Venture capital spent two decades optimising for a specific type of company: software-based, asset-light, scalable at near-zero marginal cost. The canonical success stories — Google, Amazon, Salesforce, Airbnb — demonstrated that writing code could be the path to trillion-dollar enterprises. Network effects create winner-take-all dynamics. Gross margins approach 80 percent. Capital requirements are modest relative to revenues generated. The logic was compelling, and for a time the returns were extraordinary.
At Lumino Capital, we invest in a different kind of company. Our conviction is that the deepest scientific and engineering problems — commercial nuclear fusion, photonic quantum computing, AI-driven drug discovery, long-duration energy storage — represent not just the most important technical challenges of our era but the most attractive investment opportunities available. The characteristics that make these companies difficult — long development timelines, high capital intensity, genuine scientific risk — are precisely the characteristics that create durable, irreproducible competitive moats.
This piece sets out the full architecture of our investment thesis. It draws on the publicly disclosed trajectories of companies like PsiQuantum and Commonwealth Fusion Energy to illustrate what patient, science-led investing looks like at scale, and why seed-stage entry into these trajectories is where the highest-quality returns in venture capital are currently being generated.
What Deep Technology Actually Means
The term "deep tech" is applied loosely — sometimes to any startup with a technical component, sometimes to anything that involves a PhD founder. We use it precisely. For Lumino Capital, a deep technology company has three defining characteristics that together determine investment quality.
The first is a genuine scientific breakthrough at its core: not an incremental improvement on existing technology, not a clever recombination of known approaches, but an advance on the state of the art that required years of research to achieve and that represents knowledge genuinely difficult for competitors to replicate. PsiQuantum, founded in 2016 by Jeremy O'Brien and colleagues from the University of Bristol, is built on a fundamental insight about quantum computing architecture: that photonic qubits — particles of light — can be manufactured in existing silicon photonics foundries at a scale that trapped-ion and superconducting qubit approaches cannot achieve. This is not a marginal engineering improvement. It is a structural architectural difference that, if proven out at scale, changes the economics of fault-tolerant quantum computing entirely.
The second characteristic is commercial relevance: the breakthrough must address a problem that customers will pay significant sums to solve. Commonwealth Fusion Energy's compact high-temperature superconducting (HTS) magnet technology — which enables tokamak fusion reactors at one-tenth the physical scale of ITER — is relevant because commercial electricity at competitive cost is worth trillions of dollars annually. The problem is not interesting; it is existentially important.
The third is moat durability: the technical position must be genuinely difficult for well-funded competitors to close. Here is where deep technology diverges most sharply from conventional software. A software product can, in principle, be replicated by any team willing to invest engineering time. The knowledge embedded in Commonwealth Fusion's SPARC magnet programme — understanding of how HTS tape behaves under extreme cryogenic and electromagnetic conditions, how joints are engineered to minimise resistive losses, how quench protection systems function — cannot be replicated by hiring a team of engineers. It took decades of plasma physics research, refined through thousands of hours of magnet testing, to produce.
The Return Profile of Hard Science Companies
The most common objection to deep tech investing from conventional venture managers is timeline. Software companies can reach product-market fit within two years of founding. Deep technology companies routinely take seven to twelve years to reach commercial scale. For fund managers managing ten-year vehicles with fixed LP return expectations, this creates genuine structural tension.
But this objection misunderstands what long timelines imply about return quality. The empirical record of deep technology at commercial scale is compelling. Consider the characteristics of the companies that have successfully navigated the development gauntlet.
PsiQuantum has raised approximately $665 million in total funding from investors including Microsoft, Blackbird Ventures, and the US government's DARPA programme. The company is not yet generating commercial revenue — it is building toward a million-qubit fault-tolerant quantum computer that it believes will be achievable via silicon photonics manufacturing at GLOBALFOUNDRIES. But the technical decisions made in 2016 and 2017 — the choice of photonic architecture, the bet on existing semiconductor foundry processes rather than bespoke cryogenic engineering — now represent a multi-year head start that competitors cannot replicate on any short timeline. The company that emerges from the fault-tolerant threshold will not be competing with near-term NISQ devices; it will be competing for a market estimated at hundreds of billions of dollars in pharmaceutical simulation, financial optimisation, and materials discovery alone.
Commonwealth Fusion Energy, spun out of MIT's PSFC programme in 2018, has raised approximately $1.8 billion in total funding — making it the best-funded private fusion company in the world. Its SPARC experiment, currently under construction, is designed to achieve net energy gain at a compact scale that its founders believe will enable commercial reactor designs within the 2030s. The fundraise includes not just venture capital but strategic investment from Equinor, Eni, Breakthrough Energy Ventures, and others who are making long-duration commitments based on their assessment of the technical trajectory. These are not speculative bets on an unproven technology; they are calibrated commitments by sophisticated energy industry participants who have reviewed the physics and the engineering in detail.
