Semiconductors: a new phase in the AI era

The main recent development in AI is that the semiconductor industry is moving from a chip-focused model to an integrated-system approach. GPUs remain essential due to their strength in parallel processing, but they are no longer the only critical component. Within these systems, CPUs are playing increasingly important roles. CPUs act as the “orchestrators,” managing data flows and coordinating between different components.

Also, memory is emerging as a critical component in AI-infrastructure. Especially High Bandwidth Memory (HBM) has become a bottleneck, as AI performance increasingly depends on how quickly data can be accessed and processed. In addition, there are ASICs, which are specialized chips designed for specific tasks. These chips are also gaining traction in inference workloads, where efficiency and cost per query are key. Hyperscalers such as Meta and Alphabet are increasingly adopting ASICs to improve performance and reduce costs.

Together, these components form tightly integrated systems in which performance depends on how well compute, memory and connectivity work together. As a result, value creation is moving beyond individual chips toward full systems, including networking, packaging and software. This increases complexity, but also expands the overall market opportunity. What began as a GPU-led cycle is now extending into CPUs, memory and custom chips.

From a market perspective, the semiconductor sector continues to be attractive, although the easiest phase of the rally - mostly driven by GPU growth - is likely behind us. Valuations have moved higher as investors have increasingly priced in strong AI-driven demand, but the story goes beyond simple momentum. This cycle is underpinned by structural growth, as AI spending continues to expand across hyperscalers and demands more complex, system-level architectures. Importantly, the opportunity set is broadening.

This diversification of demand suggests that earnings growth will gradually become more balanced across the sector, which is typically a constructive sign in a longer-term bull market. At the same time, after a period of strong performance, parts of the sector may look somewhat stretched in the near term, and phases of consolidation would be a natural development.

Overall, we believe the sector remains investable, although selectivity is becoming increasingly important as the cycle matures. The combination of strong demand visibility, expanding end-markets and increasing value capture throughout the ecosystem supports a constructive outlook, particularly for companies that are well positioned within the evolving AI landscape.

The AI boom is evolving into a multi-layered ecosystem, where CPUs, ASICs and memory are becoming increasingly important alongside GPUs In this new landscape, success will not be determined solely by the fastest chip. Instead, the winners will be those that can integrate compute, memory, networking and software into complete and efficient systems. As such, AI is no longer just a GPU story, but a structural transformation of the entire semiconductor industry.

For investors, this shift reinforces the attractiveness of the sector, while at the same time requiring a more selective approach. As value creation becomes more distributed across the ecosystem, opportunities extend beyond the early leaders, but differentiation between winners and laggards is likely to increase.

Joost Olde Riekerink
Global Equity Research & Advisory Expert