AI & Robotics - Innovation over taxation

Hot topics

Here is an update of the Hot Topics presented in our 2025 outlook.

Applications at a turning point

In our outlook, we noted that AI models were reaching maturity setting the stage for large-scale deployment, particularly in productivity. This has proven accurate, as evidenced by a strong run of positive earnings and news flow. ServiceNow reported robust adoption of its AI offerings: Pro Plus deals quadrupled in Q1 and Now Assist became its fastest-growing product ever, reaching $250mn in annual contract value within 18 months and targeting >$1bn by the end of 2026. Pegasystems saw its stock surge in April after a 33% revenue beat, driven by strong customer demand for generative AI solutions. Even UiPath Inc began to recover following a prolonged slump, buoyed by improved results.

This is only the beginning. AI agents, combining the cognitive power of Large Language Models (LLMs) with the execution capabilities of Robotic Process Automation (RPA), are just starting to roll out, e.g., for customer service. ServiceNow's management highlighted a spending cycle likely to last “at least the next ten years,” a forecast that seems increasingly plausible as AI capabilities accelerate. In parallel, the recent breakthrough by DeepSeek is expected to provide a lasting boost. We expect its architectural optimizations, which reduce computing power requirements for both training and inference, to become an industry-standard. This will lower operational costs and further drive the adoption of AI applications.

Hardware focus to shift

The impact of DeepSeek was also strongly felt in the hardware segment. On the trading day following the company’s mainstream debut, stocks of leading AI datacenter chipmakers fell by nearly 20%. This sharp selloff highlighted a question we raised in our outlook: where will the AI training infrastructure build-up slow or stop, especially for training?

The subsequent recovery did not provide clear answers, though the muted reaction to Nvidia's results was in stark contrast to previous years. In our view, debating about whether DeepSeek is actually using large amounts of Nvidia GPUs or not miss the larger point: the true revolution is the model’s efficiency, immediately measurable in the drop in operating costs for inference workloads. Such a change has already triggered a pricing war within the Chinese AI ecosystem.

This trend validates our view that maximizing efficiency will be the top priority for hyperscalers, especially for inference tasks, and achieving this will require custom chips. Strong results from Broadcom confirmed the momentum behind such custom silicon, while Marvell underperformed, largely due to management’s decision not to clarify their guidance for this specific segment ( although such a gap may be addressed during their “Future of Custom Silicon Technology for AI Infrastructure” event on June 17). Interestingly, Nvidia has also started to emphasize that inference now represents a growing share of workloads handled by its GPUs.

Political uncertainties 

This was arguably the point with the most uncertainty. We had correctly flagged tariffs as a risk, although we clearly did not foresee the scenario that unfolded on “Liberation Day”. While tensions have since eased somewhat, the situation remains fluid and subject to rapid change. Despite the pause in escalation, tariffs continue to loom as a threat, even though semiconductors are, for now, still exempt. U.S. policy has already created headwinds for some businesses, with chip export restrictions to China costing Nvidia and AMD significant revenue.

Conversely, many of the other concerns raised in our 2025 outlook either remain unresolved or have proven less impactful. Trump repealed Biden’s AI executive order, but retained several key restrictions in a different format. As anticipated, business realities around data use are taking precedence: The New York Times has signed licensing deals with both CNN and Amazon, though the case involving OpenAI is still ongoing. Antitrust scrutiny has not subsided, and the eventual rulings could carry meaningful consequences. Meanwhile, the feared politicization of the Justice Department did materialize, but its market impact has so far remained limited.

Still, our final message remains as relevant as ever: in this fast-evolving environment, investor success will depend on responsiveness and adaptability.

Catalysts

• Innovation trigger. DeepSeek turbocharged momentum in AI development, reaffirming that “there has never been a better time for breakthroughs to happen". More innovations are sure to follow. 

• Deregulation. The new U.S. administration repealed prior regulations but tasked agencies with setting standards across key sectors. These upcoming frameworks could significantly reshape industry trajectories. 

• Tighter implementation. Constraints within current frameworks limit AI’s full potential. However, rapid progress, particularly around AI agents, suggests mass deployment may soon unlock broader efficiencies.

Risks

• Hardware woes. Proposed U.S. tariffs and trade policy could increase hardware costs, potentially capping AI growth. They have also fueled geopolitical tensions, further disrupting supply chains. The situation remains volatile and unpredictable.

• Data wall. Market forces seem to be outpacing legal barriers, but access to high-quality data is becoming scarcer. This forces companies to explore alternative data sources and models.

• Demanding investors. The opportunity set is massive, but so are expectations. H1 results showed that companies with weak execution or failure to meet lofty forecasts faced sharp market penalties.