The return profile of such companies — when the technical bets resolve correctly — is not the log-normal distribution typical of software portfolio construction. It is a distribution with larger outlier outcomes and more durable competitive positions, because the moat created by a decade of hard technical work is qualitatively different from the moat created by network effects or switching costs. Switching costs erode as user behaviour evolves. Genuine physical and chemical knowledge does not erode. It accumulates.
The Capital Structure Question
Deep technology companies are capital-intensive by definition. Developing a novel fusion magnet, qualifying a new electrochemical process for industrial deployment, or running Phase II clinical trials for a first-in-class therapeutic are not activities achievable on $10 million. The total funding requirements for companies reaching commercial scale in hard technology are typically in the hundreds of millions to billions of dollars.
This capital structure has two important implications for seed-stage investors. The first is valuation entry: seed-stage positions in deep tech companies are typically priced at fractions of the ultimate commercial value, because the technical risk at the time of investment is genuinely high and the commercial outcome is genuinely uncertain. An investor who backed Commonwealth Fusion Energy in its early rounds at modest valuations now holds a position in a company that has demonstrated net-energy-gain-capable magnet technology and attracted $1.8 billion of total investment. The dilution from subsequent rounds is real, but the initial entry multiple is large enough to generate exceptional returns even with significant dilution.
The second implication is portfolio construction. A seed-stage deep tech fund must accept that some positions will fail — the scientific bet will not resolve correctly, or will resolve correctly but too slowly, or will be beaten by a competing approach. This is not a flaw in the strategy; it is an inherent feature of investing at the frontier of human knowledge. The portfolio must be sized to absorb failures and still deliver fund-level returns driven by the positions that succeed. At Lumino Capital, our $5 million seed fund is sized to make initial positions in eight to twelve companies, with reserves for follow-on investment in the best performers.
Why Now: Three Converging Tailwinds
Our enthusiasm for deep technology investing is not a timeless claim that science-based companies are always the best investment. It is a time-specific claim: we believe the current moment represents an unusually favourable entry point, driven by three converging forces that were not simultaneously present at any prior point in the history of venture capital.
The first is the maturity of enabling technologies that dramatically reduce deep tech development costs and timelines. Machine learning has transformed computational biology: protein structure prediction, molecular dynamics simulation, and genetic analysis that previously required years of experimental work can now be performed in hours using tools like AlphaFold2, Rosetta, and single-cell RNA sequencing platforms. The same machine learning substrate is transforming materials discovery — companies like Citrine Informatics are using AI to navigate the combinatorial space of novel materials at speeds orders of magnitude beyond what traditional experimental chemistry allows. Quantum simulation of chemical systems is already delivering pre-commercial value in pharmaceutical research, creating a proof-of-concept base that strengthens the case for investment in full fault-tolerant quantum computing.
Advanced manufacturing — precision CNC, additive manufacturing, automated metrology — has reduced hardware prototyping cycles from years to months. A company developing novel electrolyser hardware in 2024 can iterate through five generations of prototype in the time a comparable 2010 company could build one. This compression of the iteration cycle reduces the capital requirement per technical learning event and accelerates the resolution of the scientific uncertainties that determine company value.
The second tailwind is policy and capital deployment. The US Inflation Reduction Act committed approximately $369 billion to clean energy and climate technology investment over ten years — the largest single climate investment in US history. The European Green Deal commits €1 trillion of public and private investment over the same period. The US CHIPS and Science Act has committed $52 billion to semiconductor research, manufacturing, and workforce development. The UK government has committed £700 million to nuclear fusion research and identified it as a strategic national technology. These policy commitments create demand pull for deep tech companies in energy, computing, and health that is structural and durable rather than cyclical.
TAE Technologies, which has raised approximately $1.2 billion in fusion capital from investors including Google and Chevron, is pursuing a different fusion architecture than Commonwealth Fusion — a field-reversed configuration designed to eventually use hydrogen-boron fuel that would produce significantly less neutron radiation than deuterium-tritium approaches. The existence of multiple well-funded fusion approaches, each betting on a different physical configuration, reflects the genuine uncertainty about which architecture will first achieve commercial viability. For seed-stage investors, this uncertainty is not a deterrent — it is the source of the return premium. Companies that resolve the uncertainty in their favour have built positions that are impregnable to competition.
The third tailwind is structural: European deep tech has a supply-demand imbalance that creates entry points unavailable in the US market. European universities — Cambridge, Oxford, ETH Zurich, the Max Planck Institutes, Imperial College, Karolinska — produce world-class fundamental science. But Europe's venture ecosystem remains smaller than the US ecosystem relative to the scientific output it has to commercialise. The ratio of venture investment to R&D expenditure in Europe is roughly one-third the equivalent US ratio. This means that European deep tech startups have historically operated in more capital-constrained environments, producing engineering teams that are more disciplined, more capital-efficient, and more technically rigorous than their US counterparts. These are exactly the qualities that predict long-term competitiveness.
The Moat Architecture: What Protects Deep Tech Companies
Every investment thesis must answer the same fundamental question: what prevents a well-funded competitor from replicating this in three years? For software, the answer is network effects, data advantages, or switching costs. For deep technology, the answer typically involves three interacting layers that compound over time.
