🚀💥 The Case for Space-Based Compute

Happy Monday! This week’s deep dive explores the case for space-based compute, examining why orbital data centers are being considered, how power, cooling, and radiation challenges are addressed, and what near- and mid-term applications might look like.

In our spotlights, we highlight IQM’s $320M Series B, one of the largest rounds in quantum, and a SemiAnalysis report on AWS’s Trainium2 buildout with Anthropic.

The headlines cover semiconductor policy and chip strategies, new advances in quantum and neuromorphic, data center and cloud shifts.

This week’s readings cover AI chip design, U.S. semiconductor timelines, progress in quantum error correction and photonics, neuromorphic breakthroughs, and rising data center power demand (+ two interesting videos).

Funding news was highly active with 14 rounds, ranging from a €350K photonics round to Anthropic’s $13B Series F, with several landmark raises across quantum, AI, and data centers.

In our two bonus sections, we delve deeper into a semiconductor industry outlook from PwC and share a set of additional reports focused on market sizes across computing technologies - because we know, you love them.

Another highlight: We’re bringing computing and defense together at next week’s Defense x RISC-V Breakfast in Munich, co-hosted with TUM Venture Labs, Ubitium, and the European Defense Tech Hub. The discussion will center on how RISC-V architectures can power innovation across critical defense applications. Sign up here.

As Earth-based infrastructure reaches its limits in terms of power, cooling, and land use, space is emerging as a serious candidate for next-generation data center deployment. Orbital platforms promise continuous solar energy, natural radiative cooling, and fewer permitting constraints, but they also pose challenges.

Rationale for Space Deployment

Power availability: Space offers uninterrupted solar energy, especially in sun-synchronous orbits where solar capacity factors can exceed 95 percent. This is far above the ~24 percent typical for solar farms and allows for dense, energy-hungry compute clusters without relying on fragile terrestrial grids.

Cooling environment: Without convection or conduction, all excess heat must be radiated into space. Peak heat rejection is limited, but passive cooling through large deployable radiators is highly efficient. Combined with immersion cooling of the electronics, this allows thermal management at high density.

Isolation and regulation: Orbital infrastructure avoids many Earth-based constraints such as land acquisition, zoning laws, water usage, and grid connection delays. Physical separation also adds resilience in the face of natural disasters or geopolitical risks. However, operators must still comply with orbital debris regulations and space traffic rules.

Physical Architecture

Power systems: Solar arrays in sun-synchronous orbit generate near-continuous energy. Batteries cover eclipse periods, and power conditioning units ensure stable delivery to compute modules.

Thermal systems: Servers are immersed in conductive fluids, with heat transferred to large radiators that passively emit it into space. The main bottleneck is scaling heat rejection at high compute density.

Radiation tolerance: Instead of heavy shielding, systems use software-based fault tolerance, redundancy, and passive protection via cooling fluids. Off-the-shelf GPUs are made space-ready through system-level adaptations.

Network Architecture

Network performance determines whether orbital compute can participate in real-time or data-heavy workflows on Earth. High-speed connections are required to move data between satellites and terrestrial infrastructure efficiently.

Most orbital data centers use laser communication terminals to establish direct optical links with Earth or with satellite relay networks. These links can provide:

• Throughput of up to 100 Gbps per link, with potential for more throughput through parallelization

• Latency as low as 50 milliseconds under optimal conditions

Laser links are favored over traditional radio-frequency systems because they offer higher bandwidth, lower latency, and reduced risk of signal interference. Operators typically connect to commercial satellite constellations such as Starlink, but other systems like Amazon Kuiper or OneWeb may also be used.

Future Outlook

Near term: Orbital data centers will act as edge compute nodes for satellites, not replacements for Earth-based clusters. The main workloads are inference and pre-processing of sensor, hyperspectral, and radar data. Running these tasks in orbit reduces terabytes of raw downlink to megabytes of insights, saving bandwidth and speeding up delivery. Defense, Earth observation, and climate monitoring will be the first customers.

Mid term: As high-throughput laser links stabilize and thermal systems improve, orbital compute could expand beyond satellite-only workloads. This would enable Earth-directed applications such as large-scale inference, low-latency analytics, and eventually even training on space-based GPU clusters.

