Drugs, data centres and satellite factories: where does the hype begin?

On 20 October, a spacecraft launched to investigate an experimental cancer drug in microgravity. Earlier in the month, Axiom outlined two computing nodes already working in orbit. Extraordinary news. Yet between a compelling demonstration and a profitable industry lies a distance that enthusiasm alone cannot close.

An orbital laboratory is not yet a pharmacy

Varda Space Industries launched its W-7 mission on 20 October to investigate how microgravity changes the crystallisation of a class of experimental cancer medicines called PROTACs. These molecules are designed to trigger the degradation of disease-related proteins. Some are hard to formulate because of limited solubility. Different crystallisation conditions may improve their properties.

The conditional language matters. A promising experiment is not proof of clinical benefit. An orbital crystal is not an approved medicine. Manufacturing feasibility must eventually be weighed against pharmaceutical quality standards, reproducibility, regulatory approval and the economics of producing a dose that can reach a patient.

On 30 September, Varda announced a $251 million Series D round, bringing its cumulative funding to $598 million. Its platform combines microgravity processing and return-to-Earth operations. That means spacecraft, launch, thermal protection, re-entry, recovery and quality systems. It is a substantial engineering proposition — and a substantial capital requirement.

For orbital pharmaceutical manufacturing to become a repeatable commercial market, three tests are unavoidable: a measurable product advantage over terrestrial alternatives, a reproducible process that meets industry standards, and a total cost compatible with the value of the resulting treatment. A laboratory breakthrough begins that journey; it does not complete it.

Microgravity has a price. Clinical evidence has another

Improving the crystalline structure of a molecule may be a significant scientific advance. It is not enough to create a medicine. The improvement must survive analytical testing, stability studies, reproducible batches, appropriate manufacturing controls and, where required, clinical evaluation and regulatory approval. Returning a sample safely from orbit is a meaningful mission milestone. It is not proof that an orbital pharmaceutical industry already exists.

The commercial question is even tougher. Could the same result eventually be reproduced on Earth at lower cost? How much usable material is recovered relative to the mass sent to orbit? What do launch, insurance, recovery, quality control and failed batches add to the unit cost? If microgravity is used only to discover a crystal form that can subsequently be manufactured terrestrially, that is a very different business from making commercial quantities in orbit. Both may have value. They carry different margins and risks.

Varda’s sequence of missions is a way to learn, retire technical risk and offer an experimental platform. Investors must still ask where the economic value ultimately accumulates: with the orbital operator, the developer of the active molecule, the patent holder or the business responsible for regulatory approval and distribution. Science opens a door. Commercial agreements determine who gets to walk through it.

Two orbital computing nodes are not a space hyperscaler

On 6 October, Axiom Space reported two operational orbital ARC computing nodes, deployed as part of a partnership involving Kepler Communications. It reported communication between nodes and with ground systems. Separately, Axiom demonstrated a post-quantum encrypted workload migration on the ground.

That is technically interesting. Processing information close to the satellite that gathers it may avoid some unnecessary transmission and help time-sensitive applications. It is best understood as a form of orbital edge computing. It is not evidence that the vast terrestrial data centres used to train AI models are about to relocate to space. The equipment, power consumption, customer economics and scale are fundamentally different.

In orbit, the most spectacular idea does not win by default. The one that survives physics, unit economics and a demanding customer’s second order has the better claim.

Vacuum is not free air conditioning

The case for orbital data centres is easy to sell: solar energy, abundant physical space and fewer local planning objections. Then thermodynamics intervenes. In a vacuum there is no air convection to remove waste heat; cooling depends heavily on radiative surfaces. Equipment faces radiation, launch stresses, limited repair options and the eventual need for replacement or disposal.

Boston Consulting Group’s August 2026 analysis found that orbital data centres could become technically feasible at scale over time while retaining a cost premium against terrestrial alternatives, even under favourable assumptions. Earlier in 2026, the European Parliamentary Research Service also identified economics as a central obstacle, alongside regulation and environmental concerns.

These are analyses, not definitive forecasts. They nevertheless offer a useful corrective. A feasible technology is not necessarily a competitive infrastructure business. The plausible early applications may involve processing data already generated in space, including secure or time-critical workloads, rather than replacing terrestrial cloud capacity wholesale.

Orbital computing contains two markets that should not be confused

Processing imagery close to the satellite that captures it can reduce the amount of raw data transmitted to Earth. An algorithm that identifies only relevant events may save bandwidth and shorten response times. That is edge computing: a plausible market with identifiable customer problems and a development path that can proceed in manageable stages.

