14 Jul 2026
As AI adoption continues to accelerate, the focus is increasingly shifting from technology to economics. While the market is currently rewarding cyclical AI infrastructure beneficiaries, questions around future capacity expansion, monetisation, return on investment and competitive positioning are becoming more important. Against this fast-moving backdrop, Hyun Ho Sohn, portfolio manager of the FF - Global Technology Fund, talks through the key takeaways from his recent Silicon Valley trip.
Fidelity’s global technology research team recently completed its 13th annual Silicon Valley trip. Over several days in the Bay Area, we met more than 20 companies across semiconductors, internet software and frontier AI labs. These trips are an important part of my investment process, as they give us direct access to management teams and allow us to test market narratives against what companies are seeing on the ground. This helps us understand where technological change is creating real business value. Fidelity’s global research platform is a key advantage in this process, bringing together company access, sector expertise and cross-market perspectives across the technology value chain.
Hardware companies were bullish on AI demand remaining strong, which was broadly in line with expectations. Continued hyperscaler capital expenditure is translating into robust order activity across the broader technology value chain, with demand extending across a wider part of the infrastructure stack. The bottleneck is no longer limited to GPUs. Companies highlighted constraints across CPUs, memory, advanced packaging, substrates and optics. From a cyclical perspective, this remains supportive for a broad range of technology suppliers.
However, supply is also responding to this robust demand. Across several parts of the semiconductor ecosystem, companies are expanding capacity. For example, an optical transceiver manufacturer highlighted that it aims to double transceiver capacity each year. Another semiconductor equipment manufacturer confirmed that dynamic random-access memory (DRAM) producers are placing strong orders for wafer fabrication manufacturing equipment, at almost double the levels seen over the past two years.
While robust demand is driving a strong near-term outlook for many of these cyclical businesses, I believe that investors often underestimate the cyclical risks in certain parts of the semiconductor industry, such as memory. Unlike many other areas of the technology stack, memory pricing is highly dynamic and largely driven by the balance between supply and demand. When conditions are tight, vendors can raise prices quickly. However, when the cycle turns, the downturn can be significantly more severe.
Memory manufacturing is a high fixed-cost business, and producers have limited ability to control industry supply once capacity has been built. At the same time, technological advances continually improve density and lower production costs, creating a natural tendency for prices to decline over time. This dynamic discourages both producers and customers from holding excess inventory, as tomorrow’s memory is typically cheaper than today’s. Thus, periods of shortage can quickly give way to oversupply, making memory one of the most volatile areas of the semiconductor market despite the current strength in AI-related demand.
Another theme was the shift from ‘tokenmaxxing’ towards ‘valuemaxxing’. Earlier in the year, companies were encouraging employees to maximise AI usage, with token consumption often viewed as a measure of innovation and productivity. More recently, however, companies have become increasingly focused on the cost of that usage and the value being generated from it.
A useful example came from a consumer internet company, which reportedly exceeded its planned AI token budget much earlier than expected. While management noted some efficiency benefits, particularly in software development, the return on investment remains mixed and the revenue impact is not yet clearly visible. As a result, the company is continuing to monitor token costs and expects to assess the benefits more fully over the next 12 months.
We heard similar messages from other companies. AI is helping to shorten software development cycles, improve customer-service automation and generate efficiency gains in internal workflows, but very few companies can yet quantify these benefits. Several management teams acknowledged that the return on investment remains directional rather than fully measured at this stage.
This uncertainty is important to monitor closely. A significant portion of current demand is still driven by trial activity and subsidised usage, reinforced by aggressive pricing moves as AI labs consider price cuts to encourage adoption and defend market share.
Software companies and developers are using tokens to experiment and build new applications, while many end users are consuming AI services because they remain free or heavily subsidised. As cost concerns rise and companies become more selective in their usage, this could constrain the revenue-generating potential of AI model providers and application-layer software companies, raising broader questions around the sustainability of demand and the ultimate return on investment.
The impact of AI on software was another key debate during the trip. In my view, AI will be a threat for some companies and an opportunity for others.
For some business models, AI creates challenges. It is making it faster and cheaper to write code, lowering some barriers to software development. However, software is much more than code alone. Building effective applications still requires a deep understanding of business processes, workflows and customer needs, while deployed software must be integrated, secured and maintained over time.
The market is increasingly challenging the terminal value of many software businesses. While this may prove true for some categories, I believe the reality is likely to be more selective. Companies with recurring revenue models, mission-critical products and deeply embedded customer relationships may be more resilient than current market expectations imply. Writing code is becoming easier, but replicating trusted enterprise software platforms remains far more difficult. In areas such as finance, enterprise resource planning and customer relationship management, purchasing decisions are often driven as much by trust, integration and long-standing relationships as by the underlying technology itself.
This is where active, fundamental research becomes particularly important. Broad market indices and passive strategies often treat software as a single category. In reality, the impact of AI is likely to vary significantly between companies, making a bottom-up approach essential to identify differentiated outcomes. Some business models may face disruption, while others may emerge stronger by embedding AI into existing customer relationships and workflows.
I continue to believe that AI remains a powerful structural growth driver; however, it is important to remain mindful of the cyclical dynamics underpinning much of the current wave of investment.
The scale of capital flowing into the ecosystem continues to increase. Recent capital-raising activity, including equity issuance by large technology companies such as Alphabet, debt issuance by companies such as Nvidia, and the prospect of AI labs coming to market through IPOs, is providing further fuel for continued investment. While this inflow of capital may help sustain momentum in the near term, it also raises concerns about the dilution of returns on these investments.
In the shorter term, we are also seeing a market dynamic where momentum-driven and lower-quality businesses are outperforming higher-quality companies which have dominant market positions. Interestingly, Nvidia, despite being a high-quality company and a market leader over the past two years, is now underperforming the broader tech sector.
These dynamics reinforce the need for a disciplined and selective approach. While the opportunity set remains compelling, I believe it is important to balance enthusiasm for AI’s long-term potential with caution around the cyclical nature of the market and the sustainability of returns across the ecosystem.