This is the first of a series of (roughly) monthly insights coming straight from the deal room—a synthesis of the data Aexodus is seeing under the hood from private markets that most don’t.

Real labor gaps, exacerbated by long-horizon global birthrate declines, is central to our fund thesis, and teams who can bridge the gap with commercially-viable, intelligent physical automation integrations are the ones we back. This month, we are highlighting a market opportunity in one overlooked place: the shipyard.

Here’s how the traditional shipyard workflow has been done in the past for one ship (an up to 9-year project) and where robots can come in:

Three fundamentals:

• Labor shortage drives demand. There aren’t enough young laborers to replace the aging 40+ year-old shipbuilding laborers for the requisite specialized, dangerous, and unglamorous tasks like welding. That is fundamentally what’s underpinning demand for robots to buttress the labor gap today; not someday.

• Humanoids must start specialized; not generalized. High-skilled, high margin work needs specialization. Humanoids are touted as general-purpose machines, but that vision just isn’t possible today (from a technical feasibility angle—more on that in a future article).

• Threats of conflict drives maritime demand. Historically, the global shipping industry had been steady, dotted by short 3-4 year bursts of activity (1999-2001 and 2004-2008). With the closing of the Straight of Hormuz, however, this has spurred a bull cycle for the global shipping industry, with South Korea emerging as a winner amidst the chaos.

WHY A HUMANOID, AND WHY HERE

A shipyard is built around people: their reach and balance, the ladders, the scaffolding, the tight access. ABS makes the point itself that humanoids suit spaces designed for human workers, including confined and awkward spaces that fixed welding cells a la Path Robotics just can’t sufficiently service alone. This answers the usual objection to legs and hands, that a simpler robot or fixed assembly line arms would do. In a shipyard, moving and manipulating are all part of the job.

The bar to payoff is lower than it looks. A yard does not need a robot to replace a full-time welder’s entire job category. One that can run repeatable weld passes, carry inspection sensors, log usable data, or hold a hard position for part of a shift already unlocks more out of the existing skilled people in the workforce. The first win is simple: more output while incurring less injuries per welder.

THE WELDER SHORTAGE NO ONE CAN HIRE THEIR WAY OUT OF

The American Welding Society projects the country needs 320,500 new welders by 2029 just to keep up. That’s a gap hiring alone won’t close, in work that is dangerous, hard to staff, and rife with attrition.

GAO’s March 2025 Navy shipbuilding testimony says workforce and infrastructure limits are dragging on shipbuilding performance, and that yards are struggling to replace experienced skilled workers. CSIS’s December 2025 report goes further, describing a US naval shipbuilding enterprise that cannot produce at the scale, speed, or cost now being demanded of it.

THE BIGGER PICTURE

The macro market and geopolitical backdrop has created a tailwind for global shipping that wasn’t present just two years earlier. Shipbuilding saw a run when the Strait of Hormuz closed in early 2026. The IEA called the resulting oil-supply shock the largest disruption in the history of the oil market, where the resulting scramble for oil and ships shot VLCC (Very Large Crude Carrier) demand through the roof.

A US guided-missile destroyer can take up to nine years to deliver. Automation in this use case is one of the credible ways to scale capacity, cut build times, all while maintaining costs at par with traditional methods, even given the lack of laborers.

We see the market starting to fill out with startups addressing the demand. Blue Water Autonomy, Saronic, BlackSea, and Anduril are all in the water. Saronic has the most capital, the largest shipyard footprint, and the only large-scale Navy production contract among the list so far. The rising tide of US federal government demand is, invariably, raising all boats as it were.

CHINA BUILDS MOST OF THE SHIPS

In 2024 China took 53% of global shipbuilding output by tonnage and roughly two-thirds of all new orders, and remains the only major producer still expanding capacity. The gap is stark: over the past decade, Chinese yards delivered 6,765 commercial ships to the United States’ 37.

