Signal from the Edge: What Space Tech Expo Told Us About the Next Compute Frontier
- 20 hours ago
- 13 min read
By Nikki Liu │ ICM HPQC Fund │ June 2026
“Written by humans, please don’t blame the robots for our typos”
The conversation at Space Tech Expo was not about whether orbital compute is coming. It was about whether you'll be ready when it arrives.
Across panels, demos, and exhibition booths, a consistent signal cut through: the foundational questions of space-based data infrastructure – power, thermal rejection, radiation tolerance, launch economics – are no longer being theorised. They are being answered in real time, by companies already flying hardware and closing partnerships. The window, as Rama Afullo of Satlyt put it, is now. "If you're thinking about building something, please do it and do it now."
This sense of urgency did not emerge in a vacuum. It is the product of a fundamental shift in the economics of operating in the space.
Why Space Tech Expo Is Busier Than Ever: The Economics Behind the Momentum
The rapid decline in launch costs provides important context for understanding the level of commercial activity visible across Space Tech Expo. For decades, the economics of many space-based systems were constrained not by technical feasibility, but by the cost of transporting hardware into orbit. As shown in Figure 1, estimated launch costs to Low Earth Orbit (LEO) have fallen substantially across successive generations of launch vehicles, driven primarily by reusability and increasing launch cadence.
Launch cost functions as a tax on every kilogram deployed to space. Historically, that tax was high enough that many space-based business models remained economically marginal regardless of their technical merits. As launch costs fall, a broader range of applications – including communications systems, Earth observation platforms, in-space manufacturing, power infrastructure, and orbital compute – can be evaluated on commercial rather than purely technical grounds. Business models that previously failed economic scrutiny are increasingly being reconsidered. NASA analyses of life-support system economics have similarly found that lower launch costs can materially alter engineering tradeoffs, reducing the extent to which transportation cost dominates system-level design decisions.1
This shift was visible throughout the Expo. While launch vehicles remain foundational, much of the industry’s attention has moved to the technologies required once hardware reaches orbit: power generation, thermal management, communications, radiation tolerance, spacecraft subsystems, and software infrastructure. In this sense, much of the activity visible across Space Tech Expo can be understood as a second-order consequence of declining launch costs rather than an independent trend.
Figures 1 and 2 illustrate the economic shift driven by falling launch costs.
Much of the industry's optimism is tied to Starship. Current launch economics have already improved materially relative to previous generations, but Starship's stated objective of approximately 100 metric tons of payload capacity combined with substantially lower cost per kilogram would represent another order-of-magnitude reduction in transportation cost4. If achieved, concepts that remain difficult to justify under today's economics – including large-scale orbital power systems, space manufacturing, and orbital compute infrastructure – move materially closer to commercial viability. Whether Starship ultimately reaches its published targets remains to be seen, but its potential impact on the economics of operating in space explains why it was an implicit reference point in many conversations throughout the Expo.
The Central Constraint: Thermal, Always Thermal
If there was one sentence that unified every technical conversation at the event, it was this: thermal management is the first problem. If you haven't solved it, nothing else matters.
This is not an abstraction. In space, there is no atmosphere to carry heat away – heat ultimately has to be rejected by radiation. There are no fans, no HVAC. Every watt that enters a compute node as power must eventually exit as infrared photons through a radiator panel. At the power densities required for serious orbital compute workloads – we are talking hyperscale-adjacent infrastructure – this creates an engineering constraint that defines the entire system architecture.
Rob DeMillo, CEO of Sophia Space and a former researcher at NASA's Jet Propulsion Laboratory and MIT Lincoln Laboratory, laid it out starkly: "Those are your only two options. The first option is a giant thing with a lot of radiators. Option two is you get a whole bunch of CPUs, GPUs – tiny little satellites – and you throw those into orbit, you connect them all with laser optics. That's it." Either you build massive passive radiating surfaces, or you distribute the heat budget across a constellation where each node manages only a fraction of it.
Both paths are capital-intensive. Both require materials and system architecture that simply did not exist – at commercial price points – five years ago. The materials science has genuinely moved, with a generation of new materials finally replacing the 1960s-era components that have underpinned rocket science for decades.
For investors, the implication is direct: companies with proprietary thermal management IP – whether in advanced heat pipe design, novel radiator materials, or thermal-aware operating systems – occupy a structural chokepoint in the space compute stack. This is not a commodity layer.
The Tile Model: Modular, Distributed, Orbital
One of the most concrete architectural visions discussed was Sophia Space's modular tile approach to orbital compute – metre-scale server modules, each housing enterprise-grade processors, designed to be assembled into orbital data centres, passively cooled and modular so that failure of one tile does not cascade.
