Deep Tech Monthly in Review - December 2025
Field notes from last month in Deep Tech startups and private markets — a strategic recap for Builders and Backers.
Dear Builders and Backers,
As we open a new year together, we’d like to use this edition of our Deep Tech Monthly in Review to wish you a great 2026!! 🚀✨
May it bring you real opportunity, clearer conviction, and the kind of calm focus that lets you navigate volatility without losing sight of the long game.
As you know, at the beginning of each month we publish this review to take stock of what has unfolded in Deep Tech over the past few weeks, so the new month can begin with context, composure, and a clear sense of direction.
So here we are again!
December delivered a rare kind of clarity. Not the “everything is up and to the right” clarity, but the industrial kind—where incentives, procurement logic, manufacturing constraints, and balance sheets show their teeth.
Three patterns stood out:
Balance sheets became active allocators of frontier-tech risk.
One large semiconductor player’s decision to sell a mature business line and reinvest the proceeds into AI-era infrastructure was emblematic of a broader shift: late-stage deep tech outcomes increasingly look like negotiations between cap tables and corporate treasuries, not “IPO by default.”“Scale is the product” quietly replaced “the prototype is the product.”
A recent battery de-SPAC made clear that the real story is not chemistry but financing the expensive middle—yields, reliability, integration, and factory truth arriving quickly enough to withstand public-market scrutiny.The “reindustrialization” slogan hardened into balance-sheet commitments.
Government-backed funds, export-credit structures, and buyout-led rollups in aerospace and energy all treated factories, reactors, and supply chains as national infrastructure, pushing capitalization terms closer to project finance than to traditional venture.Enjoy the read,
— Giulia & Nicola
December: The Big Reality Check
December felt like a month where multiple sectors hit the same wall from different angles: physics and supply chains refuse to negotiate, while capital increasingly demands execution evidence.
Aerospace and Space Systems
The space stack continued its evolution from “missions” to repeatable industrial services. Reditus Space’s push toward reusable microgravity missions reinforced a pragmatic wedge: selling experiment cadence and return logistics as a product, rather than betting everything on a future in-space economy. The technical center of gravity here is not exotic propulsion; it is thermal management, re-entry engineering, and refurbishment discipline—the unglamorous mechanics that make unit economics real.
Defense-adjacent Earth observation also kept moving toward subscription-like demand, with ICEYE’s SAR constellation scaling under a defense and intelligence pull. SAR’s value proposition—imaging regardless of clouds and darkness—reads like a procurement-friendly capability, not a speculative “space app.”
Meanwhile, aerospace finance looked structurally healthier than many expected. Commercial aerospace PE deal flow broadened, with 96 deals in the first nine months of 2025 vs. 88 in 2024, and Q3 alone at 44 transactions, against 9.8B passengers projected for 2025. The implication: aviation constraints (aircraft supply, labor, sustainability mandates) are pulling capital toward software, data, MRO, and efficiency infrastructure, not just OEM exposure.
Materials and Advanced Industrial Inputs
Materials innovation showed up less as “new molecules” and more as new bottlenecks becoming investable. Photonic interconnect is the cleanest example: the constraint is no longer transistor density alone—it is moving data fast enough, cheaply enough, with tolerable energy per bit. Marvell’s acquisition of Celestial AI framed optical interconnect as a control point for AI infrastructure, not a feature.
In parallel, the “materials” story also included industrial process infrastructure—platforms that create repeatability. The quiet meta-pattern: the market is paying for manufacturability and deployment interfaces, not for novelty in isolation.
Energy: Fission, Fusion, Geothermal, and the Infrastructure Lens
Energy in December split into financeable wedges—each with its own risk curve and contracting pathway. Microreactors continued to mature as “portable firm power,” with Radiant raising over $300M and targeting a factory buildout and a 2026 demonstration at Idaho National Lab, while positioning deployment timelines explicitly into the back half of the decade.
Antares’ microreactor round reinforced a second theme: reactors are being designed for multiple theaters—land, sea, and space—which forces a convergence of licensing, logistics, and high-reliability manufacturing.
Geothermal, meanwhile, looked increasingly like contractable capacity rather than “promising baseload.” Fervo’s $462M Series E was framed deployment-first, tied to execution milestones, with Cape Station expected at 100 MW in 2026 and scaling toward 500 MW by 2028—the kind of schedule utilities and capital providers can underwrite.
Storage also moved toward an infrastructure mental model: grid-scale batteries increasingly treated as a standalone infra asset class with contracted revenue stacks—while battery swapping reminded everyone how fast a capital structure collapses when the edge device (the vehicle platform) is not controlled.