Chinese yards are sitting on close to four years of forward orders. By segment, Chinese yards took 81% of new containership orders and 74% of new tanker orders that year. The US does not buy from Chinese yards directly, but its supply chain exposure is dubiously wide open. About one in five of the goods imported into the US in 2024 arrived on a Chinese-built ship.

In a world teetering on the edge of conflict, that is a single point of failure. The COVID-era supply-chain snarls were only a preview of what happens if those routes are cut.

ENTER KOREA

South Korea is the world’s second largest shipbuilder by order value, and is the country the US is leaning on as a drop-in for China. Toward the end of 2025, as the industry experienced whiplash from the on-again, off-again tariffs, South Korean shipyards emerged as the winners as the market braced for punitive tariffs that President Trump might impose on Chinese ships calling US ports.

South Korea still owns LNG, holding 62% of new carrier orders in 2024 to China’s 38%, and its tanker orderbook is recovering after falling from 54% to 16% between 2021 and 2024. The Big Three, HD Hyundai, Samsung, and Hanwha, are all moving into the US naval and repair market, each having signed a US defense or shipyard partner in 2025. Hanwha buying Philly Shipyard is the clearest tell.

THE DARK HORSE

If the humanoid race is won in the 2030s by whoever achieves reliability, autonomy, generalization, and scale without running out of money first, the race, we anticipate, will rhyme with how frontier AI labs has unfolded: a concentration of North American winners and a slightly longer list of Chinese competitors, with Chinese EV companies as some of the frontrunners.

One team building directly for the shipbuilding labor gap is Persona AI, based in Houston, TX. It manufactures humanoid robots for heavy industry, by aiming toward creating the world’s largest repository of industrialized skills and tool usage like welding, grinding, painting, cutting, assembling, and other critical tasks. We would underwrite Persona as a picks-and-shovels play for the growing demand in shipbuilding as opposed to a pure humanoid play where narrative-driven valuations have far exceeded fundamentals.

The company’s financial projections are thus: a bill of materials (BOM) around $50k at scale against a robot-as-a-service (RaaS) price of $75k-$100k per unit a year. Persona reports their humanoids having a best-in-class 2kWh battery, good for over 4 hours of aggressive runtime and can last longer with lower grade work. They were designed for fast battery swaps where thermal performance is optimized for continuous operation.

Persona has signed an agreement with HD Hyundai’s shipbuilding group to deploy robots into their shipyards. Its second-generation prototype is currently welding and walking, and it is targeting first deployments in 2027, working with marquee partners including POSCO and ABS. The founders have built this class of machine before: CEO, Nic Radford, led NASA’s humanoid work, including the Valkyrie program, then founded Nauticus and Boardwalk Robotics, which got acquired by Foundation. CTO, Jerry Pratt, was formerly the CTO of Figure AI and a fellow NASA alumni to Nic during their tenure at Johnson Space Center together.

The US’ weak shipbuilding competitiveness score is the gap that Persona is built for. BRS scores US shipbuilding competitiveness at 46 out of 100, level with Vietnam, citing labor, supply chain, and design capability as variables in the scoring. That supply gap is the very chasm that Persona, Path Robotics, Saronic, and the MASGA policy push are all aiming to bridge.

WHERE THIS GOES NEXT

Morgan Stanley projects the humanoid market could reach $5 trillion by 2050, if more than a billion units are in use. China alone shipped roughly 90% of the world’s humanoids in 2025, but captured only 30% of the dollars because hardware will inevitably get commoditized.

The capital behind humanoids is betting on the directionality of where robotics is going as a whole. Figure AI, Skild, Apptronik, Agility, Neura, Physical Intelligence, and Boston Dynamics have all taken money from strategic industrials, Mag 7 companies, and AI and AI-adjacent VCs. Many of the backers like Toyota, Nvidia, Amazon, Samsung, SoftBank, Foxconn, Hyundai read like they’re hiring external robotics companies that may become acquisition targets. That matters because the late-stage outcomes here are likely to result in RaaS exits, M&A deals, secondaries offerings, or, in Agility’s case, SPAC IPOs.

