Investment Manager's Report

Two product releases in 2025 changed the way software gets built. In February, Anthropic launched Claude Code as a research preview: the first agentic tool that could take instructions, execute actions on a developer's computer, and deliver working code. In November, they released Opus 4.5, a model capable enough that it settled the question of whether AI could write production code. By the end of the year, AI-assisted development had gone from experiment to default.

Then, in February 2026, Anthropic released CoWork, extending the same agentic capability to every knowledge worker. In the weeks that followed, roughly $1 trillion was wiped from software company valuations. The market grasped what developers had known for months: AI agents can do much of what traditional per-seat software does, and the licensing model that built the SaaS industry is under genuine threat.

These events occurred after the reporting period, but they are the context in which the Fund's 2025 results should be read.

This is the "SaaSpocalypse." It is real, but it is also overdone.

The panic is selective. Horizontal point-solution SaaS vendors with low switching costs face genuine disruption. An AI agent that can process expenses, schedule meetings, or manage basic workflows makes the dedicated tool redundant. Atlassian fell 35% in days. Intuit fell 34%. The North American Tech Software Index dropped roughly 30% from its September 2025 peak.

The broader picture tells a different story. Gartner forecasts worldwide software spending will grow 14.7% in 2026 to over $1.4 trillion, roughly $180 billion in net new spending in a single year. Forrester projects global SaaS spending will nearly double from $318 billion (2025) to $576 billion by 2029. The software market is getting larger, not smaller.

What is changing is where the value sits. There will be a rich software industry on the other side of this transition, built by companies that step up to provide the agent workflows, orchestration, and intelligence around core data platforms that enterprises need. But it will belong to companies that can make the transition. Those that cannot will be left behind.

In the PC era, value shifted from hardware to software. In the internet era, from infrastructure to applications. In the AI era, value is moving from model providers to those who make AI deployable in specific sectors. The winning companies will build applications with domain expertise, verification, and the decision scaffolding that makes AI outputs trustworthy.

The deployment gap

The capability constraint is over. The models are here. Agentic harnesses work. Open-source models match proprietary systems at a fraction of the cost. China's Moonshot released Kimi K2.5, comparable to the best proprietary systems at agentic work and built for running up to a hundred agents in parallel. Raw capability is no longer the bottleneck.

And yet 95% of enterprise AI projects still fail to reach production.

MIT's 2025 research found the primary obstacles are data quality (43%), lack of technical maturity (43%), and skills shortage (35%). S&P Global reports 42% of companies now abandon AI initiatives before production, up from 17% the previous year.

Software development transformed first because the models were trained on verification systems (test suites, type checkers, continuous integration). Through reinforcement learning, they got fast feedback on what works and what does not. Other sectors lack this. Institutional knowledge lives in people's heads and walks out the door when they leave. Decision logic sits in spreadsheets no one maintains. The verification systems were never built.

Portfolio Performance

At the Company level, SuperSeed Capital Limited reported net assets of GBP 3.24 million as at 31 December 2025, with the fair value of investments at GBP 3.29 million. Net asset value per share was GBP 1.3688 (fully diluted: GBP 1.33), and total comprehensive income for the year was GBP 270,618. The NAV increase reflects continued value creation within the Fund's underlying portfolio.

During the year, the Company managed its liquidity through issuance and repayment of convertible loan notes and selective sales of interests in the Fund to the Investment Manager. The aggregate principal of the convertible loan notes was reduced from GBP 1 million to GBP 400,000, with the redemption date extended to February 2028. The Investment Manager, SuperSeed Ventures LLP, is the counterparty to these transactions.

At the Fund level, Q4 was the strongest quarter since inception, capping twelve months of accelerating growth:

·     Net IRR: 21.1%

·     TVPI: 1.25x

·     DPI: 0.13x

·     Portfolio CARR: $22 million (up 69% from $13 million at the start of the year)

Portfolio revenue grew at nearly 100% annualised in Q4, up from 50% year-on-year in 2024. Over four years, the Fund has delivered 21.1% net IRR, top quartile for its vintage year according to Cambridge Associates. That compares with 10.6% annualised from the S&P 500 over the same period.

Popp had a breakout year as its AI-powered recruitment platform found clear product-market fit. Finteum brought Goldman Sachs live on its distributed ledger platform for intraday liquidity management; four of the world's largest banks are now active on the platform. Duel closed a $16 million Series A, with enterprise clients including Estee Lauder, Lush, and Victoria's Secret. OctaiPipe secured ABB as both strategic investor and distribution partner for AI-powered data centre cooling.

A small number of portfolio companies had a harder year. That comes with the territory at pre-seed. Where commercial traction has been slower than hoped, we are working with founders on repositioning and restructuring. There is potential in several of these situations going into 2026.

New investments

2025 was the Fund's final year of active investment. We completed seven new investments, with two more closing just after year-end, bringing the portfolio to 27 companies.