Achieving this requires not just technical progress but also further reductions in launch costs, since multi-gigawatt clusters remain uneconomical under today’s deployment economics. This will take some more time.

If you would like to learn more about orbital data centers, read our interview with Starcloud co-founder and CEO Philip Johnston. Starcloud has already raised $21M, one of the largest-ever seed rounds post-YC.

Other companies exploring data centers in space include Axiom Space, Sophia Space, and ADA Space.

⚛️ IQM raises record $320M Series B to cement Europe’s place in the quantum computing race (Tech.eu)

“Today IQM Quantum Computers announced that it has raised $320 million in venture capital, the largest Series B raise ever in the quantum space, both in Europe and outside of the US. It brings the company’s total funding raised to date to $600 million.
[… ]
I spoke to Dr Jan Goetz, co-founder and co-CEO of IQM Quantum Computers, to find out more.
[… ]
‘Our key differentiator is error correction. We’ve developed a path that’s about ten times more hardware-efficient than surface codes used by others like Google. We’ve also invested heavily in product maturity. Our systems are standalone, self-calibrated machines designed to fit seamlessly into supercomputing centres with a standard 19-inch rack footprint. That makes them very competitive globally.’”

🦾 Amazon’s AI Resurgence: AWS & Anthropic’s Multi-GW Trainium Expansion (SemiAnalysis)

From one of our favorite sources for deep technical and market analysis: SemiAnalysis takes a detailed look at Amazon’s bet on Trainium2, its in-house AI training chip, and its partnership with Anthropic, which has emerged as AWS’s anchor customer. The report highlights AWS building over 1.3 GW of datacenter capacity for Anthropic, set to host the world’s largest cluster of non-Nvidia AI chips.

While Trainium2 lags Nvidia’s GB200 in FLOPs and memory bandwidth, it shows a memory bandwidth per TCO advantage that fits Anthropic’s reinforcement learning roadmap. SemiAnalysis notes that Trainium2 is converging toward an Anthropic custom-silicon program, making Anthropic one of the only AI labs, alongside Google DeepMind, benefiting from tight hardware–software co-design.

Last week’s headlines covered government moves and new chip strategies in semiconductors, advances in quantum and neuromorphic computing, shifts in data centers and cloud, and another multibillion-dollar AI valuation.

🦾 Semiconductors

Why the US government is taking a stake in Intel (TechCrunch)

Congress Considers Forcing Nvidia to Sell Leading GPUs to Americans First (PCMag)

TSMC, like South Korean rivals, has US fast-track export status for China revoked (Reuters)

Europe enters the exascale supercomputing league with ‘JUPITER’ (European Commission)

OpenAI is reportedly producing its own AI chips starting next year (Engadget)

Tesla Dojo: The rise and fall of Elon Musk’s AI supercomputer (TechCrunch)

Green Critical Minerals’ VHD graphite proves superior in thermal management (The Mercury)

⚛️ Quantum Computing

U.S. and Indian investors launch $1 billion-plus India Deep Tech Investment Alliance (The Quantum Insider)

Norma completes quantum AI algorithm validation on NVIDIA (The Quantum Insider)

Quantum Circuits integrates with NVIDIA CUDA-Q to advance creation and testing of first quantum applications based on dual-rail qubits (The Quantum Insider)

IonQ breakthrough in synthetic diamond materials accelerates quantum networking scale and production (Business Wire)

8 female founders to watch in quantum (Sifted) (Paywall)

Researchers demonstrate new control of magnon-photon hybrids for quantum tech (The Quantum Insider)

Researchers Achieve Universal Quantum Computation by Introducing ‘Neglectons’ to Ising Anyons (Quantum Zeitgeist)

🧠 Neuromorphic Computing

BrainChip’s Cloud Platform Lets Developers Use Its Neuromorphic Tech (All About Circuits)

Accelerating AI: scientists combine the most accurate neuron simulator with DRAM (ITC.ua)

💥 Data Centers

Infineon and Delta collaborate on high-density power modules to accelerate data center power architecture (Infineon)

OpenAI plans 1 GW India data centre in Stargate AI push (Data Centre Magazine)

☁️ Cloud

SAP to invest over €20 billion in sovereign cloud in Europe (CNBC)

Cloud provider Lambda may be gearing up for an IPO (TechCrunch)

Microsoft says Azure affected after cables cut in the Red Sea (TechCrunch)

🤖 AI

Mistral, the French AI giant, is reportedly on the cusp of securing a $14 B valuation (TechCrunch)

The reading list covers AI in chip design, novel materials and timelines in semiconductors, advances in photonic and quantum computing, neuromorphic breakthroughs, and rising power demands in AI data centers.