Agrifood and Bio-Industrial Systems
Agrifood’s December signal was less about consumer-facing narratives and more about industrial biology as operational leverage. A recurring motif across the month’s work: biology becomes investable when the delivery mechanism fits existing procurement and distribution rails.
Examples clustered around:
crop protection shifting toward data-driven discovery pipelines and scalable manufacturing logic,
alternative protein and cultivated approaches inching from “promise” toward cost-optimized bioprocessing constraints, and
cold chain and packaging efficiency being treated as energy and margin infrastructure (notably, refrigeration’s outsized energy footprint in food systems).
Defense and Dual-Use
Defense remained the sector where procurement pull converts technical progress into revenue clarity faster than in most civilian markets. ICEYE’s defense-tilted trajectory, Moonshot Space’s defense-linked hypersonic testing infrastructure, and the general drift toward autonomy stacks that demand vertical integration all pointed toward the same reality: in dual-use, the buyer’s contracting model is often as important as the technology itself.
Quantum and Compute Infrastructure
Compute in December was defined by a single question: where does scaling actually come from now? Photonics and packaging are part of the answer; software rails are the other.
Lightmatter’s story captured the industrial approach: silicon photonics designed to ride existing foundry infrastructure, paired with packaging partnerships and a phased go-to-market strategy that starts with interconnect before compute replacement. This “complement-first” posture is less romantic than “GPU killer,” but structurally more compatible with how data centers adopt new hardware.
On the quantum side, Horizon Quantum Computing’s hardware deployment in Singapore highlighted a regional pattern: sovereignty and operational readiness matter, and owning the stack changes credibility—from middleware to reference infrastructure.
Biotech and AI-Native R&D
Biotech’s December pattern was not “AI is coming.” It was “AI is already outrunning the wet lab, CMC, and regulators,” and the industry needs organizational interfaces that keep the whole pipeline from becoming slideware.
Terray Therapeutics’ milestone with Bristol Myers Squibb illustrated the direction of travel: integrated platforms that behave like installed capital equipment—automation, proprietary datasets, models—embedded into pharma workflows rather than bolted on as vendors. The business model implication is compounding: value accrues over multiple programs once trust and integration are established.
In parallel, “self-running labs” and experiment-as-a-service models (e.g., ChemLex and Pauling.AI) suggested a new procurement object: lab hours and iteration velocity as a contracted service, not just a cost center.
Manufacturing and Industrial Execution
December reinforced a blunt truth: manufacturing is no longer a back-office function—it is the moat. The month’s work repeatedly returned to the same axis: hard infrastructure + operational discipline beat pure innovation narratives.
That showed up in capital structure too. Factorial’s de-SPAC mechanics effectively priced the market’s skepticism: public capital is available for factories, but only when manufacturing evidence is arriving quickly enough to survive the quarterly cadence.
Transportation and Mobility
Mobility continued its hard reset. The interesting parts were not “EV vs. not EV,” but who controls the platform and which revenue stacks look contractual.
Natilus’ blended-wing-body ambitions—paired with airline appetite for efficiency bets—kept airframe architecture on the table as a competitive variable again, not just fuels and engines. Meanwhile, battery swapping’s fragility underlined the platform-control lesson: infrastructure economics fail fast when standardization and distribution are not owned.
Last Month in Data
$4.5B — Unconventional AI’s $475M seed round was reported at a $4.5B valuation, indicating continued investor appetite for early-stage AI and compute infrastructure platforms.
~750 miles — Factorial reported a Mercedes test vehicle nearing ~750 miles of range using its cells, illustrating the performance potential of next-generation batteries pending large-scale, repeatable production.
100 MW (2026) — Fervo’s Cape Station is planned to deliver 100 MW in 2026, with a ramp to 500 MW by 2028, positioning enhanced geothermal as a capacity asset with a defined deployment schedule.
600+ miles — Heven’s Z1 hydrogen platform is characterized as providing more than 600 miles of range and over 10 hours of flight time, targeting long-endurance unmanned operations.
$13M — Onepot.AI, based in San Francisco, launched with $13M to automate chemical synthesis for drug discovery, combining a 1.9M-compound digital library, an in-house AI engine (“Phil”) for route design, and a robotic synthesis platform (POT-1) capable of delivering new molecules in roughly five days.
$115B+ — According to Crunchbase’s 2025 snapshot, startups under three years old raised over $115B in $100M+ megarounds, exceeding the 2021 peak, with AI-focused companies accounting for a significant share of this activity.
Policy and Regulation Watch
A focused extract from our wider regulatory observatory—spotlighting a few policy moves that quietly set the terms for Deep Tech ventures.
