Newsletter 043

👋 It’s Crusey — it’s been a minute. The word of the year is gigawatt. Everyone's talkin' energy infrastructure — for datacenters, for fabs, for the grid.

For me, the questions have become: What does power look like when you want your datacenter in orbit? What does this mean for today’s sun-harvesting stack? And what do we need to build?

I've been in solar since 2008. I had a front-row seat at DTE Energy in Detroit as the great terrestrial solar boom took off — and then watched 🇨🇳’s economic strategy crater the market in slow motion.

So when all of the SpaceX orbital datacenter news hit, the first thing I thought about was solar cells.

It’s time we chat about the sun-harvesting stack, precursor materials, manufacturing throughput, and $/W...

👇 So let’s do it.

IN THIS WEEK’S EDITION:
☀ Monitoring the “Space Power” situation
🚢 The non-oil commodity hostages of Hormuz
🎖 Drone Dominance & Gauntlet Round 1 results
📰 Robotics holy war, “weird space stuff,” packaging
🏰 Wins for allied science & The arena of legends

Forwarded this email? Subscribe to Per Aspera here.

Earth installs the equivalent of all orbital solar power every hour and a half. The combined solar capacity of every active space system is smaller than a single mid‑size solar farm in West Texas. And SpaceX already accounts for, let’s say, ~90% of that orbital total.

AI has dramatically expanded the Overton Window in the last 6 months, to the point that space-based datacenters have been normalized among the powers that be in Silicon Valley, the space industry, and everywhere in between. Discussion of the challenges (thermal, Kessler, etc.) is fairly well-saturated today – including in Per Aspera’s own pages (see our #2 most popular Antimemo, Realities of Space-Based Compute, published May 2025, well before the current craze kicked up).

But what’s largely escaped the online food fights conversations about “space datacenters” and associated constraints (thermal, Kessler, etc.) is the sun‑harvesting stack: cell chemistry, precursor materials, manufacturing throughput, and $/W.

…and from Q1 2026 alone, there’s a lot to report back on.

Back at Davos, Elon announced that SpaceX and Tesla would each target 100 GW/yr of U.S. solar cell manufacturing within three years (for reference, the U.S. installed 43 GW last year). Around the same time, SpaceX filed with the FCC for a 1M-bird space-datacenter constellation, and revealingly, sent a delegation to China’s solar belt, visiting with the world’s largest producers and inspecting heterojunction and perovskite lines. Finally, SpaceX is hiring…a lot…for a solar cell factory in Bastrop, TX.

Your friendly neighborhood VC, Per Aspera cofounder, and coauthor here — Jeff Crusey — has been involved in solar since the 2000s, way before it was cool and costs had collapsed 90%. So today we put pen to paper and take a closer look at the fundamentals of (literal) space power:

Unless you’ve been living under a shipping container, you know a closed Strait of Hormuz puts a huge damper on global oil and gas supplies. Less appreciated are all of the other commodities being blockaded. To wit:

The modern chip is one of the most extraordinary things humans make, and it is also one of the most globally interdependent, requiring exotic gases, rare metals, and specialized chemicals sourced from dozens of countries. So it knows a thing or two about force majeure and exogenous shocks. There was obviously Covid in 2020. Two years later, Russia’s invasion of Ukraine took half of the world’s semiconductor-grade neon offline within days (and prices had spiked 600% during Crimea in 2014). Both times the chip industry absorbed the hit. Fabs stockpiled, new sourcing came online, and the supply chain came out more distributed than it went in.

Today we see signs of the same: fragile in the moment, antifragile over time. TSMC says it doesn’t expect a near-term production impact. South Korean chipmakers sit on ~6 months of helium inventory. North American Helium is fast-tracking wells in Saskatchewan and building Canada's first liquefier. The Strait will likely reopen before long, as no party to this conflict benefits from a prolonged closure. But each crisis, Crimea to Covid to Ukraine to Hormuz, reminds us why we cannot run on single-chokepoint dependencies.

Gauntlet 1 results from the Drone Dominance competition are in. As a reminder, Drone Dominance is the DOW’s open-vendor competition to bootstrap America’s drone industrial base via power of the purse and successively larger POs. The $1.1B program runs across four rounds, expanding volume from 30,000 to 150,000 drones and cutting unit costs by 54%.

Per Aspera’s 2¢: More procurement should be done like this. Open the door to American (and allied) startups, field-deployed and battle-tested technologies, and tiny shops that nobody in DC has heard of. Hand the scoring to the warfighter (which is how it’s done in Gauntlet, with Army, Marines, and SOF operators scoring each entrant’s system after two hours of training). Take away the lobbying, incumbency advantages, and decade-long acquisition timelines.

Let the games begin. Up next, Phase 2 arrives in ~6 months with 50,000–60,000 more drones and a harder gauntlet: GPS denial, comms jamming, and a full EW environment. Oh, and Chinese-origin motors and batteries are banned.

…while we’re here: The Pentagon wants way more LUCAS drones, which as we wrote a couple weeks back, is a remarkable American success story. LUCAS is made by SpektreWorks, which until recently, could have been considered one of those “tiny shops that nobody in DC has heard of.”

