Q1 2026: AI’s Industrialisation, VC Bifurcation, and the French Tech Paradox
The headline this quarter: AI’s industrialisation is well underway, but the gap between what models can do and what enterprises actually deploy keeps widening. Capability accelerated roughly 2x in 2024, driven not just by raw compute but by new paradigms: reinforcement learning, data quality improvements, and test-time compute (spending more FLOPs at inference rather than training). At current trajectory, the Epoch Capabilities Index should reach 160+ by end of 2026, a step change that would unlock meaningful agentic and scientific automation.
Yet enterprise adoption remains strikingly concentrated. Enterprise AI spend hit $37bn in 2025, up from $1.7bn in 2023, but the distribution reveals how concentrated real deployment still is. Coding alone represents $4bn (55% of departmental AI spend), making it the single largest application category; 50% of developers now use AI coding tools daily, and startups captured 71% of this market. Beyond coding, production deployments concentrate in healthcare ($1.5bn, 43% of all vertical AI spend) and in narrowly scoped horizontal tasks: market research, content summarisation, and simple copilots for customer support and sales. True multi-step agents remain early: only 16% of enterprise deployments qualify as genuine agents; most are still fixed-sequence workflows.
The VC bifurcation
On the venture capital side, the bifurcation is becoming unmissable. Global VC surged to $512bn (+28% YoY), but strip out AI mega-rounds and funding is essentially flat. AI captured 64%+ of all capital deployed globally, up from ~39% in 2023 (for comparison, AI in France is reaching 43% of VC money, roughly a one-year lag). The top five companies alone raised $84bn, roughly 20% of all 2025 VC. France mirrors this concentration: Mistral’s €1.7bn Series C in September represented over 20% of the €8.2bn raised across the entire French tech ecosystem in 2025. Capital is concentrating at the extremes: mega-rounds dominate the top end, while seed and early stages are being squeezed. The middle is hollowing out, and industries outside AI and deeptech are struggling to secure funding.
The French Tech paradox
French Tech is a particular concern, especially in a broader context where the rest of the market is recovering. The US surged +95% YoY to $339bn (64% of global VC), Europe grew +7% to €44bn, yet France is the only major European ecosystem to decline year-on-year (€6.7bn raised, down ~5%). The headline figure masks a much deeper structural shift. The real crisis is especially stark at seed: a 38% collapse, from €0.8bn to €0.5bn. This is compounded by a global GP fundraising collapse ($118bn raised in 2025, down 46% YoY, fund count down 88% from the 2022 peak) that is polarising capital toward megafunds and starving seed of oxygen everywhere, but France feels it acutely.
What we observe locally is a difficult restructuring: fewer funds chasing larger rounds, numerous funds writing small cheques, and a growing tendency to homogenise taste and decision-making. Small funds all want to be followers; large funds want to deploy very large amounts alongside an industry specialist, who vouches for the risk-taking; fewer and fewer seed investors want to write the first meaningful cheque alone without strong traction from other investors. For everyone else outside the narrow profile the market currently rewards, it’s a terrible period. Our concern is that herding behaviour in finance is seldom a good sign, and that fewer seeds today means fewer fundable companies in 24 months. Mega-rounds mask the current fragility of our tech ecosystem.
The assets are real
And yet France has real assets, starting with the talent base. The country has produced world-class AI researchers and engineers at HuggingFace, Owkin, Dataiku, Mistral, Photoroom, and across leading US and French labs (Google DeepMind, Kyutai, Meta FAIR). Strategic sectors are growing at exceptional pace: Harmattan, our portfolio company, became France’s first defence unicorn at a €1.4bn valuation, less than two years after its founding in April 2024, backed by a €200m Series B led by Dassault Aviation. The key ingredients are there. What’s missing is not capability, but enough diversity and density of conviction-driven capital willing to take early technical risk.
For us at Galion.exe, this environment reinforces our thesis. In a market that increasingly rewards followers, we remain committed to writing first cheques on (often technical) conviction. We continue to focus on our sweet spots: vertical foundation models, AI infrastructure, critical industries (defence, human health, and plant biology), and teams building agents at the earliest stage. Our portfolio is still recent, but already showing strong early signals: four uprounds at approximately 3x (Harmattan, Dune, Generare, and 2501), and an average ARR multiple of ~3.4x between entry and end of 2025 across the portfolio, for a portfolio largely deployed in 2024 and 2025.