Germany consolidates deep tech public capital: DTCF management moving under HTGF starting February 1, 2026, with a broader seed-to-growth platform goal—practically, a tighter pipeline from early checks to later-stage financing within a single state-backed architecture.
U.S. defense posture shaping space and missile-tech procurement: the FY2026 NDAA’s “Golden Dome” framing signals sustained demand gravity for tracking, sensors, and defense-linked space infrastructure—relevant for dual-use companies that can meet procurement timelines and security constraints.
EU launches EIT Water as a long-horizon water-innovation platform: EIT Water, run by the Allwaters consortium, is set up as a 15-year Knowledge and Innovation Community to move water-tech from lab to market—targeting climate-driven extremes, ecosystem restoration, and “blue economy” circularity, while providing non-dilutive support, incubation, and cross-border deployment pathways from 2026–27 onward.
UK deploys Great British Energy with an $11B+ clean-energy mandate: through GBE, the UK commits more than $11B over five years to co-develop and own wind, solar, and storage projects totaling an extra 15 GW by 2030, aiming to crowd in ~£15B of private capital and use project income and local-jobs narratives to anchor long-term power-sector investment.
Singapore RIE2030 and semiconductors as a national program: the S$37B plan and its semiconductor flagship reflect a practical global trend—talent, fabs, and applied research bundled into a single competitiveness instrument.
Canada’s research talent push: $1.7B earmarked to attract top researchers functions as deep tech industrial policy via people—the scarcest input for frontier compute, biotech, and advanced manufacturing.
Three Breakthroughs
What changed, technically—and why it matters
1) Photonics becomes infrastructure—not an “adjacent” component
The month kept pointing to the same compute truth: moving data is becoming as hard as computing it. Optical interconnect, photonic fabrics, and silicon photonics aren’t new ideas—but December showed they are increasingly treated as system-level infrastructure.
How it works (technical core): photons carry information with far less resistive loss than copper over distance, enabling higher bandwidth density and better energy-per-bit in the right architectures.
Why it matters now: data center scaling is hitting interconnect, packaging, and thermal walls. That pushes value toward the companies that make racks behave like coherent systems, not loosely connected boards.
Market implication: moats shift toward packaging ecosystems, integration know-how, and deployment compatibility—less “single chip advantage,” more “system throughput advantage.”
2) “Factory-grade nuclear” takes a step closer to an equipment category
December treated microreactors less like policy theater and more like an emerging equipment class.
How it works: small reactors aim for standardized builds, repeatable manufacturing, and transportable deployment profiles—reducing bespoke engineering and compressing deployment cycles (at least in theory).
Why it matters: standardized systems create clearer cost curves, clearer licensing roadmaps, and clearer procurement pathways—especially for remote, industrial, and defense-adjacent loads.
Market implication: competition shifts from “who has the most elegant reactor concept” to “who can build, certify, and deliver reliably.”
3) Space is closing the loop: return logistics and test cadence become strategic
Reusable re-entry, microgravity logistics, and hypersonic testing infrastructure are converging into something that resembles an industrial supply chain.
How it works: re-entry vehicles and reusable platforms turn microgravity experiments and payload returns into repeatable cycles, while hypersonic testing infrastructure reduces time-to-iteration for defense and aerospace programs.
Why it matters: cadence compresses development timelines and turns “space” into an operational layer—where buyers pay for turnaround time, reliability, and integration.
Market implication: companies that own the logistics layer can become the rails on which multiple industries iterate (pharma, materials, semiconductors, defense).
Ten Lessons Learned
Industrial proof is back as the credibility currency.
Manufacturing competence is increasingly the defensible moat.
Compute is a power-and-cooling problem as much as a software problem.
Defense remains the cleanest demand signal when systems are procurement-ready.
In energy, financeability follows capacity + timeline + buyer, not narrative.
In biotech, AI advantage depends on lab/CMC/regulatory interfaces.
Space is industrializing wherever cadence and refurbishment are designed in.
Sovereignty is becoming a product requirement, not a policy footnote.
Capital structures are tightening around milestones and near-term evidence.
The best deep tech companies increasingly look like operating companies early.
Last Month on The Scenarionist
Deep views on Deep Tech to see clearly, decide with discipline, and move first.
DEEP TECH BRIEFING
What Actually Moved in the Deep Tech World
Weekly briefings for the moments when someone around the table asks, “What really changed this month?” and you need a single coherent narrative.
If you sign off on megawatts, missions, or data centers, start here.
