My friend Tobias Huber wrote an entire book on the idea, titled “Boom: Bubbles and the End of Stagnation.” Highly recommend a read!

To quote Tobias’ book preview:

“In a series of case studies tracking some of the most significant breakthroughs of the past 100 years—from the Manhattan Project and the Apollo program to fracking and Bitcoin—they reverse-engineer how transformative progress arises from small groups with a unified vision, vast funding, and surprisingly poor accountability. They conclude that financial bubbles, while often maligned as destructive and destabilizing forces, have in fact been the engine of past breakthroughs and will drive future advances.”

In other words: Bubbles aren’t all bad. :)

A VC saying this is blasphemy, I am aware. But it feels great to get my view out in public.

The last two decades of software made investors lazy. 80% gross margins meant “this is investable.” There was no need to do the harder underwriting of balance sheets, return on capital, and how nascent markets would evolve.

But in this new venture era where investing in hardware is back in vogue and balance sheets are getting bigger, we need to primarily care about two things:

1) Aggregate profit DOLLARS (“Contribution Margin”); and

2) Total capital invested to get there

Profit dollars fuel continued R&D, which deepens the moat and increases enterprise value. And the more profit dollars we generate, the fewer equity dollars we need.

The optimization should always be for the most DURABLE profit dollars generated as QUICKLY as possible with the LEAST about of capital required.

High gross margin (%) is just a proxy for doing that well, but focusing just on that will cause one to miss great opportunities.

1. Building before planning. You ship sub-optimally, then you can’t hire because no one wants to own the mess.

2. Raising while building. You come to market with an MVP and no talent moat. The story doesn’t close/lacks credibility.

3. Building because it’s comfortable. Technical founders are good at it. So they do it. Then they can’t let go and the company stalls.

4. Getting pulled into product minutiae. Small fires feel enormous when you’re the only one fighting them. They don’t when you have a real team around you.

This is why we feel so strongly about founders focusing on hiring early (“First Five”). It’s the difference between a founder running a company and a founder running a product indefinitely.

Instead of building early, hire for the roles that unlock everything else. Do it earlier than feels comfortable.