Greypoint: Building Hardware in an AI Batch

MilesFive days out from Demo Day, and my inbox is still full of pitch decks that all start the same way.

GrantLet me guess. We're the AI layer for-

MilesFor something that didn't need an AI layer. Yeah, that's every third deck this batch.

GrantMm-hmm.

MilesI'm Miles.

GrantI'm Grant, and this is the show where we follow one startup through the one decision that actually mattered.

MilesBut before we get to our founder, I wanna sit in the room for a second because underneath all that AI wrapper noise, something else was happening on that stage.

GrantThe soldering irons.

MilesThe soldering irons. Booths with actual metal on the table, CNC parts, circuit boards, not a single dashboard screenshot in sight.

GrantWhich raises the question I can't shake. Is that a real shift in what YC is funding, or did I just wander into the one corner of the floor that had hardware in it?

MilesThat's the first thing we're chasing today.

GrantAnd the second, even if it's real, can a team of four people with one CNC machine actually go up against Defense primes that have been doing this for decades?

MilesBig contractors, government relationships, decades of institutional weight.

GrantAgainst four people and a machine shop.

MilesAnd somehow that's the pitch everyone in the room wanted to hear more about.

GrantWe've got the founder behind one of those pitches coming up.

MilesBut first, Grant, walk me through what's actually in the data because the vibe in that room and the numbers behind it might be two different stories.

GrantPull it up. Let's see what the Batch actually did because right now this is all just vibes and soldering irons.

MilesSo I actually went and pulled the numbers this morning.

GrantAnd?

MilesTwenty-three percent.

GrantTwenty-three percent of what?

MilesOf the whole batch. Two hundred forty-five companies launched, and a researcher named Heyman, he's with Rebel Fund, counted robotics, manufacturing, hardware, energy, defense, logistics, all lumped together as industrials. Twenty-three percent.

GrantIs that a lot?

MilesThe five-year average for that category is six point six percent. This is the highest share he's tracked.

GrantWait, so it basically tripled overnight?

MilesBasically, and here's the flip side. SaaS, the dashboards and subscription stuff, dropped to twelve percent. The historical average there is twenty-eight.

GrantSo software didn't just plateau, it got cut in half.

MilesCut in half roughly, yeah.

GrantOkay, that's not noise. That's a real move.

MilesAnd industrials is doing a lot of work as a label. That's six different kinds of company getting rolled into one bucket.

GrantRight. It's not just drone makers. It's anyone building an actual physical thing, whether that's a factory line or a power system or hardware for a warehouse.

MilesSo the label's broad, but the direction's still the same, away from pure software.

GrantAway from pure software towards stuff you can actually hold in your hand.

MilesThere's a piece from a writer named Verma that frames the whole Demo Day around exactly that, hard tech and defense pulling ahead of the usual AI software story.

GrantRight, which lines up with what we were saying in the open.

MilesThere's a second analysis that gets more specific. Same twenty-three percent figure, but it says that's up from twelve point eight percent last batch.

GrantSo it wasn't a batch blip, it doubled batch over batch.

MilesAnd it names names. At least eight companies in this batch building precision munitions, drone swarms, counter-UAS hardware.

GrantCounter-UAS meaning?

MilesAnti-drone systems, stuff that shoots other drones out of the sky.

GrantHuh.

MilesAnd that's just the ones explicitly building weapons systems. The broader industrials number includes plenty more than defense. It's the whole category rolling up together.

GrantSo defense is the loud part of the story, but it's riding on a much bigger wave.

MilesExactly. Defense gets the headlines because it's dramatic. The manufacturing and energy founders are just as much a part of that twenty-three percent. They just don't get videos of drones getting shot down.

GrantFair point. Still wild that a batch that used to be almost entirely software now looks like this.

MilesSo when we talk about looking past the, you know, the AI-wrapper around a model pitches, that's what we mean. Founders building energy hardware, manufacturing lines, defense systems, physical stuff you can drop.

GrantNot a dashboard you log into.

MilesNot even close.

GrantOkay. Numbers are numbers. I wanna actually sit with one of these companies.

MilesYeah?