Training terrestrial-scale artificial intelligence models in orbit is an entirely different proposition. It requires substantial electricity, large radiators, high-capacity interconnections, hardware replacement, radiation protection and the ability to move enormous datasets. The US Government Accountability Office warned in April 2026 that rejecting heat in vacuum is difficult and that large-scale systems remain unproven.

The relevant comparison is therefore not between an Earth server and an identical server supposedly cooled for free by outer space. It is between the full costs of two infrastructures, including maintenance and service continuity. Any orbital data-centre proposal should explain which customer problem it can solve more economically than a terrestrial alternative. Without that comparison, the word ‘cloud’ may simply describe a more distant place to incur familiar costs.

Satellite factories are an entirely different business

Back on Earth, the industrial story looks less cinematic and potentially easier to measure. Open Cosmos raised €300 million to expand production and satellite-enabled intelligence and connectivity services. EnduroSat secured $205 million to advance standardised platforms and higher-volume manufacturing. The basic product is tangible: spacecraft, subsystems, infrastructure and services bought by customers with specific missions.

Risks remain. Orders may be concentrated, programmes can slip, and factories consume cash before capacity becomes revenue. Nevertheless, investors can study indicators familiar from advanced manufacturing: assembly-line yield, testing time, defects, utilisation, schedule adherence, margins by contract and cash conversion.

These figures do not tell us whether microgravity will improve a drug. Nor do they predict whether an orbital data centre will be competitive. They tell us that three activities grouped under one glamorous heading have different business models. Treating them as the same asset class merely because they all involve space is a category error.

What NVIDIA’s satellite interest actually signals

On 19 October, Reflex Aerospace described a planned NVIDIA minority investment intended to support manufacturing expansion in Bavaria and onboard processing. Satellite-based AI has a straightforward rationale: select, compress or analyse sensor data before downlink, improving usefulness and reducing unnecessary communications for certain applications.

It does not follow that NVIDIA has built a global cloud data-centre network in orbit. A proposed investment, a prototype, an operational demonstrator, a signed sales contract and a completed transaction are separate milestones. Industrial reporting should keep those distinctions visible, especially when technological ambition is driving valuations.

Four tests that separate opportunity from a bubble

First, repeatability. Can the technological result be produced consistently, under meaningful operating conditions, and compared fairly with terrestrial alternatives? A one-off demonstration proves possibility, not industrial reliability.

Second, genuine demand. A memorandum of understanding is not an order. An order is not revenue. Revenue is not margin. And reported margin does not always translate into available cash.

Third, full unit economics. Count the cost of launches, power, insurance, ground stations, manufacturing, maintenance, re-entry where relevant and end-of-life disposal. A model that works only with implausibly cheap launches or permanent subsidies is not yet resilient.

Fourth, the capital path. How much equity is needed before a commercially repeatable product exists? Who can finance later stages? What is the route to exit? These questions are especially important for investment funds with defined lifetimes and liquidity obligations.

The real dividing line is the customer who comes back

These three industries require different kinds of capital. A satellite manufacturer can be judged on bookings, defect rates, factory yields and delivery performance. An orbital computing operator must prove the value generated by processing relative to the costs of hardware and connectivity. Space-based pharmaceutical work may involve scientific uncertainty, regulatory timelines and product adoption. Applying the same valuation method to all three would ignore the nature of their future cash flows.

Funders should look for an orderly progression: demonstration, repeatable second mission, paid order from an independent customer and eventually reproducible margin. Partnerships can speed up development; they do not constitute revenue in themselves. Public finance can support learning and remove some early technical risk. It cannot make customers purchase a product that fails to solve their problem.

Europe should not retreat from ambitious projects. It should finance each stage with an instrument appropriate to the uncertainty involved, rather than forcing private equity to behave like a research grant or venture capital to pretend it owns a mature industrial asset. The question separating revolution from bubble is not how far technology can travel. It is who pays to use it again, and why.

One label, three very different futures

Microgravity research may unlock valuable scientific advances, while significant clinical and industrial work remains. Orbital computing has credible specialised applications without necessarily replacing terrestrial infrastructure. Serial satellite production is already an observable industrial transition, but it will reward execution rather than slogans.

Capital should resist both reflexive cynicism and reflexive excitement. The task is to value different risks with different methods. A bubble begins when a prototype is sold as a market. An industrial revolution begins when customers come back and buy the product again.