The demand is coming from industrial buyers. The physical AI startups who can secure contracts with customers early (as in, now), deploy their hardware into those idiosyncratic environments, accumulate and compound the data collected on-site, integrate their software into legacy workflows, who ultimately achieve vendor lock-in are the teams we are backing to make it past the 10-year event horizon. Surviving beyond that point very probably means becoming the sector’s future incumbents.

For Aexodus, because we think of physical AI as a tool to augment the skilled labor gap that is currently below workforce replacement, exposure here means owning a slice of future GDP that the post-manual world is ultimately going to depend on.

This does leave us with an open question to ponder about stepping into the ensuing decade: could the US doubling down on physical AI stand to re-shore commercial shipbuilding domestically, or would that still not be enough to stop outsourcing its needs to countries like South Korea?

If any of this was useful, forward it to someone who’d find it interesting. That’s how the newsletter grows.

Invest in this space? We keep a smaller circle for people building and backing physical AI. Reply to this email, tell us who you are and we will be in touch.

Until next month,
The Aexodus team

_{Disclosure: Aexodus is an investor in Persona AI and Agility Robotics.}

_{This material is for informational purposes only and does not constitute an offer to sell or a solicitation of an offer to buy any security. Any offer will be made only pursuant to definitive offering documents. Investing involves substantial risk, including possible loss of all capital. Past performance is not indicative of future results.}

SpaceX’s IPO marks a structural transition in the space economy: from a market defined by a small number of government programs and speculative commercial ventures to one with a $1.77 trillion anchor company, a growing defense investment anchored by Golden Dome, and myriad active commercial developments of orbital data centers, in-space manufacturing, and autonomous in-orbit infrastructure.

The space infrastructure market, valued at $157.4 billion in 2026, is projected to grow to $373.7 billion by 2034. The enabling sub-markets, in-orbit servicing robotics, space-based compute, and new altitude regimes such as VLEO, are each growing at rates between 10% and 67% annually on current projections, reflecting early market formation dynamics rather than mature-market growth.

The capital allocation question the IPO raises is structural: as SpaceX transitions from a private vehicle to a public equity, institutional exposure to the broader space infrastructure stack through early-stage companies becomes the primary mechanism for accessing the pre-consensus formation phase of the orbital economy’s next layer of enablement.

With SpaceX entering the public markets, institutional attention is turning to the underlying infrastructure market the company helped create and to the segments of that market that remain in early formation.

The Space Economy: Current Scale and Growth Trajectory

The total global space economy was valued at approximately $626 billion in 2025, according to Novaspace’s twelfth annual Space Economy Report, roughly double its size a decade prior. The Space Foundation, using a separate methodology, put the global 2024 figure at $613 billion, and McKinsey and the World Economic Forum projected that the entire global sector will reach $1.8 trillion by 2035.

Government spending accounts for a substantial portion of this total. Global government space budgets reached $135 billion in 2024, with the United States responsible for approximately $79.7 billion, roughly 59% of worldwide government space expenditure. Defense applications represent the largest single segment of government spending: global defense space budgets reached $73 billion in 2024, a figure that is rising as multiple nations accelerate sovereign space capabilities.

Market Structure: A Word of Caution

The commercial NewSpace sector, including SpaceX, Rocket Lab, Planet, Maxar, and a growing roster of venture-backed startups developing launch services, satellite operations, and adjacent applications, are increasingly eating up a growing share of the space infrastructure market that’s been traditionally dominated by established primes like Lockheed, Northrup, and Boeing.

The in-orbit manufacturing market was valued at $1.21 billion in 2025 and is projected to reach $3.51 billion by 2030 at a 23.6% CAGR. The broader in-space manufacturing, servicing, and transportation market is projected to expand from $2.09 billion in 2025 to $5 billion by 2034.