🦾 Semiconductors

AI’s Value in Chip Design Depends on Data Availability (SemiEngineering) (13 mins)

A timeline of the U.S. semiconductor market in 2025 (TechCrunch) (13 mins)

Miniaturising semiconductors: IIT-Gn develops nanosheets as insulators (Times of India) (7 mins)

⚛️ Quantum Computing

Preparing for the Quantum Computing Age (SemiEngineering) (Video - 24 mins)

Quantum error correction: computing qubits requires tens of thousands (fault-tolerant) (Quantum Zeitgeist) (9 mins)

⚡️ Photonic / Optical Computing

Silicon Photonics and Co-Packaged Optics at the Heart of Next-Generation AI-Driven Data Infrastructure (Yole Group) (4 mins)

Analog optical computer for AI inference and combinatorial optimization (Nature) (23 mins)

Integrated lithium niobate photonic computing circuit based on efficient and high-speed electro-optic conversion (Nature) (22 mins)

🧠 Neuromorphic Computing

DNA-based Neural Network Learns from Examples to Solve Problems (Nature) (30 mins)

A metal-organic framework neuron with dopamine perception (EurekAlert!) (3 mins)

Neuromorphic Research Slashes Energy Use in Machine Vision (MVPro Media) (3 mins)

💥 Data Centers

AI data centres will drive a 165% power demand: Explained (AI Magazine) (7 mins)

💾 Memory

Challenges In Stacking HBM (SemiEngineering) (Video - 15 mins)

A very active week with rounds across the full spectrum, from a €350K Venture round in photonics to Anthropic’s $13B Series F. Activity spanned early Pre-Seed to multi-billion later-stage financings, with several landmark raises in quantum, AI, and data centers.

Last week PwC published its Global semiconductor industry outlook 2026 report. The paper provides a strategic outlook on the $1T semiconductor industry of 2030, with implications for capital allocation, geopolitical exposure, and long-term growth. It dissects demand drivers across end-markets, highlights supply chain realignments, and identifies future technology inflections.

Key takeaways:

• The semiconductor market is projected to grow from $627B in 2024 to $1.03T by 2030 (8.6% CAGR).

• Servers & networking chips (11.6% CAGR) and automotive chips (10.7% CAGR) will be the fastest-growing segments, fueled by AI data centers, EVs, and autonomous driving.

• AI accelerators are expected to account for ~50% of all data center chips by 2030.

• China is on track to become the world’s No. 1 in semiconductor capacity by 2030, with Asia-Pacific dominating manufacturing and packaging.

• New demand drivers include AR/VR, personal robots, renewable energy, medtech, and defense applications.

Source: “Global semiconductor industry outlook 2026” (PwC, 2025)

Several new market size reports dropped last week, ranging from explosive neuromorphic growth forecasts to strong gains in semiconductor sales, foundries, and advanced packaging.

Global semiconductor sales increase 20.6% year-to-year in July (Semiconductor Industry Association)

Global semiconductor equipment billings increase 24% year-over-year in Q2 2025 (SEMI)

Top five wafer fab equipment manufacturers’ revenue up 20% YoY in Q2 2025 (Counterpoint Research)

2Q25 Foundry Revenue Surges 14.6% to Record High, TSMC’s Market Share Hits 70% (TrendForce)

Advanced packaging market set to reach $79.4 billion by 2030 (Yole Group)

Neuromorphic Computing Market to Reach USD 13,596.67 million by 2032, Growing at a CAGR of 85.93% (OpenPR)

Neuromorphic Sensors Market Size and Share Forecast Outlook 2025 to 2035 (Future Market Insights)

❤️ Love these insights? Forward this newsletter to a friend or two. They can subscribe here.