GrantOne of the Counter-UAS ones, ideally. I wanna know what it's like being four people trying to out-build people with actual defense budgets.

MilesI know exactly who you're thinking of.

GrantGood, because a number's one thing, but I wanna know if a team like that can actually compete, not just exist.

MilesThat's exactly the question we're about to test.

GrantBring 'em in. Let's see if the pitch holds up.

MilesOkay, so the company is Greypoint Industries, four people this year based in San Francisco.

GrantFour co-founders too, Nic Chu, Jordan Khoo, Steven Xu, Darcy Zhang, all from this year's batch.

MilesThat's a lot of names for a company that small.

GrantRight. Everybody's a co-founder when there's four of you and a lathe.

MilesSo what's the actual product? This is the part that got me. YC's own company page says the thing they've built is called LEGION. It's a drone swarm that finds enemy jammers and command posts by picking up the enemy's own radio signals.

GrantWait, so it's not shooting anything down. It's listening.

MilesThat's what YC's page says, yeah. It's a geolocation swarm.

GrantHuh. Because the press write-up I read on them called it something totally different, autonomous interceptors that shoot down attacking drones cheaper than the drones they're killing.

MilesRight. That's the tension. One source has them hunting jammers, the other has them as kill vehicles.

GrantThose aren't the same mission.

MilesNo, and I don't think we should pretend we know which one is the real product or if it's both.

GrantCould be both. Find the emitter, then hand off to something that ends it.

MilesCould be, but nobody said that out loud in either piece.

GrantFair. So we flag it and move on.

MilesYeah. What we do know is why they started it.

GrantGo on.

MilesJordan Khoo, one of the co-founders, quit General Dynamics to do this.

GrantHe left an actual defense prime to go build in a garage?

MilesBasically, and in a LinkedIn post, he frames the whole mission around Canada's Avro Arrow program. The government killed that fighter jet program back in nineteen fifty-nine.

GrantI remember that story. Canada had a genuinely advanced interceptor and just shelved it.

MilesKhoo's argument is that's the pattern repeating. Big institutions can design the thing and still fail to build it.

GrantSo his fix is what? Faster paperwork?

MilesNo. His words were first principles engineering, real factories, not another software layer on top of the problem.

GrantThat's a hell of a mission statement for four people to sign up for.

MilesIt is, but it also tells you why he left a company with actual production lines to go do this with a small team instead.

GrantBecause the big company wasn't moving.

MilesThat's his read on it, at least.

GrantOkay, so we've got one team, real product ambiguity, and a founder who quit a prime out of frustration.

MilesRight. And here's the thing that's nagging me. Greypoint isn't unusual in this batch. It's one of several teams telling almost this exact story.

GrantSame shape. Small team, hardware, a founder walking away from something bigger because it wasn't shipping.

MilesSo it's not one scrappy shop we happened to find.

GrantIt's a pattern nobody's actually said out loud yet.

MilesWhich means the real question isn't Greypoint, it's whether this was always going to happen. Remember our Nine Mothers episode, the founder who raised two hundred million and everybody treated it like some freak outlier?

GrantYeah, the one where we kept saying, "How did that even happen?"

MilesRight. I think we had it backwards. That wasn't a fluke, Grant. That was the front edge of this whole cohort.

GrantThat's a big leap from one company.

MilesIs it, though? Look at the numbers we just walked through. Industrial share, SaaS share, all of it moving the same direction at the same time.

GrantOkay, but a batch-wide shift and a two hundred million dollar check are different animals.

MilesThey're the same animal at different ages. Hard tech eats capital the way SaaS never did. No recurring revenue cushion, no ninety percent margins, just steel and machinists and burn.

GrantYeah. That's the part nobody wants to say out loud when they're pitching. You're not funding a product, you're funding a factory.

MilesThink about what that actually looks like month to month. You're not paying for another engineer's laptop, you're paying for a Batch of machined parts that might come back the wrong tolerance.

GrantAnd then you scrap the Batch and order again.

MilesRight. There's no rolling back a bad shipment of aluminum the way you'd roll back a bad deploy.