Presently, while the markets appear bullish, Aexodus maintains a sober stance about the macro recessionary overhang. In market booms, SpaceX and its adjacent companies are built up and prosper. A rising tide raises all boats, as they say. When markets inevitably collapse, and they always do, the less prepared and the most financially exposed get wiped out. Space technology has extremely high upfront capital expenditures with decade-long horizons to profitability. Only a handful of startups will manage to survive a receding tide. Of the ones that do, the new incumbents attain a moat that few new entrants will ever cross.

It’s important to be mindful about making any space investment in the public markets at today’s retail values. We perceive the latter half of 2026 to represent a peak of the broader market cycle in actuality. We are instead observing this boom carefully and wisely, preparing to brave the repricing that lies ahead. The prescient question is: how far away is it, and how hard will it hit.

Defense Application: Golden Dome Initiative

The single largest new demand signal in the space infrastructure market is the Golden Dome missile defense initiative, established by executive order in January 2025 and seeded with an initial $24.4 billion through the 2025 reconciliation law. The program represents a strategic pivot toward space-based interceptors and persistent orbital sensing as the foundation of U.S. homeland defense.

The White House has pegged the program’s total cost at $175 billion, with the administration targeting a fully operational system in roughly three years. The Congressional Budget Office (CBO) has projected costs of roughly $1.2 trillion over 20 years, with the space-based interceptor layer accounting for approximately 70% of acquisition costs.

Orbital Data Centers Enable the AI Inference Layer

Beyond defense applications, the same space infrastructure can be applied toward dual-use in applications such as orbital data centers. The space-based data center market was valued at $1.28 billion in 2025 and is projected to reach $3.81 billion by 2034 at a 12.96% CAGR, according to Fortune Business Insights. Over 50% of planned satellite missions and space initiatives by 2030 are expected to incorporate robotic servicing or debris mitigation technologies, according to SNS Insider.

As of early 2025, 47 orbital computing payloads were in active development or had been launched, compared to just 3 in 2021. In January 2026, Axiom Space launched the first two orbital data center nodes to LEO, operating within Kepler Communications’ optical relay network with 2.5 Gbps optical links for Earth-independent processing. In December 2025, Nvidia-backed Starcloud successfully trained an AI model while in orbit using commercial H100 GPUs, the first empirical validation of orbital AI compute at commercial hardware specifications. SpaceX filed plans with the FCC in January 2026 for an orbital data center constellation of up to one million satellites. Venture capital investment in orbital data center startups has exceeded $1 billion cumulatively; in March 2026, Starcloud raised a $170 million Series A at a $1.1 billion valuation, making it the category’s first unicorn.

In-Orbit Assembly: A New Category of Spacecraft

Rendezvous Robotics, an Aexodus portfolio company, is pioneering a new category of spacecraft: intelligent modular systems that launch as individual tiles and autonomously assemble in orbit. Its spacecraft tiles integrate sensing, compute, power, control, autonomous docking, and electromagnetic assembly, enabling them to dock, self-correct, and form larger spacecraft platforms without human intervention.

The infrastructure enabling orbital data centers including power, thermal management, data relay, and physical assembly at scale, represents a distinct but critical infrastructure layer from the compute hardware itself. Low-latency data transport between orbital clusters and terrestrial networks is a critical dependency, as is the autonomous assembly capability required to build and operate structures at the aperture sizes needed for large-scale orbital compute.

The structural case for in-orbit assembly rests on a physical constraint: spacecraft are constrained by the launch vehicle’s fairing diameter. SpaceX’s Starship offers a 9-meter fairing, the largest commercially available. A thermal management radiator for a 100-kilowatt orbital data center requires the order of 100 square meters of surface area. The fairing constraint does not disappear with Starship. It shifts the threshold at which in-space assembly becomes necessary relative to pre-integrated launch.

