Newsletter 061

👻 Hi everyone, it’s Joy (cofounder here at PA… I know… I’m a ghost!). Filling in for our usual crew up here as I come bearing a mea culpa: we apologize for going dark the last few weeks. As a rule, we only skip for holidays, but in our defense — we had two!

① Ryan Duffy married his lovely, beautiful, hilarious bride Anna in an Austin-based Tuscan-style ceremony. Jeff/Branden/I were lucky enough to be there, to witness and celebrate the two. And, his company Array Labs just raised a new $21M round!

② Dan Goldin turned 86 — a perfect excuse to gather some of the PA community and celebrate 86 years of ‘setting the angel free’.

Send them your congrats! Also note that we’ll be off for the rest of August to prepare for a new-and-improved Per Aspera in Q4 and beyond.

Now I'll hand it to Dan who, during his b-day celebration, was thinking about the bottlenecks in AI chip production… 👇

Forwarded this? Join Per Aspera today.

GOLDIN HERE. I spend a lot of time (as many of you do, too) on AI’s power problem — both the enormous electricity demands posed by the technology, as well as the massive inefficiencies of current approaches. In search of what comes next, I’ve gone the distance on orbital solutions, photonics, and even exotic neurology-inspired semiconductor design.

But even if we solved the electron problem, we’d still be bottlenecked by flops (a function of how many chips the industry can actually produce). So I’ve been looking into this side of the equation, which starts in the Dutch town of Veldhoven. There you will find ASML, holder of one of the purest monopolies in industrial history. Every EUV machine on Earth — the machines which prints leading-edge AI chips at scale — comes from ASML’s assembly halls.

ASML’s Extreme Ultraviolet (EUV) machines are the most precise tools humans have ever built for printing circuits. They draw with 13.5 nm light — a 14× shorter wavelength than the 193 nm Deep Ultraviolet standard before it. Shorter wavelength means finer lines → smaller transistors → denser chips → more AI compute.

High-volume EUV mfg really only began in ’18–’19, laying the groundwork for today’s AI accelerators. Without EUV, we would have no Nvidia H100, no Blackwells, and no AI supercycle.

The thing is, each machine borders on impossible to build:

Assembly alone takes months, and ASML’s backlog already runs well into ‘27 for TSMC, SK Hynix, Samsung, Intel, and Micron.

ASML shipped 48 of these in 2025. ~50 machines a year is a brutal upstream constraint on new leading-edge AI capacity, worldwide. (With 65 planned for next year and 85 the following. And not one EUV machine has been licensed for export to China.)

But I must ask — does the question necessarily have to be: how many machines can we get out there?

Or is it: how much ultraviolet light for printing geometry on silicon can we generate?

Today’s answer to my question — how much light can we generate? — is: barely any.

How it works today: You take droplets of molten tin, each a third the width of a human hair, and fire them across a vacuum chamber 50,000 times a second. A laser strikes each droplet twice, once to flatten it, once to vaporize it into plasma 40× hotter than the surface of the sun. The plasma emits a faint flash of the required 13.5 nm light, and a collector mirror gathers what it can.

All in, less than 1% of the electricity feeding the source comes out as usable EUV.

An extraordinary system of physics, chemistry, and engineering — and a reminder of what babies we still are as creators of compute systems!

For decades, physics labs have made light with free-electron lasers (FELs), which accelerate electrons to nearly the speed of light and send them weaving through a magnetic array, where they radiate intense, coherent light — tunable to exactly the wavelength you need. Applied to chipmaking, this means far more usable EUV, no tin, no mirror-eating debris, and a source that sits outside the scanners (meaning one FEL can feed several machines at once).

This approach does not attempt to replace the printing machines themselves. Rather, it replaces only their weakest component: the light source, where improvement flows straight into output.

There’s a full-circle quality to this. EUV was born as an American project, with Intel, AMD, Motorola, and the Energy labs (Livermore, Sandia, Berkeley) proving the physics in 1997. Then, a Dutch company spent two decades industrializing the technology. The FEL is American-born, too: invented at Stanford in the ‘70s.

This time around, I have high hopes we can scale FEL at home. I see at least two guys marching in the right direction:

No FEL has yet run a production line, but my belief is when not if, and that the obstacle here is engineering and $$$. Meaningfully more light per machine means meaningfully more chips from our fabs’ existing install base — without needing to build a single new machine.