GrantThat's real money sitting in a bin somewhere, not a Git commit you can revert.

MilesAnd Greypoint isn't alone in that factory.

GrantWho else is in it?

MilesThere's a whole defense cohort this Batch. ISENGARD, Earendil Robotics, GUILD, Hop Aero, all building some version of the same thing Jordan's team is building.

GrantWait, Earendil too? I didn't clock that one.

MilesMhm. Precision munitions, drone stuff, counter systems, different shapes, same bat.

GrantSo it's not four kids with a CNC anymore. It's basically a sub-industry inside one batch.

MilesAnd it's not just one flavor of hardware either. Swarms, interceptors, munitions, sensing, all sitting under that same industrials umbrella.

GrantSo it's not one company getting cloned four times. It's a handful of different bets on the same underlying shift.

MilesThat's the reframe. Nine Mothers wasn't the exception that proves the rule, it was the rule showing up early.

GrantOkay. I'll give you that. But early doesn't mean right.

MilesNo, it doesn't.

GrantThese companies are burning cash on tooling, on materials, on testing hardware that either works or it doesn't. There's no soft landing where you pivot the UI and call it a win.

MilesAnd you can't beta test your way out of it either. A software bug you patch overnight. A structural failure in a drone chassis, you're grounding the whole unit until you re-qualify it.

GrantWhich eats weeks, not hours.

MilesWhich is exactly why I keep coming back to Jordan's line about real factories. If the capital's not there, the whole thing stalls on the shop floor, not in some Slack channel.

GrantRight. And investors know that, so either they've decided the margins eventually get there or-

MilesOr they're betting on defense budgets during the heavy lifting instead of gross margin.

GrantProbably both.

MilesProbably both.

GrantOkay, so here's my actual hangup. A story is one thing. Twenty-three percent of a batch is one thing.

MilesBut?

GrantBut none of that tells me if Greypoint specifically has customers, orders, anything real. A pattern is only as good as the proof underneath one company in it.

MilesSo does Greypoint actually have traction, Grant, or is this just a really good pitch with a compelling founder story attached?

GrantThat's exactly what I wanna know.

MilesSo I went and pulled Darcy Zhang's LinkedIn last night. Full nerd move, I know.

GrantOf course you did.

MilesSomebody's gotta check the receipts. But it's real information. A LinkedIn post from his team lays out that Greypoint's already got a contract with the Canadian government.

GrantWait, an actual signed contract, not a letter of intent?

MilesThat's what the post says. Signed contract, plus a demo with the Canadian Armed Forces where they geolocated an actual target with the aerial platform.

GrantA live target. Okay, that's not a slide deck anymore.

MilesRight. That's the part that stops you. Four people, one machine, and they're already flying with soldiers watching.

GrantWho's actually building this thing, though? Because a contract is only as good as the team executing it.

MilesThe same post breaks that down. You've got someone out of the Canadian infantry, somebody from General Dynamics, that's Jordan, we already covered him, plus a founder from Bosch Quantum Sensing, and one from TerraSense Analytics.

GrantQuantum sensing and geospatial analytics sitting next to an infantry background. That's an odd little roster.

MilesIt's exactly the roster you'd want if you're trying to find a drone in the sky and put a number on where it is.

GrantOkay, but I wanna slow down for a second. This is one LinkedIn post from one team.

MilesSure.

GrantIt's not a press release from the Canadian government. It's not a filing.

MilesDarcy's writing about his own company.

GrantYou're not wrong. Nobody's independently confirmed a contract value, the timeline, any of it.

MilesRight, and founders self-report wins all the time. I've watched people call a pilot program a partnership.

GrantSure, but a demo with an actual armed forces unit isn't something you just invent. You can fake a testimonial. You can't fake a country's military showing up to watch your hardware fly.

MilesFair, and honestly, even the skeptical read still lands somewhere good for them. Worst case, it's a credible pilot with a national government. That's still further than ninety percent of this batch gets by demo day.

GrantThat's basically my point. I don't need this to be a nine-figure defense contract to take it seriously.