Rendezvous is commercializing self-assembling spacecraft technology originally developed at MIT and advanced through multiple ISS demonstrations. The system has flown on two ISS missions, SpaceX CRS-20 in March 2020 and Axiom Ax-1 in April 2022. According to the company, those flights validated autonomous assembly over 30 days with more than 100 docking cycles, and error-correction work that improved docking success rates by 20 -30%

VLEO: The New Orbital Regime of Next Generation Space Infrastructure

Very Low Earth Orbit (VLEO), typically defined as 150 to 250 kilometers altitude, has historically been avoided for sustained operations precisely because of this constraint. Atmospheric density at 200 km creates continuous drag sufficient to deorbit a conventional satellite within weeks. Atomic oxygen at these altitudes corrodes exposed surfaces. Aerodynamic torques complicate attitude control. Traditional spacecraft address these challenges by carrying onboard propellant, which converts any low-altitude mission into a countdown to depletion.

The physics advantages of operating at this altitude, however, are considerable. For a sensor at only the 200 km band versus at the higher altitude band of 1,000 km, infrared target brightness increases by approximately ~20 times. Imaging resolution improves proportionally with altitude reduction for a fixed-aperture sensor. One-way signal propagation latency (measured) from 200 km in VLEO is approximately 10 ms, compared to 20-50 ms from 1,000 km LEO, 130-170 ms from medium Earth orbit (MEO), and 500-700 ms from geostationary (GEO) orbit. These differences are mission-relevant for time-critical applications including hypersonic threat tracking, fire control, and directed-energy cueing.

The Golden Dome architecture’s cost uncertainty is partly a function of orbital altitude. The CBO’s analysis of the space-based interceptor layer specifically notes that satellite lifetime in low-altitude regimes is constrained by atmospheric drag, making constellation replenishment a significant and ongoing expense. Cost aside, lower latency data transport between orbital compute and ground users is a potential commercial differentiator for latency-sensitive AI inference applications.

Vaxon Space, an Aexodus portfolio company, is developing an air-breathing electric propulsion (ABEP) satellite bus for the VLEO band. The company has filed provisional patents on its inlet and compression architecture. According to the company, its commercial wedge puts missile defense, Golden Dome-adjacent applications, and orbital data centers as the primary applications, with a $175 billion TAM, expanding to bus sales, ISR, AI connectivity, and low-latency relay applications as the platform matures.

ABEP is an emerging technology category designed to make sustained VLEO operations commercially viable. ABEP systems ingest residual atmospheric particles at orbital altitude, ionize them, and accelerate them electrically to generate thrust without carrying onboard reaction mass. The atmosphere that causes drag becomes the propellant source, enabling theoretical steady-state altitude maintenance in a regime that conventional propulsion cannot sustain.

ABEP as a category is not exclusive to Vaxon. ESA has funded ABEP research programs, and academic institutions including CU Boulder’s LASP and Rutgers are active in the VLEO environment. The technology is at an early readiness level across the industry, with no operational systems yet flown. U.S. Space Force officials have identified VLEO as a strategically relevant orbital regime. The service’s deputy chief science officer has said satellites operating below 250 kilometers could support a more resilient and survivable architecture, and congressional defense committees have directed the Space Force to explore commercial VLEO capabilities.

The VLEO and ABEP market has no established commercial incumbents with operational systems, representing an early-formation category. Adjacent academic and government programs have validated the physics; commercial operationalization remains in development.

Aexodus Capital is a seed-stage venture fund focused on physical AI, robotics, and sovereign infrastructure. Vaxon Space and Rendezvous Robotics are portfolio companies of Aexodus Capital Fund I. This article is for informational purposes only and does not constitute an offer to sell or solicitation of an offer to buy interests in any investment fund. Market data cited from Novaspace, Space Foundation, McKinsey, SNS Insider, Research and Markets, Fortune Business Insights, Grand View Research, Uptime Institute, Mordor Intelligence, and Breaking Defense. Past performance is not indicative of future results.