Now relieving the production bottleneck just returns the binding constraint back to energy. And, of course, two developments I’m obviously and overtly excited about (since we can’t seem to stop writing about them!):

You’ll see me going back and forth between the energy required to run AI compute and the actual manufacturing of the chips that make it possible. But if you’re working on the production side, I’d love to hear from you.

The team finally decided I’m responsible enough to have an email address 😂, so reach out! 💌 [email protected]

DAN GOLDIN’s 86th — Angel in the Marble. At PA, we spend so many, probably too many, of our brain cycles on how we can bring our community together in meaningful ways not done before — so on Fri Jul 31, Per Aspera threw its first house party 😛🍾 to celebrate Dan Goldin’s 86th lap around the sun.

Friends and family of PA flew in from all over the country and drove in through the LA traffic 🥹. Naturally, we had to make the theme: Angel in the Marble, celebrating 86 years of DG setting the angel free. Pls look up the attributed Michelangelo quote if this is new to you.

Thank you to everyone who came out — and special TY to those who've burned the midnight oil with Dan across the decades (and, for the last ten years, with Joy too) and gave reflective toasts. In true Per Aspera fashion, there were many:

Mark Albrecht — Executive Secretary of National Space Council under Bush 41 and hired Dan into NASA. You can thank or blame this guy; Leon Alkalai — met Dan as a twentysomething at JPL, now building Sophia Space together; David Dreier — an ally of Dan's on the Hill in the '90s and still cooking up schemes together as recently as last week; Darren Garber — CEO of Revolution Space and lucky him, Dan is his Chairman; Mike Fong — CEO of Privoro who is mentored and advised by Dan; Jordan Blashek — GP Overmatch Ventures / Chairman of Endless Frontiers / Co-founder America's Frontier Fund (!) and dear friend + collaborator of DG!

Of everything that came out of the event party 😛 — the big thing we learned was: celebrating someone's life is indeed special in itself. But getting the people you've burnt your hands with, fought battles with, lost with, and even better, won with — all in one room, to celebrate a lifetime of contribution to this country? That's something else — 10/10 highly recommend!

Special s/o to Out of This World Design for helping us bring this special event together.

001 / A GANGBUSTERS IPO… On Monday, China’s first mainland-listed humanoid-robot company — Unitree Robotics — saw its Shanghai IPO retail tranche oversubscribed by 5,526x at ~$9B valuation. In a category that has run for a decade-plus on the narcotic of VC money and viral demo reels, Unitree’s financial filings are a story of staggering, bureaucratic business boredom — it is a company that ships products, collects cash, and turns a profit! The company realizes luxury handbag-level margins (~60%) on robots, grew revenue 335% in 2025 to ~$250M, and has slashed its ASP (average selling price) 71% in 24 months while expanding margins 16 points.

002 / MARKET TALK… This (^) listing should make us stop and reread our note on American IPOs. In the Western model, an IPO is mainly an exit. In China's, it's an industrial-policy tool — policy priority → early capital → IPO liquidity → scale → national capability — mobilizing domestic risk capital around a strategic narrative (which, what do you know, typically coincides with a Five Year Plan-blessed critical technology). It also hands retail concentrated exposure to emerging technology through a different class of equity … no SPACs required!

At the outset of this movement, we, like many of you, were guilty of poking fun at financialization in favor of building hard things … but of course, just like atoms/bits, that’s a false binary. Doing the hard things tends to be capital-intensive. Our job as builders is to capitalize the work efficiently, then go faster-better-cheaper. The actual assignment is to use every tool at our disposal to do hard things, and win.

The original Arsenal of Democracy was financed by the public: 85M Americans bought war bonds. The modern version can draw on the deepest capital-formation machine on Earth: our public markets. Time to enlist them as part of our competitive arsenal!

003 / HANWHA BIDDING MORE SHIPYARDS… On Aug. 11, Hanwha Defense USA made a preliminary, non-binding offer of ~$1.1B to acquire Austal USA — the Mobile, AL shipyard — from its Australian parent. This adds to the ever-growing Korean partnership with the U.S. shipbuilding industry: Hanwha acquired Philly Shipyard for $100M in 2024 and is investing to modernize and expand U.S. production.

South Korea lost its position as the world’s largest shipbuilder by volume to China in 2010. In 2025, PRC yards received roughly 63% of global new-vessel orders, measured in compensated gross tonnage, vs. 21% for South Korean yards.