At a physical level, Sophia's tiles resemble a three-layer sandwich: solar panels on the front face capturing approximately 1.36 kilowatts of power coming in, four enterprise-grade servers packed into the middle converting that into 250-300 watts of usable compute, and a passive radiator on the rear face continuously shedding waste heat as infrared light into deep space – the same one-sided exchange that makes space lethal to an unprotected human works here in the tile's favour, radiating heat outward with almost nothing coming back.5 The components are commercial off-the-shelf where possible, radiation-tolerant rather than radiation-hardened, and designed for straightforward replacement and incremental scaling.
What makes this architecture compelling from an investment lens is not any single component – it is the system integration problem it represents. Getting off-the-shelf GPU silicon to operate reliably in LEO requires deep co-engineering between processor vendors, thermal specialists, power conditioning designers, and software teams writing operating systems that actively manage thermal headroom. That integration knowledge is hard to replicate and slow to commoditise.
Thermal: Solved at Current Power Levels
Among companies with flight heritage in the room, there was a clear view: thermal management at current orbital compute power levels is no longer the open question it once was. A common misconception is that space being cold makes cooling easy – in reality, space is a vacuum, and a vacuum conducts no heat away from anything, which means the cold is effectively irrelevant to the engineering problem.
No one put it more directly than Rob DeMillo. For Sophia's tile architecture specifically, thermal management is in the rearview mirror. The tile design – passive rear-surface rejection into the void, combined with OS-level thermal scheduling – has closed the problem at their operating power levels. "It's a belief that we solved," DeMillo said. "For us, it's in our rearview." That confidence carries weight coming from a team with hardware already in orbit.
The temperature environment in LEO is nonetheless far from benign. "It's not calm. There's no atmosphere for the weather...it's like a boat on the ocean. There are swells, there are surprise storms." You trade Earth weather for space weather. Solar flux variability, driven by the sun's eleven-year activity cycle and the unpredictability of solar flares, means the thermal load on an orbital system is never truly steady. Add the violent swing between orbital day and night, and atomic oxygen erosion at very low altitudes, and active management is non-negotiable – the hardware handles the baseline, the operating system handles everything the sun decides to do.
Radiation is a different matter. Unlike thermal, radiation does not become more manageable with better geometry or smarter software – how tractable it is depends heavily on mission profile, altitude, duration, and chip choice. At Sophia's current operating parameters it is manageable, but it is the less closed of the two problems, and it moves to the front as ambitions scale.
Radiation: The Constraint That Moved to the Front
Thermal and radiation are distinct problems and it is worth being precise about why. Radiation is not a heat management challenge – it is the space environment itself bombarding microelectronics with cosmic rays, solar proton events, and trapped Van Allen belt particles, causing bit flips, latch-ups, and long-term material degradation.6 It is why chips designed for Earth can fail in orbit, and it is an entirely separate engineering discipline from thermal management.
DeMillo noted that for Sophia's current tiles, radiation tolerance at their operating parameters is sufficient, but was careful not to universalise that. Rama Afullo, Founder and CEO of Satlyt – a former Starlink Product Manager at SpaceX who is building an open, decentralised software infrastructure layer for interoperable in-space computing – echoed this directionally: the industry is watching the radiation question closely, particularly as compute density increases and mission durations lengthen. The question of how far commercial radiation-tolerant silicon can be pushed – before mission profiles demand the more expensive and less performant radiation-hardened parts from the defence supply chain – remains an open question.
NVIDIA's launch of the Space-1 Vera Rubin Module earlier this year was a significant signal for the industry – not because the platform is immediately deployable at scale, but because of what it represents strategically.7 For decades, space compute has relied on processors originally developed for defence and aerospace applications, procured through specialised supply chains at defence-sector prices and timelines. Commercial silicon vendors largely ignored the orbital market because the volumes were too small to justify dedicated product development. NVIDIA's decision to introduce a platform specifically designed for space environments suggests that calculus may be changing. The launch signals a belief that orbital computing is becoming a sufficiently large and commercially relevant market to warrant purpose-built hardware from one of the world's leading AI computing companies. That is a form of market validation, not just a product launch.
The Viability Case: Launch Economics and Sustainability
Launch costs have fallen dramatically over the past decade and are widely expected to continue falling. The business models that do not close at today's prices may well close within the next five to ten years – and the companies building the enabling technology layers now are positioning themselves ahead of that inflection.
That dependency is currently concentrated to a remarkable degree: BryceTech estimates that SpaceX accounted for approximately 86% of global spacecraft upmass in Q1 2026.8 As a result, much of the industry's economic thesis remains tied to the performance, pricing, and execution of a single launch provider. That concentration is a real near-term risk. But the direction of travel is clear. Rocket Lab's Neutron program, Blue Origin's New Glenn, and emerging reusable launch providers such as China's LandSpace all reflect a broader effort to replicate the cost reductions unlocked by reusability.9,10 While none currently approaches SpaceX's scale or cadence, launch remains a critical cost input for space infrastructure, creating strong economic incentives for new entrants to compete on price and performance over time.