The first layer is fundamental intellectual property — patents covering novel physical structures, chemical compositions, biological sequences, or engineering architectures. PsiQuantum holds extensive patent positions in silicon photonics-based quantum computing architectures. Quantinuum, the quantum computing company formed by the merger of Cambridge Quantum and Honeywell Quantum Solutions, holds significant IP in trapped-ion qubit systems and quantum error correction. These patents provide time-limited legal protection, but their more important function is as markers of the technical knowledge frontier — evidence that the company was first to solve a specific problem and therefore understands it most deeply.
The second layer is process knowledge — the accumulated, tacit understanding of how to build and operate the physical systems at the core of the business. This knowledge is not patentable because it is procedural rather than structural. It lives in the engineering team's hands, in process documentation, in the institutional memory of failure modes and their solutions, in vendor relationships developed over years of qualification work. Quantinuum's trapped-ion system performance — currently the highest-fidelity gate operations of any commercial quantum computing platform — reflects not just the architectural choices made at founding but years of accumulated process learning about laser control systems, ion trap fabrication, and error correction implementation. A competitor who reads Quantinuum's published papers and patents cannot replicate this without years of their own iteration.
The third layer is ecosystem position — strategic relationships with customers, regulators, research institutions, and talent pipelines that compound over time. Helion Energy, which has raised approximately $500 million including a $375 million investment led by Sam Altman, has signed a power purchase agreement with Microsoft for fusion electricity delivery by 2028. This agreement — even if the technical timeline proves optimistic — creates a customer relationship, a public commitment, and a reputational position that competitors cannot easily replicate. It signals to the talent market, to subsequent investors, and to industrial partners that Helion's fusion approach is taken seriously by sophisticated counterparties.
Risk Calibration: What Can Go Wrong
An honest deep tech investment thesis must confront the genuine failure modes. Not all scientific bets resolve correctly. Not all technically successful companies become commercially successful companies. Not all commercial successes deliver venture-level returns to seed investors after multiple dilutive rounds.
The most common failure mode in deep tech is timeline extension. A technology that works in the laboratory at small scale frequently encounters engineering challenges that delay commercial deployment by years beyond initial projections. Nuclear fusion has a long history of optimistic timelines: the joke that commercial fusion is always thirty years away dates to the 1970s. The current generation of fusion startups — Commonwealth Fusion, Helion, TAE, General Fusion — is betting that a combination of high-temperature superconducting magnet technology, computational plasma physics simulation, and private capital discipline will break this historical pattern. The bets may be right. But investors must price in timeline risk and ensure that capital structures can sustain companies through extended development periods.
The second failure mode is technical substitution — a competing approach resolves the same problem faster or cheaper, eroding the value of the first mover's technical position. In quantum computing, the competition between superconducting qubits (IBM, Google), trapped ions (IonQ, Quantinuum), neutral atoms (Pasqal, QuEra), and photonics (PsiQuantum) represents precisely this risk. Multiple architectures are pursuing fault tolerance via different physical mechanisms, and the architecture that first achieves million-qubit fault-tolerant operation will likely capture a disproportionate share of the initial commercial market. Investors who have diversified across architectures are hedged against this substitution risk; investors with concentrated positions in a single approach must have high conviction in its structural advantages.
The third failure mode is market timing: a technically successful company may succeed too early or too late for its market. A company building industrial electrochemical processes for green hydrogen that are technically superior but cost-competitive only at electricity prices below $20/MWh may need to wait for grid electricity economics to align before commercial deployment makes sense. Patient capital — and fund structures that allow holding periods beyond the conventional ten-year horizon — is essential to capturing value from companies that are technically ready before their markets have matured.
Conclusion: The Compounding Value of Hard Things
Deep technology investing is not for every venture manager. It requires genuine scientific literacy — the ability to evaluate whether a claimed breakthrough is real, reproducible, and commercially relevant. It requires patience — the willingness to hold positions through development timelines that conventional investors find uncomfortable. And it requires conviction — the confidence to maintain a position in a company that has not yet generated revenue while the technical barriers to commercialisation are being systematically resolved.
But for investors willing to develop these capabilities and exercise this patience, the deep tech opportunity is exceptional at this particular moment. The science emerging from academic and industrial laboratories is genuinely world-class. Policy environments in the US, Europe, and increasingly Asia are creating structural demand for the technologies these companies are building. The capital deployment into deep tech globally — visible in PsiQuantum's $665 million raise, Commonwealth Fusion's $1.8 billion, TAE Technologies' $1.2 billion — demonstrates that institutional conviction in this asset class is at an all-time high.
At Lumino Capital, our seed fund is positioned to capture this opportunity at the stage where the return premium is greatest: before technical proof points are established, when genuine scientific uncertainty keeps valuations modest and moat-building has barely begun. The companies we back today — working on problems at the frontier of what physics, chemistry, and biology will permit — represent our best assessment of where the next generation of durable, transformative value is being created. That is the deep tech investment thesis in full.