MilesNo, but I wanna be honest about what we actually know versus what we're inferring from one post. What we know, a government engagement, a live demo, a founder team stacked with infantry, defense industry, and sensing backgrounds.

GrantWhat we're inferring is that it scales past one contract.

MilesRight. And either way, look at what year one means for these guys. It's not a dashboard shipping to beta users. It's a physical product in front of a military.

GrantWhich is a completely different clock than a SaaS company running on, and it's not one they get to reset. So if that's Greypoint's year one, what does that look like for the rest of this defense and hardware cohort we've been naming?

MilesSo zoom out with me for a second. Every year on this show, we've told the SaaS founders year one, free trial, first paying customer, board deck. This batch, that story doesn't cover most of what actually happened on the floor.

GrantBecause the press wrote the AI story. That's the piece everyone filed five days ago.

MilesRight. And underneath it, you've got hardware, defense, biome, fintech founders whose year one looks nothing like a dashboard demo.

GrantDifferent capital, different timeline, different everything.

MilesDifferent everything. Greypoint's year one is machining tolerances and a government contact, not AB testing a landing page.

GrantWhich is why four guys with a CNC machine are suddenly one of the hottest tickets in the Batch. Think about that for a second. A team that size building physical hardware, pulling the kind of attention that used to go to whoever had the slickest onboarding flow.

MilesIt's not about slick anymore.

GrantIt's about does the thing actually fly, actually geolocate, actually work when someone points it at a drone.

MilesIt's a bar is higher and lower at the same time.

GrantHow do you mean?

MilesHigher because you can't fake a part tolerance in a slide deck. Lower because nobody's asking you to prove product market fit with growth charts. They're asking you to prove the thing doesn't fall out of the sky.

GrantAnd that's a completely different kind of proof.

MilesCompletely different. So here's the part that doesn't get said out loud enough. What does a founder actually sign up for when they pick this path instead of SaaS?

GrantA different kind of investor, for one. The people who write checks for hardware already know the capital intensity going in. That's not news to them as it is to a first-time founder.

MilesRight. It's not the same crowd chasing monthly recurring revenue by month six.

GrantNo. And the fundraising math follows from that. Longer rounds, more of them before anyone knows if the thing works at scale.

MilesSo when a founder picks this over SaaS, and clearly this batch a lot of them did, they're not picking the easier path.

GrantNo. They're picking the one with a harder physics problem and a slower fundraising clock, and somehow more attention right now than the software pitches.

MilesIt's a strange trade to make on purpose, but it's the trade.

GrantEvery founder in that cohort is making it, whether they've done the spreadsheet math on it or not.

MilesEyes open or eyes closed, either way, that's the deal they signed up for.

GrantYeah. And that's not a bad place to leave it before we do the actual math on what it actually costs them dollar for dollar. Here's the number that sticks with me. If Greypoint's contract falls through, that capital doesn't come back as a lesson. It comes back as scrap.

MilesThat's brutal.

GrantIt's just the math. Doesn't mean they lose. It means the runway math is completely different, and nobody on that demo day stage was saying it out loud.

MilesAnd that's true for the whole cohort we named, not just them. ISENGARD, Earendil, GUILD, Hop Aero, every one of them is running that same scrap or scale bet right now.

GrantWhich is why I keep coming back to the twenty-three percent number. That's not twenty-three percent of a batch choosing the easy choice. That's twenty-three percent of a batch choosing the harder clock on purpose.

MilesSo do they make it?

GrantI genuinely don't know. Ask me in a year.

MilesFair.

GrantThat's the honest answer, and I think it's the right one to leave hanging. This show's whole premise is that year one doesn't resolve on a five-day timeline, no matter how good the demo day pitch was.

MilesYeah. Well, if you're out there and you know a YC founder living through their first year right now, hardware or otherwise, we wanna hear it. Write in, tell us what it actually looks like from the inside.

GrantEspecially if it's messier than the highlight reel.

MilesEspecially that. And if you liked this one, leave us a review. It genuinely helps people find the show, and it helps more of these hardware stories get told instead of just the AI ones.

GrantGenuinely.

MilesDon't steal my line.

GrantToo late.

MilesWe'll catch you next time.