The AI Arms Race: A Guide to Navigating the Noise

We are in the midst of what many describe as an “AI Super-Cycle.” It began with the release of ChatGPT in 2023, a significant technological breakthrough that quickly moved from early interest to widespread adoption, and has since evolved into intense competition among companies and governments. For investors, this environment is both exciting and challenging. Share prices of the largest technology companies have risen sharply, shaping behaviour across both corporate decision-making and capital allocation. 

Before responding to the excitement, it is worth stepping back to examine the industry's structure and to revisit past technology cycles. Understanding how these cycles tend to unfold can help separate long-term signal from short-term noise. 

Technology Cycles Revisited

Every major technology shift over the past 40 years, from the personal computer to smartphones to cloud computing, has followed a broadly similar structure. Each wave develops in distinct layers, with different levels of technological change, competitive advantage, and risk. Understanding where value tends to accrue within these layers helps clarify both the opportunities and the pitfalls for investors. 

1. Semiconductors (the foundation): Semiconductors are the essential inputs that make modern technology possible. Every device, data center, and AI model relies on these chips. This layer often benefits early in a technology cycle, as it provides the foundational building blocks. Proprietary standards and scale can offer some protection from competition, but leadership changes as technology evolves. Intel and AMD dominated the PC era, ARM and Qualcomm led in mobile, and today companies such as Nvidia and Broadcom play a central role in AI. These shifts make semiconductors inherently cyclical, with sharp swings in supply and demand that can create volatility even during long-term growth periods. 
2. Infrastructure, devices, and models (distribution): This layer functions as the roads and utilities of the digital economy, connecting users to technology and services. Building this infrastructure is capital-intensive and risky, particularly before standards and ecosystems are established. Competition is often fierce, and many participants fail before the market consolidates. Past cycles saw well-known names disappear as technology advanced. Today, large cloud providers such as Amazon, Microsoft, and Google operate the core infrastructure that supports AI. While scale provides advantages, the capital required to build and maintain these platforms increases the importance of disciplined investment and realistic return expectations. 
3. Software and services (applications): Software and services represent the user-facing layer, where people interact with technology on a daily basis. This layer evolves most quickly and often attracts the greatest enthusiasm, offering both high growth potential and high risk. Past cycles left many casualties before a small number of platforms achieved durable positions. Over time, companies that attract sufficient users and developers can benefit from powerful network effects. In AI, this layer remains unsettled. New models and applications continue to emerge, and leadership is still fluid as the market works toward longer-term winners. 

Red Flags and Lessons from Past Technology Cycles

Much of the recent increase in AI investment has been driven by what are known as “scaling laws,” the idea that investing more in computing power leads to more capable and effective AI systems. Capital spending on data centers is now estimated to exceed an annual run rate of $400 billion in 2025, nearly tripling since 2022. This represents roughly 1.4% of U.S. GDP. 

For context, this level of investment is still below that of several past technology buildouts. The railroad expansion accounted for roughly 3-4% of U.S. GDP, auto infrastructure around 2%, and the information technology and telecommunications boom just over 1%. While not definitive on its own, this comparison suggests the AI investment cycle may still have room to develop before reaching a ceiling. 

Sources: Bloomberg, Goldman Sachs

With the sharp increase in spending and rising valuations among large technology companies, it is natural to ask whether parts of the market are becoming overheated. History shows that technology bubbles often form when barriers to entry are low, innovation cycles are rapid, and companies borrow heavily to keep up.  

Today, several of these conditions are evident in AI infrastructure. Competition among infrastructure providers such as NeoClouds and AI models has intensified. Companies like Oracle, Meta, and CoreWeave are taking on more debt to fund expansion. OpenAI has announced plans for $1.5 trillion in spending, backed by equity commitments from Nvidia. At the same time, new AI models are emerging rapidly as advances lower barriers to entry. Taken together, these reinforcing trends resemble past technology cycles, where aggressive buildouts supported by creative financing ultimately led to excess capacity, unsustainable business models, and difficult corrections. 

Fear-of-missing-out (FOMO)

Human behaviour is a powerful signal of where we are in the market cycle, and one we monitor closely. Right now, technology leaders and investors appear to be driven by a fear of missing out (FOMO). Corporate executives justify aggressive spending to avoid being left behind, while investors worry about underperforming if they are not sufficiently exposed to large technology companies.  

This dynamic creates a form of prisoner’s dilemma, where capital continues to be allocated even as return visibility declines and supply risks outpace demand. The result is an environment where stretched valuations and competitive urgency reinforce each other, sometimes at the expense of rational decision-making.  

The Big Unknowns and Economic Reality

Network effects occur when a product or platform becomes more valuable as more people use it, like a messaging app that gets better when all your friends join. Early in a technology cycle, standards are less established, markets are fragmented, and switching between providers is easy. As adoption grows and standards begin to settle, network effects start to matter, making it harder for new competitors to catch up and allowing a small number of dominant winners to emerge, as we have seen with platforms such as Nvidia’s CUDA, Apple’s iOS ecosystem, Amazon’s marketplace, and Google’s search engine.  