001 / A Specialist-Shaped Problem… The U.S. Navy announced the service’s largest robotics deal to date on Tuesday, awarding a $71M IDIQ to Gecko Robotics (no affiliation with Warren Buffett's lizard-avatared insurer). Gecko designs specialized robots, sensor payloads, and software for industrial asset inspection. It hails from Pittsburgh, home to a dense cluster of robotics firms tracing back to CMU’s Robotics Institute, which sent cleanup bots into Three Mile Island, won the DARPA Urban Challenge, and helped put rovers on Mars. We’re seeing something like an intellectual holy war emerge in robotics — humanoids vs. specialized bots — with each camp hell-bent on proving their approach is the right one, and Team ‘Specialized Bots’ watching with dismay as humanoids inflect their way up the robotics hype cycle. Let this Gecko/Navy contract serve as a useful reminder that it’s unlikely to resolve cleanly in either direction. Yes the built world is designed for humans. Yes many automatable jobs are not human-shaped. Both can be true! In this case, magnetic wall-crawlers mapping corrosion on the hulls of 18 ships in the U.S. Pacific Fleet — i.e., Gecko’s specialized robots — will get the job done just fine.

002 / CALL A SPADE A SPADE… Typically, FCC filings are where space nerds go to unearth architectural details about new satellites/constellations (e.g., SpaceX’s aforementioned FCC petition & Starcloud filing this week for ≤88,000-satellite compute cluster), or where space lawyers go to duke it out over spectrum. In a most-welcome and very entertaining departure from this normal state of affairs, the FCC is circulating a draft Notice of Proposed Rulemaking for “Spectrum Abundance for Weird Space Stuff” (h/t Mach33’s Joseph Ibeh). This is real:

003 / PACKAGING RACE & RAISE … At GTC, the Super Bowl for AI, this week, Nvidia unveiled Vera Rubin: 288 GB of HBM4, 22 TB/s of bandwidth per GPU, rack-scale NVL72 "AI factories," and even a Space-1 module for orbital data centers, with up to 25× more inference compute than the H100, in orbit. Under the hood is advanced 2.5D “chip‑on‑wafer‑on‑substrate” (CoWoS) packaging — a physically demanding, specialized and hard-to-scale process of assembling huge AI chips and stacked memory. (TSMC in Taiwan is where most of that capacity sits; Goldman Sachs recently raised its TSMC target, partially due to their strength in packaging.) In Issue #038, we flagged a 🇺🇸 win with Micron climbing in global memory market share. And we’re here to flag two domestic bets worth watching as they become 🇺🇸 wins themselves — Amkor's $7B advanced packaging plant in Arizona and SK Hynix's $4B facility in West Lafayette, explicitly an advanced packaging facility for HBM in partnership with our friends at Purdue University and their Purdue Research Park — exactly the kind of industry-academia, national-interest plays we’re giving love to from afar.

CHEMISTRY FROM SCRAP HEAP
We recently flagged chemical processing as one of America's most under-discussed bottlenecks. Good news on this front! Oak Ridge National Laboratory has demonstrated that low-grade mine waste can be chemically upgraded into high-purity rare earth products using standard reagents. Working with simulated tailings from Missouri’s Pea Ridge iron mine — grading just 2.4% total REEs, well below the 5-10% yield you’d see in a typical commercial deposit — the team produced rare earth concentrates at ~92% purity of targeted elements. Still bench-scale, still simulated feedstock, and still open questions around scaling costs & real-world impurities. But a win’s a win. And this is a big step forward for American chemistry that has been badly in need of one!

STEM CELLS FROM THE LAB BENCH
Japan's health ministry has approved the world's first commercially available iPS stem cell therapies — one for Parkinson's, one for severe heart failure. The Parkinson's treatment, made by Sumitomo Pharma, implants lab-grown precursors of dopamine-producing brain cells to replace what the disease has destroyed. Current treatments only manage symptoms, whereas this goes after the underlying cellular loss. A second approval went to startup Cuorips for ReHeart, heart muscle sheets grown from iPS cells that restore function in failing hearts. Both are conditional (the Parkinson’s trial covered seven patients) and Japan’s fast-track pathway for regenerative medicine accepts thinner clinical evidence than the FDA would (feature or bug, depending on your priors). Japan is now the proving ground for an entire class of therapy that’s the closest current thing medicine has to a universal spare part — reprogrammable into nearly any cell type in the body. If these hold up, and ‘if’ is doing a lot of heavy lifting here, the implications would run far beyond these two very prevalent, and very devastating, diseases.

Lastly, in the latest addition to software giants who've stepped into the arena: Uber founder Travis Kalanick. Though as TK put it, he “never left.” After spending 8 years in stealth, he’s unveiled Atoms — formerly City Storage Systems and CloudKitchens — now targeting specialized, wheeled industrial robots across food, mining, and transportation. A win for the Renaissance that is currently stacking legends from ecommerce, search, AI: Jeff Bezos (Project Prometheus), Eric Schmidt (Relativity Space), Bob McGrew (Arda). “The tech stack for industrial progress machines is not for the faint of heart,” TK writes. Amen. And as we all know, the only way to the stars, Ad Astra, is through hardship, Per Aspera…