AI & Robotics: Time for the application boom

Allocation preferences 

Understanding AI & Robotics

What is it all about?

The AI & Robotics strategy provides exposure to advanced automation technologies. The investable universe is structured around four main pillars:

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

Investors are demanding tangible results 

The launch of ChatGPT in November 2022 acted as a significant catalyst for what may ultimately be a historical technological transition. Its impressive capabilities brought AI back to global attention, sparking a wave of substantial investments.

This impact was particularly profound for the semiconductor industry, where leading technology providers like Nvidia achieved spectacular stock performances. In 2024, this momentum continued, with the frenzy extending to the broader data center supply chain.

Eyes are now turning towards applications, a segment that concentrates the most innovation. Investors are starting to demand returns on their investments and are no longer satisfied with the promises of longer-term perspectives. This put pressure on companies to deliver tangible results, penalizing players with inconsistent execution (e.g., UiPath Inc) but handsomely rewarding top performers (e.g., Palantir Technologies Inc).

It became clear that AI is not a fad but a durable trend. Faced with secular headwinds, such as an aging population and inflation, companies (and even countries) have no other choice than to rely on automation technologies to fuel growth and productivity gains. Recent innovations are poised to rapidly and significantly make inroads in the global economy, with notable convergence already taking place in sectors like healthcare and robotics. 

Hot topics

Applications at a turning point

Rapid improvements in large language models (LLMs) and other generative AI-based technologies have dramatically increased application capabilities. While these models have not realized their full potential, they are mature enough to enable large-scale deployment. We believe this rollout will incredibly impact productivity and sectors such as healthcare and security. 

On the productivity front, businesses (including ourselves) leverage LLMs for tasks like content generation, customer support, and data analysis to streamline operations and enhance decision-making. In healthcare, the focus is on improving patient outcomes through personalized medicine and efficiency gains through the automation of administrative tasks. Finally, in cybersecurity, AI can detect threats faster than traditional methods, while visual recognition algorithms massively improve the efficiency of surveillance systems.

Beyond the ability to sustain innovation, the key differentiating factor is how effectively these applications can integrate into existing workflows and frameworks: a powerful chatbot is useless if users don’t bother using it due to a lack of convenience. However, companies that successfully address this challenge, as shown by ServiceNow, can capitalize on the strong demand for differentiated technologies - provided they solve another critical issue: monetization. In this regard, additional innovation might be required to find the best solution with end-customers’ aspirations and budgets.

Hardware focus to shift 

Hardware players, and Nvidia in particular, have been the clear, undisputed winners of the recent AI cycle, as a massive capacity build-up was necessary to power the technology. This surge in demand led semiconductor valuations to reach historic highs. Questions remain regarding the capacity required for training, although new generations of GPUs from Nvidia and AMD launching next year are expected to drive further growth. Yet betting on GPU suppliers is no longer a no-brainer.

However, what is certain is that with AI applications becoming widespread, the demand for inference chips (designed to run AI models rather than train them) will only increase. While GPUs can handle inference tasks, specialized chips, often custom-designed for specific end-users, are far more efficient. We see such custom architectures proliferating in data centers, not only for inference but also for specialized functions such as networking. With data center operators under pressure to minimize operating costs, maximizing efficiency will be a top priority. Custom chip designers such as Marvell are poised to benefit.

Political uncertainties

The new U.S. administration poses a dilemma for investors. On one hand, its official hands-off approach contrasts with the Biden administration’s caution. Repealing the implemented safeguards would undoubtedly fuel innovation. On the other hand, proposed tariffs will likely drive up the cost of computing chips, which are predominantly manufactured outside the United States. 

But the elephant in the room is the potential judiciary disruption. Several legal procedures have been engaged against AI giants for the non-consensual use of data to train models, which could rewrite the business playbook and topple current business models. In the most extreme outcome, such as in the procedure opposing the New York Times newspaper to OpenAI, this could force model developers to delete the problematic data from their training set and existing models, implying a costly re-training phase. 

At the same time, concerns exist regarding a potential weaponization of Justice by the new administration with the entry of major stakeholders in the government, which could favor some players and negatively impact others, resulting in major instability that would delay innovation. However, this could work in several ways, e.g. with antitrust procedures likely to end up stimulating innovation if enacted. Once again, responsiveness and adaptability will be key for investors to navigate this period successfully.

Catalysts

• Innovation trigger. Successfully developing new technology building blocks and combining existing ones can make the difference between a curiosity and a success story. Considering the current effervescence, there has never been a better time for breakthroughs to happen.

• Deregulation. The Biden administration created safety barriers to control the development of AI. The new U.S. administration may choose to remove these barriers, favoring innovation over caution.

• Tighter implementation. AI is already powerful, but implementation constraints in existing frameworks limit its effectiveness. Breaking this barrier would lead to a landslide adoption of the technology.

Risks

• Hardware woes. The proposed U.S. tariffs may lead to higher hardware costs, thus capping the growth potential. The new administration's policies may also lead to geopolitical tensions, further disrupting existing supply chains. 

• Data wall. Data is already emerging as a limiting factor in certain cases for effectively training AI models. This challenge could be exacerbated if legal actions regarding unauthorized data usage succeed, potentially restricting access to critical datasets. 

• Demanding investors. The strategy's potential and addressable market are huge, but so are investors’ expectations. Therefore, the tiniest execution mistake could quickly spiral out of control from a stock market perspective, regardless of the underlying business’s actual performance.