The sustainability case for orbital compute is more compelling than it first appears. The intuitive objection – that rocket launches are carbon-intensive – is true but incomplete. An orbital system pays much of its carbon cost upfront: launches happen, emissions are emitted, and then the system runs on solar power with no electricity grid draw and no water-intensive cooling infrastructure on the ground. A terrestrial data centre, by contrast, continues consuming electricity to power servers and cooling systems throughout its operating life. Recent studies have suggested that, under certain assumptions, this could reduce some of the environmental burdens associated with large-scale computing infrastructure. Whether that translates into a net sustainability advantage remains uncertain and will depend heavily on launch emissions, deployment scale, and operational lifetimes. If launch costs continue to fall and orbital infrastructure scales successfully, sustainability could emerge as an additional advantage alongside performance and economics rather than a constraint on adoption.
From Edge Processing to Orbital Cloud: Where We Actually Are
The most important thing to understand about orbital compute is that it is actually two stories on different timescales.
The near-term reality is edge processing. Satellites already generate enormous volumes of sensor data in orbit, but the bandwidth to transmit it all to Earth is limited, and for many applications the latency of a ground round-trip is simply too high. Missile tracking, maritime surveillance, autonomous systems – anything requiring immediate action cannot wait. The answer is to process the data where it is generated, on the satellite itself, and send only the result down. This is already happening: HPE and KIOXIA's Spaceborne Computer-2 has demonstrated onboard AI processing that reduced a sample Earth observation dataset from approximately 1.8 GB to 92 KB before transmission to Earth, dramatically reducing bandwidth requirements while accelerating analysis.11 The use cases are proven, and the urgency is real.
The customers writing checks for this today are not enterprises renting general-purpose cloud compute – that is the 2030s vision. The near-term paying customers are defence and intelligence agencies. The use cases are well-defined and funded: satellites collecting vast volumes of sensor data that need to be acted on faster than a ground round-trip allows – missile tracking, maritime surveillance, ISR, autonomous target recognition. Commercial Earth observation is a parallel market, with companies already selling processed satellite imagery to governments, insurers, and commercial customers, where on-orbit processing reduces latency and increases the value of the data product.12 The general enterprise cloud vision is real and will likely find its customers – but not yet. For now, the paying demand is concentrated in defence, intelligence, and commercial Earth observation, where the use cases are funded, the urgency is operational, and the value of faster, lower-latency data is already well understood.
The longer-term vision – orbit functioning as a general-purpose cloud platform where any company can rent compute the way they rent servers from AWS today – is a 2030s story. DeMillo's own roadmap puts Sophia's transition to full orbital data centres in the early 2030s. The technology, the launch economics, and the software abstraction layer all need to mature further before that model is viable at scale.
What bridges the two is a hybrid architecture that is likely to define the first generation of serious orbital compute: satellites preprocess and compress data in orbit, handling the time-critical and bandwidth-intensive work at the source, then stream the condensed output to ground for the heavier workloads. The hybrid model falls short of the full orbital cloud vision, but it is buildable today and addresses the use cases that matter most in the near term.
What makes this model interesting is that it requires space and ground systems to function as a single computing environment. The company that defines how that hybrid layer works – the protocols, data pipelines, and software abstractions that allow workloads to move seamlessly between orbital and terrestrial infrastructure – captures durable value. That problem remains largely unsolved and sits directly above the hardware layer that companies like Sophia are building today.
Investment Signals Worth Tracking
Several structural observations from the Expo are worth carrying into any early-stage deep-tech investment framework.
Thermal management is solved at current power levels – but will reassert itself as compute density increases. Sophia's tile architecture demonstrates that passive thermal management is achievable today. The next generation of orbital workloads will push power densities higher, and the thermal problem will return. Companies with proprietary solutions in radiator materials, heat pipe design, or thermal-aware OS scheduling are building IP in a layer that is structurally constrained and not obviously commoditised.
Radiation tolerance is becoming a commercial market. For decades, space-grade chips meant defence procurement – expensive, slow, and a generation behind commercially available silicon. That is changing. The shift toward commercial radiation-tolerant silicon, validated by NVIDIA designing explicitly for the orbital environment, opens a new class of investable companies outside the traditional aerospace prime structure. Fabless chip designers and IP licensors at this intersection are worth watching.
