AI & Robotics: A New Center Of Gravity

What Is It All About?

The AI & Robotics strategy targets the emergence of advanced automation technologies. The portfolio is positioned to capture the development of the broader AI as well as the robotics segments that stand to benefit from their growing convergence. The investable universe is structured around four main pillars:

• Semiconductors: providing the computing power for automation tasks.
• Data: training AI models and enabling their efficient use for every user and use case.
• Hardware Applications: supporting physical automation in industries like manufacturing.
• Software Applications: empowering smarter, faster decision-making everywhere.

A Look In The Rearview Mirror

As of 30 November, our strategy’s outperformance vs. the peer group reached 18% over two years (57% vs. 38%) and 49% over three (116% vs. 67%). 2025 has, so far, resulted in a return to a performance more in line with peers. The reason is clear: since the beginning of this year, the market has rallied along larger caps and hardware infrastructure names. Many of our competitors have the majority of their holdings exposed to the so-called magnificent seven, to the broader semiconductors ecosystem and to the industrial robotics supply chain. Contrary to our peers, we deliberately favored software players (~60% of our portfolio), notably in the data supply chain (e.g., Snowflake) and in AI platforms (e.g., ServiceNow), as we believe they are the prime beneficiaries of the emerging large-scale application rollout. Hardware is not forgotten in our allocation but targets the Chinese robotics ecosystem and AI pure-play chipmakers. All in all, we are where we want to be exposed: to certain winners, not to potential ones.

What We Are Watching

All Eyes On Inference Infrastructure

Capital spending by hyperscalers and major chatbot developers has dominated investor attention in recent months. This theme will remain central in 2026, though with more nuance.

As we have highlighted before, the industry’s focus is gradually shifting from training to inference. This does not imply that training is becoming irrelevant. Rather, the ecosystem is preparing for the large-scale deployment of AI applications. In this context, Nvidia is no longer the only game in town. With hundreds of billions of dollars flowing into AI infrastructure, more companies now have both the scale and the incentive to build custom chips. This trend is already benefiting leading suppliers such as Broadcom and Marvell, both well positioned to ride this wave and occupying prominent positions in our portfolio.

The expected surge in computing demand is also driving strategic moves once considered unlikely, as illustrated by Alphabet opening parts of its infrastructure to Anthropic. More surprises are likely, underscoring the importance of staying alert to new opportunities. At the same time, excesses are visible across several parts of the market. Supply chains are under strain: memory prices are soaring, pushing this corner of the semiconductor industry’s traditional boom-and-bust cycles to unprecedented levels. Energy constraints are emerging, raising concerns. Announced data center capacity expansion plans are sometimes extremely ambitious and may face delays. And with valuations elevated by AI enthusiasm, any execution misstep could be costly.

Talk of a bubble is becoming more common, fueled by financing behavior reminiscent of the dot-com era, but we are not predicting a repeat of the 2000-2002 correction, as the underlying demand trajectory for AI is stronger and more immediate than bandwidth demand was in 2000. If a bubble does exist, in our view, it sits primarily within the chatbot infrastructure segment. A correction, whatever its trigger, would likely affect valuations across the sector, including ours. However, our software positioning should prove more resilient than hardware-concentrated competitors, and would not change the bigger picture: an underlying structural shift toward operational AI application deployment.

Vertical AI Is Already Knocking At The Door

Chatbots continue to dominate the narrative, but are in essence trees hiding the forest. High-profile players such as OpenAI and Anthropic attract most of the capital, driven by ambitions to reach Artificial General Intelligence (AGI). Yet innovation appears to be slowing. As a result, market positions are less secure than before, as illustrated by Alphabet’s Gemini catching up with the latest versions of ChatGPT, whose recent releases have underwhelmed.

Ambitious, general-purpose models risk becoming “jack-of-all-trades”: strong on the surface but lacking depth in specialized tasks. Their complexity also requires vast computing resources, a need reinforced by the shift toward reasoning capabilities. This raises questions about the path to monetization. OpenAI reportedly loses money on its highest subscription tier. Without breakthrough optimization, substantial price increases may be the only option, which would have clear implications for demand.