But Korea is particularly strong at ships that are difficult to engineer and build reliably:

Our naval industry power comes down to more than just hull count. Ultimately the equation we need for the foundation is:

Allied naval strength = ships + repair capacity + trained labor + supplier capacity + production speed.

001 / ANDREW + ISAAC (+RYAN) RAISE $21M FOR ARRAY LABS… Disclosure with a grin: this one’s ours. Array Labs — where your editor spends his days slinging space radar — closed an oversubscribed $21M strategic round anchored by Mitsubishi Electric, following a $20M S-A earlier this year. The two will work together to leverage Array’s radar technology and develop space-based ship- and aircraft-tracking products for APAC customers. We goin’ global! Congrats Andrew Peterson, Isaac Robledo, and of course our very own Ryan Duffy ++ rest of team on this killer win!

002 / OMAR GOES VC… Shout out to Omar Pimentel for your new role as Director of Frontier Strategy at Type One Ventures — thank you for your service at DIU 🫡. (Type One… you got a good one!)

003 / SPACE CAPITAL’s Q2 REPORT… For those who want to catch up on the state of the space economy — ie. where capital markets are rewarding companies now that SpaceX has completed its category-defining IPO — read Space Capital’s Q2 Report. (TY Chad Anderson and Brendan Guiney for sending this along!)

Branden here. I, like most of my military brothers and sisters, have been a big fan of the All-American, Made-in-USA tactical brand Crye Precision for years. Despite our stereotype, military folks actually care a lot about great design, taste, and quality 🙂.

I recently came across a photo of their HQ in the Brooklyn Navy Yard — have to say, it takes the cake:

You wouldn’t expect to find American special operators walking among the engineers and creatives who now work out of the revived Brooklyn Navy Yard. But inside this former shipbuilding factory, Crye Precision runs an 87k-sqft, vertically integrated design-and-manufacturing operation developing gear for our American (and allied) forces.

Crye Precision’s founders are an NYC-based art student and mechanical engineer — both Cooper Union graduates — who, c. ’99–’00, decided to build for American troops at a time and in a place where making things for the military was decidedly not in style. Initially, they had set out to improve the portable computers used for mortar-ballistics data (👋 former mortarman here) and eventually found their way to textiles.

When I joined the Army, we wore ACU (Army Combat Uniform) — the gray-green-tan pixelated camouflage.

The idea was to field one Army-wide pattern for almost any environment: desert, woodland, or urban terrain — simplifying procurement/inventory/deployment logistics. Sounded great… did not work. (The UCP pale gray+tan palette contrasted with the Afghan backdrop + the tiny pixels blended together at distance, leaving soldier silhouettes insufficiently broken up.)

Crye took a different approach: seven colors and irregular shapes at two scales. Broad color fields help the pattern read at distance; smaller shapes break up those fields up close — managing contrast and disrupting a soldier’s outline as terrain and light change. The result was MultiCam, which the Army designated the Operation Enduring Freedom Camouflage Pattern (OEF-CP) when it began issuing it to troops deploying to Afghanistan in 2010.

(Of course, none of this is easy — the Berry Amendment requires DoW-purchased textiles and clothing to be domestically sourced and manufactured, hence the beautiful, vertically-integrated Crye facility above ↑.)

Bottom line: we should be working with our American artists and designers more! (PS. @Crye Precision, we’d love to visit!)

¹ BTW: the Crye HQ is quite reminiscent of the TRW office space in Lyndhurst, OH (which was demolished in 2023, RIP).

Congratulations to AVIDrone, champions of the DARPA Lift Challenge, for achieving a 3.84:1 payload-to-aircraft-weight ratio over the Challenge’s five-nautical-mile course. If y’all remember, the Challenge goal was to get drones to a 4:1 payload/weight ratio, and while we didn’t exactly hit it — clearly we’re close and there’s a path here.

And big s/o to Phillip “Donna” Smith (U.S. Army Reserve lieutenant colonel, now in DARPA's Tactical Technology Office) and his team, who architected and led the program. They successfully turned the heavy-lift drone problem into a public sport: live runs, visible scoreboards, commentators, field reporting, recaps, highlight reels, and engineers crashing out arguing online in real time.

ESPN for dual-use products — genius! 🤌