In Q1, we completed three investments. Biographica has built a foundational model and graph-based ML system for crop gene identification that cuts the process from six months to two weeks. Bench applies AI to aerospace and automotive design, compressing multi-variable optimisation from months to hours.

In Q2 and Q3, we added two. Ploy maps every employee permission across every enterprise system and automates just-in-time access and revocation. As AI agents proliferate and the number of non-human identities in enterprise systems multiplies, this problem is growing fast. Fit Collective uses 348 million fit data points to fix clothing sizing before production. About half of online fashion returns are fit-related; Fit Collective eliminates the problem at source.

In Q4, two more. Hive Autonomy is building a universal platform for remote and autonomous operation of industrial vehicles. One operator, multiple machines, from a centralised station. Forklifts, excavators, diggers: these machines move trillions of pounds of materials globally, and most operate exactly as they did fifty years ago. SOLVE Chemistry combines automated high-throughput flow chemistry with machine learning, compressing pharmaceutical process development from 9-24 months to weeks and generating 20x more data than traditional methods. Already working with BASF, AstraZeneca, and Hovione.

Shortly after year-end, the Fund completed its final two investments: Cursive, a DeepMind spin-out building real-time generative AI models, and All3, a construction technology company that extends our Physical AI thesis into one of the world's largest industries.

The investment period is now complete.

Portfolio spotlight: Physical AI

I want to draw attention to three portfolio companies that show where AI creates defensible value in the physical world.

Hive Autonomy works where traditional automation fails. Automated guided vehicles need controlled environments and fixed paths. Hive's AI operates in unpredictable conditions: active warehouses, construction sites, factories. Every site is different. The hard part is not the model. It is making autonomy work where scripted automation breaks down.

OctaiPipe tackles one of AI's ironies: its own energy consumption. Data centres now consume approximately 1.5% of global electricity, approaching 3% by 2030. OctaiPipe's AI cuts facility power consumption by up to 30%. With ABB now distributing the product to the world's largest operators, hyperscale opportunities are in the pipeline for 2026.

All3 brings the same logic to construction. Paper processes, tribal knowledge, and manual coordination waste billions a year. All3's system spans AI-generated designs, robotic timber manufacturing, and autonomous on-site assembly. Construction has seen essentially zero productivity improvement in fifty years. All3 is built to change that.

What these three have in common: they operate in physical industries where failure is expensive, environments are messy, and knowing the domain matters more than having the biggest model.

Outlook for 2026

The shakeout in legacy software will continue. Gartner predicts 35% of point-product SaaS tools will be replaced by AI agents by 2030. But the market for software that helps organisations deploy and verify AI is growing fast. Our portfolio companies are building that software.

The AI infrastructure buildout continues at a staggering pace. The major cloud providers have announced 2026 infrastructure spending totalling over $600 billion. What AI applications actually generate in revenue is a fraction of that. Whether this becomes the foundation for the next computing platform or one of the largest capital misallocations in history depends on how fast the application layer catches up. We are building on the application layer.

Geopolitics is, counterintuitively, working in favour of European software. The Middle East conflict has destabilised energy markets; Europe entered 2026 with gas storage well below prior-year levels. The Trump administration has tied EU digital regulation enforcement to tariff negotiations. France has ordered 2.5 million civil servants off American cloud platforms by 2027. The EU's forthcoming Cloud and AI Development Act is expected to require European infrastructure for EU customers. The European sovereign cloud market is projected to grow from EUR 20 billion to EUR 100 billion by 2031.

Christine Lagarde made the point well in her March 2026 ECB speech: unlike in the 1920s, today's AI capabilities depend on global supply chains, so geopolitical fragmentation would erode them immediately. The 1920s are the cautionary tale precisely because policymakers treated technology and international order as separate domains. That was the mistake. For European software companies, though, data sovereignty, regulatory divergence, and local procurement mandates are expanding the addressable market. Every new regulation increases demand for compliant, European-built software.

Three Fund portfolio companies are approaching Series A rounds, with significant fundraising activity expected in H1 2026. As these rounds close at valuations reflecting the revenue growth already achieved, we expect further expansion in the Fund's TVPI.

Conclusion

We began investing in January 2022, nearly a year before ChatGPT's release. Since then, the model layer has produced enormous winners: OpenAI's valuation approaches $1 trillion, and Anthropic is not far behind. Building foundation models requires tens of billions in capital and is a race with very few entrants. For venture investors at the earliest stages, the opportunity sits in the application layer: companies that take these powerful models and make them work in specific industries, with the domain expertise and verification that turns raw capability into reliable software.

The current disruption is destroying value in commodity software. The infrastructure buildout is funnelling hundreds of billions into the compute layer. Application-layer companies with proprietary data and deep domain knowledge are where we believe lasting, investable value will concentrate.

That is what we built the portfolio to do. With three companies approaching Series A in H1 2026, we expect the next twelve months to translate portfolio momentum into marked-up valuations.

Mads Jensen

Managing Director

SuperSeed Ventures LLP