AI is still very early in the technology cycle. Core concepts such as “super-intelligence” remain loosely defined, market leadership continues to shift, and disruption is very much still on the table.  History suggests that even when a handful of long-term winners eventually emerge, there are usually many more casualties along the way, well before investors gain any real visibility. 

Value and economic capture

The disconnect between value creation and economic capture is another major unknown. We see this gap at both the macro level and the company level. While large technology firms are spending hundreds of billions on AI infrastructure, U.S. GDP growth outside of data center investment has been closer to 2%. 

Source: Financial Times

Our internal analysis suggests that more than $3 trillion in AI-related revenue, roughly 11% of incremental U.S. GDP, would be required to generate adequate returns on the scale of infrastructure now being built. At the same time, earnings growth for the broader market, measured by the S&P 500 excluding the top seven technology stocks, has been modest, at roughly 2% over the past three years. 

Source: Data from Bloomberg

ROI (return on investment) in AI is still unclear. Despite rapid growth in usage, most companies haven’t shown they can turn that demand into lasting profits. Many are prioritizing market share and innovation over immediate financial returns, so investors don’t yet have visibility on whether these investments will deliver sustainable, long-term economic value.  We worry that this will create a significant oversupply in AI infrastructure, destroying economic value for all stakeholders involved.   

All of this raises a fundamental question. Is AI acting as a true multiplier for economic growth, or is it becoming a capital-intensive race to the bottom? 

Factors that may delay a bubble

While there are signs that the technology cycle may be approaching a turning point, several factors could slow the formation of a widely feared AI bubble. Physical constraints, such as power availability and land, mean that expanding data center capacity takes time. Chip supply is also limited by highly specialized manufacturing, much of it governed by TSMC. Higher interest rates further reduce the availability of cheap capital, forcing builders and investors to be more selective. 

In addition, most AI spending today is concentrated among financially strong technology firms with established earnings, which reduces the risk of a severe and sudden fallout. Continued innovation also suggests the AI cycle remains relatively early, with further demand still to come. Finally, technology valuations, while elevated, remain more reasonable than those seen in past bubbles. 

Nicola Wealth's Approach to Investing in the AI Era

Many clients gain technology exposure through our Nicola funds, particularly through our U.S. Equity Fund. As a result, our underweight position in technology warrants explanation. More importantly, it provides an opportunity to share how we think about major technology cycles and how we navigate them as investors. 

We believe the AI cycle is still in its early stages, with an unsettled competitive landscape and a growing risk that excess infrastructure spending could weigh on future returns. Rather than attempting to predict outcomes that remain highly uncertain, we rely on a consistent investment approach focused on business quality, capital allocation, and investing at reasonable valuations. Our strategy emphasizes strong fundamentals and diversification across the technology stack to manage risk over time. We aim to invest in companies that can generate attractive risk-adjusted returns over a five-year horizon, while recycling capital as better opportunities emerge.

How we are positioned

• Semiconductors – Within AI, semiconductors represent what are often referred to as the “picks and shovels” of the cycle. Rather than relying on which applications ultimately succeed, these companies supply the essential components required to build and operate AI infrastructure. Our exposure includes Nvidia, which provides computing hardware, Broadcom, which designs custom chips, Synopsys, which supplies chip design software, TSMC, which manufactures advanced chips, ASML, which produces critical lithography equipment, and memory suppliers such as Samsung and SK Hynix. Together, these businesses provide the foundational tools needed to power AI within data centers. 
• Distribution: We are invested in the major hyperscalers, including Microsoft, Google, and Amazon, which build data centers and distribute AI services at scale. These companies benefit from existing platforms, strong balance sheets, and the ability to fund large infrastructure investments internally. We have chosen not to own Oracle, where we see more commodity-like economics and greater reliance on external financing to support expansion. 
• Software and services: In the applications and services layer, we focus on companies with established products and recurring demand, where AI is likely to enhance existing offerings rather than depend on unproven business models. Our holdings include ServiceNow, Adobe, Netflix, Meta, and Gartner. These companies are positioned to benefit as AI adoption increases across enterprise and consumer use cases.  
• Power and energy: We also maintain exposure to companies that support the growing power and infrastructure needs of data centers. This includes Schneider and Siemens, which manufacture critical components used to deliver, manage, and stabilize power as demand from AI-related infrastructure continues to rise. 

Our greatest strength as investors lies in understanding our limits. In an AI-driven environment, the ability to predict earnings trajectories and long-term competitive positioning becomes increasingly difficult. As a result, we prefer to allocate capital to companies with clearer earnings visibility, even if that means remaining cautious on parts of the technology sector for a period of time. 

The patience of our clients aligns closely with this approach and remains central to our ability to protect and manage capital through changing market conditions. With that foundation, our objective is to deliver consistent, long-term results, regardless of how unpredictable the technology cycle may become.