Meanwhile, smaller specialized models are advancing quickly, though with less visibility. Their narrower scope makes them cheaper to train, more efficient, and more reliable, while reducing development risks. The strength of “small AI” lies in its practicality: meaningful results do not require hundreds of billions in investment. Our own experience developing AI-powered investment tools such as Fundy confirms that meaningful applications no longer require frontier-model scale. Other companies have deployed AI to their advantage, such as CrowdStrike in security or Pegasystems in productivity.

Thousands of similar initiatives are approaching launch, as the gap between impressive demonstrations and production deployment has narrowed dramatically in the past 24 months. This momentum is beginning to show in quarterly results, with AI-related business segments accelerating across the data-infrastructure supply chain, while “legacy” operations remain strong. The future clearly belongs to vertical AI.

The Year Of China

Our exposure to China stands at roughly 25% including Hong Kong, spanning AI platforms with domestic revenue focus, semiconductor equipment benefiting from localization imperatives, and robotics manufacturers serving strong domestic industrial demand.

China has indeed long been a force in innovation, and its recent advances cannot be explained by geopolitics alone. It is true that restricted access to cutting-edge chips spurred the rise of DeepSeek earlier this year, reshaping thinking around AI training and triggering major market reactions. This mindset has since become embedded. In November, Moonshot AI, another Chinese player, captured attention with a model that outperformed most competitors in reasoning, and once again, like it was previously the case with DeepSeek, at a fraction of typical costs.

This focus on efficiency, born from scarcity, combined with a culture of transparency (both DeepSeek and Moonshot released their models as open-source) and strong government support, may prove to be a winning combination. Even U.S. start-ups are reportedly relying more on Chinese models running locally. A similar dynamic is unfolding in robotics, where the government’s clear priorities have stimulated a capable supply chain across both utilitarian applications (industrial robots) and frontier technologies (humanoid robots).

Importantly, both the AI and robotics ecosystems benefit from a large domestic market that increasingly favors local champions while still offering room for international expansion. At the portfolio level, this brings diversification and reduced correlation. Furthermore, the value proposition is reinforced by the significant valuation gap: our Chinese AI & Robotics universe currently trades at a P/S of 2.3x, (10-year average at 2.4x), compared with 7.1x (10-year average at 4.6x) in the U.S., despite limited differences in terms of EPS growth prospects.

Where We Stand, And What’s Ahead

We enter 2026 maintaining our core positioning, with players already witnessing actual revenue pickup from AI deployment and not relying on speculative longer-term projections:

• Enterprise AI platforms that enable a boost in productivity (ServiceNow, Pegasystems);
• Vertical software that brings a decisive competitive advantage, in security (CrowdStrike, Palantir) or healthcare (Pro Medicus);
• Data infrastructure (Datadog) and data management (Snowflake that enable AI applications to scale and become ubiquitous;
• Pure-play AI semiconductors enabling training and, most importantly, inference (Broadcom, Marvell);
• We also remain extremely constructive on China at current valuations, be it in the semiconductor manufacturing segment (Naura), the AI application (Kingdee) or the robotics ecosystem (Inovance).

We see no reason for this positioning to change in the short term. Nevertheless, potential triggers for a change to a more defensive stance include hardware and AI-model architectural breakthroughs (potentially impacting capex flows), as well as potential cracks in the funding structure of the hardware ecosystem (potentially impacting valuation multiples). Furthermore we closely monitor the chatbot sector’s advances towards profitability, as it will have a critical impact on investors’ sentiment towards valuation levels. Similarly, we closely watch figures related to Chinese internal demand and geopolitical factors.

All in all, the structural case for AI deployment continues to strengthen. Staying on the sideline is not appropriate given the evidence accumulating in corporate earnings and enterprise adoption metrics, a situation embodied by attractive relative valuations, with a PEG of 1.55 (vs. 1.78 for the Nasdaq 100). But the path will not be linear, and portfolio construction matters. Our positioning reflects the conviction that the next phase rewards companies deploying AI profitably and not merely investing in its possibility.