This Briefing reframes factory-built microreactors, waste solutions, geothermal districts, and reusable reentry as one infrastructure story—where firm power, orbital logistics, and AI-native tooling move from climate slogans to multi-decade assets on the same footing as fabs and ports.
For when “fusion revenue” and space power show up in a board memo.
This edition ties together fusion PPAs, nuclear fuel in orbit, geologic hydrogen, and power beaming as early examples of how offtake, fuel logistics, and orbital generation are starting to be structured like real contracts—not just technology bets.
Mandatory if you care how public markets, automation, and infra interact.
Here, a solid-state battery de-SPAC, autonomous chemistry labs, photonics asset recycling, hydrogen exploration, and defense logistics are treated as one risk surface—showing how capex-heavy platforms, self-running R&D, and security demands are converging into an integrated industrial stack.
DEEP TECH CAPITAL MOVEMENTS
What Capital Believes Before the Narrative Catches Up
A weekly X-ray of where serious money is really going in Deep Tech, and what that implies about how the next cycle will be priced.
The issue to read before you call any deep tech play “too heavy.”
This edition tracks capital from pre-seed to Series D into SMRs, counter-UAS stacks, energy, compute, and oncology—useful shorthand for when investors accept regulatory and hardware risk because the output looks like contracted power, deterrence, or clinically validated pipelines.
For when decks start mixing excavators, drones, quantum, and clinical data.
Seed rounds in construction AI, autonomy, and industrial software sit alongside later-stage space surveillance and clinical networks, showing how capital is treating workflow AI, dual-use autonomy, and data platforms as infrastructure rather than point tools.
The reference point for oversized early rounds in hard tech.
A $475M “seed” in compute, a $462M geothermal round, and large tickets into satellite manufacturing and quantum systems are unpacked as one pattern—when a platform, capital intensity, and demand justify compressing multiple financing stages into a single step.
For when late-stage deep tech risk feels opaque on both tech and policy.
This piece follows capital clustering around microreactors, ag autonomy, AI-native security and governance, and multi-program biotech, and treats them as infrastructure layers—power, food systems, and trust—priced with utility-like expectations.
DEEP TECH CATALYST
How to Actually Build and Back Deep Tech Companies
Operator-grade conversations and playbooks for the moments when physics is not the problem anymore, but contracts, capex, and integration are.
The episode that resets what “Series A” means for hardware and infra.
This conversation makes the case for using Series A to prove demand, ship product reliably, and run a “workshop as first factory”—and only then stepping into real plants once bookings, working capital, and burn efficiency justify it.
For when a family office or foundation appears in the syndicate and the rules change.
The episode maps how patient capital underwrites healthcare and AI differently from closed-end funds—timelines, data requirements, governance—and how founders can match structures, missions, and risk appetite to long-cycle clinical and regulatory realities.
Required listening before signing a term sheet around next-gen nuclear.
Licensing, siting, capital stacks, and industrial partners for Gen IV reactors are treated as design parameters, not afterthoughts—offering a practical checklist for what has to be true for nuclear ventures to be financeable, deployable, and durable.
INSIGHTS & ANALYSIS
Capex, Patterns, Exits: The Long Game
Slow thinking on the parts of Deep Tech that don’t fit into a weekly news cycle: industrial capex, process technology, and how exits are really engineered.
The piece to keep nearby for any non-GPU compute investment memo.
A reverse-diligence walk through Lightmatter and Mythic as case studies in photonic and analog AI hardware—covering architectural moats, foundry dependence, capital needs, and what it really takes to challenge a GPU-centric ecosystem.
Non-optional if the edge depends on a factory, process, or test stand.
This analysis explains how fabs, pilot lines, and industrial processes become the moat in deep tech, and offers a framework for when building hard infrastructure in-house turns from a cost line into a long-term strategic advantage.
A year-end reflection for anyone thinking in industrial cycles, not quarters.
From Bell Labs’ transistor demo to the 2025 deep tech landscape, this essay traces how infrastructure-heavy technologies move from fragile prototypes to system-level commitments, and what the past year signaled about power, compute, and manufacturing as the next industrial wave.
Final Thoughts
December didn’t shrink the opportunity set—it sorted it. The market rewarded technologies that can be explained in industrial terms: production, integration, regulation, procurement, and timelines. The rest remained interesting, but less financeable.
That sorting mechanism is unlikely to soften in early 2026. If anything, it is becoming the defining feature of the cycle: deep tech is entering a phase where the edge belongs to teams and investors who treat execution and capital structure as first-class technical problems.
See you soon!
— Giulia Spano, PhD & Nicola Marchese, MD
“What’s dangerous is not to evolve.”
— Jeff Bezos