The software layer that ties orbital and ground infrastructure together is largely unbuilt. The companies that establish the protocols, data pipelines, and abstractions that allow orbital compute to function as part of a unified system – rather than a collection of isolated aerospace assets – are solving hard problems with strong network-effect characteristics. This is where the non-obvious entry points are likely to appear.
The business case depends on launch costs – and launch costs are falling. Business models that do not close at current launch economics are expected to close within five to ten years as the market becomes more competitive. The companies building in the enabling technology layers – thermal management, radiation-tolerant silicon, software infrastructure – are doing so ahead of a market inflection that has not yet arrived but is becoming harder to dismiss.
One note of caution worth holding alongside the optimism: the space industry has a long history of ambitious timelines that slipped. Companies have promised delivery windows they could not hold, and burned credibility in the process. What is different now is that the evidence is no longer theoretical – hardware is in orbit, partnerships are signed, and the materials science has genuinely moved. The question is not whether to believe the direction, but how hard to push on the timeline.
Whether orbital compute ultimately becomes a large standalone market remains an open question. What is increasingly difficult to dismiss is that many of the technical constraints that once kept the concept in the realm of speculation are now being actively addressed by commercial companies – and when Jensen Huang unveiled the Vera Rubin Space-1 module to 30,000 developers at GTC 2026, with Sophia Space among its six launch partners, that question moved materially closer to being answered. The opportunity may not lie in betting on orbital data centres directly, but in identifying the enabling technologies – thermal management, radiation-tolerant silicon, software abstraction layers – that become indispensable if the market develops.
Nikki Liu
Intern, ICM HPQC Fund
ICM Global Funds Pte Ltd
June 2026
Sources:
1. NASA analyses of life-support system economics have similarly concluded that declining launch costs can materially alter engineering trade-offs, with launch cost becoming a less dominant design constraint under Falcon-era transportation economics. https://ntrs.nasa.gov/api/citations/20230015110/downloads/Take%20or%20Make%20ASCEND%20charts.pdf
2. Historical launch-cost figures reported in 2017$ were converted to May 2026$ using the U.S. Bureau of Labor Statistics CPI Inflation Calculator to place all values on a common 2026$ basis. Available at: https://www.bls.gov/data/inflation_calculator.htm
4. Falcon 9 and Falcon Heavy figures are reported directly from SpaceX 2026 investor materials. SpaceX investor materials project approximately 100 metric tons of payload capacity for Starship V3 and target a 99%+ reduction in launch cost per kilogram relative to historical launch averages. The value shown (<$185/kg) is derived from SpaceX's stated cost-reduction target and represents forward-looking management guidance rather than demonstrated operational performance. Available at: https://content.spacex.com/cms-assets/FINAL_Documents%20and%20Updates/SpaceX%20IPO%20Roadshow.pdf
5. Rob DeMillo, CEO, Sophia Space. Panel discussion at Space Tech Expo USA 2026.
10.https://www.reuters.com/science/how-landspace-became-spacexs-biggest-chinese-challenger-2025-12-03/
13.Samtec FireHawk ruggedized optical transceiver platform photographed by the author at Space Tech Expo USA 2026. FireHawk optical transceivers are designed for high-speed data communication in demanding aerospace, defense, and industrial environments. Product information available at: https://www.samtec.com/kits/optics-fpga/firehawk/
Important Note:
The information in this article should not be considered an offer or solicitation to deal in the ICM HPQC Fund (Registration number T22VC0112B SF003) (the “Sub-fund”). The information is provided on a general basis for informational purposes only and is not to be relied upon as investment, legal, tax, or other advice. It does not take into account the investment objectives, financial situation, or particular needs of any specific investor. Investors should seek relevant professional advice before making any investment decision. The information presented has been obtained from sources believed to be reliable, but no representation or warranty is given or may be implied that it is accurate or complete. The Investment Manager reserves the right to amend the information contained herein at any time, without notice. Investments in the Sub-fund are subject to investment risks, including the possible loss of the principal amount invested. All forms of investments carry risks, including the risk of losing all of the invested amount. Investors should read the prospectus before deciding whether to acquire the units in ICM HPQC Fund. The value of investments and the income derived therefrom may fall or rise. Past performance is not indicative of future performance. This document is intended solely for institutional investors and accredited investors as defined under the Securities and Futures Act 2001 of Singapore. The whole or any part of this work may not be reproduced, copied or transmitted or any of its contents disclosed to third parties without ICM Global Fund’s express written consent. This advertisement or publication has not been reviewed by the Monetary Authority of Singapore.
ICM HPQC Fund a registered Sub-Fund of the ICMGF VCC (the VCC), a variable capital company incorporated in the Republic of Singapore. The assets and liabilities of ICM HPQC Fund are segregated from other Sub-Funds of the VCC, in accordance with Section 29 of the VCC Act.
Get to know us better:
Comments