AI, Capex, and the Shifting Foundations of Investment Research

Three Transformative Trends

Three intersecting trends are transforming the way equity analysts approach fundamentals. Artificial intelligence is accelerating the demand for semiconductors and forcing a re-evaluation of digital business models and financial modelling assumptions.

As adoption grows, legacy frameworks built around static depreciation schedules, long-lived moats, and search-based monetisation begin to break down. Analysts and investors must evolve with them.

Semiconductor Capex: An Unprecedented Boom

The semiconductor industry is entering a historic investment cycle. Combined capex by major hyperscalers like Microsoft, Google, Amazon, and others is projected to exceed $350 billion this year, with a path toward $400 billion by 2030.

Microsoft alone could contribute roughly $85 billion. Much of this is directed at building out generative AI infrastructure, particularly through large-scale GPU purchases from Nvidia.

This raises serious questions about how to model depreciation. Traditional three-to-five-year asset lives may be unrealistic when hardware cycles shrink, and replacement comes faster. Standard accounting treatments overstate margins and understate economic reality if the equipment becomes materially obsolete within 18 to 24 months.

Google's Strategic Crossroads

Google faces a more existential disruption. As user behaviour shifts toward AI chat interfaces, the search engine's role in the digital experience is no longer guaranteed. This challenges its foundational monetisation model: search-based advertising.

Although Google's AI capabilities remain technically strong through its Gemini platform, public adoption has lagged behind competitors. The company is navigating a classic innovator's dilemma: invest in a future that might cannibalise its most profitable business.

The outcome depends on whether AI-enhanced search can generate more valuable advertising through greater personalisation. If not, the erosion of user intent data may impact its pricing power and ad yield.

A Wake-Up Call for Investment Frameworks

Traditional investment research is struggling to keep pace. Models that rely on linear assumptions about margins, capex efficiency, and R&D amortisation no longer match the operational realities of high-velocity AI investment cycles.

Equity research has long been anchored in structured financial data and deterministic spreadsheet modelling. However, as qualitative data grows in influence—management language, strategic tone shifts, non-obvious competitive moves—analysts need tools to interpret large volumes of context-rich information.

Natural language processing, sentiment analysis, and machine learning inference are no longer niche capabilities. They are now foundational to building a complete view of company performance and strategic trajectory.

Apple and Meta: Diverging Paths on AI Spend

Not every major player is joining the GPU arms race.

Apple remains focused on device-side intelligence and continues to minimise capex exposure by outsourcing infrastructure. Meta has made AI core to its ad platform but does not match hyperscalers in GPU procurement.

This divergence reveals varying capital allocation philosophies. Apple's cautious approach may preserve returns but also risks lagging in platform development. Meta's ROI on AI is closely tied to maintaining advertising efficiency in a shifting competitive landscape.

For analysts, the contrast highlights the need to assess AI positioning not just through spend levels but through the context of business models, margins, and competitive timelines.

AI in Equity Research: A Short-Term Edge

Today, AI-enabled tools offer a meaningful edge in the investment research process. Analysts can process earnings transcripts, filings, and commentary at scale, identifying risks and trends that would otherwise go unnoticed.

But that advantage will not last.

As adoption increases, speed and coverage become standard. Differentiation will once again come from how insights are framed and how quickly analysts can pivot to ask better questions. The key is combining machine-level coverage with human-level interpretation.

This is a narrow window. Early adopters stand to benefit, while laggards risk losing informational alpha.

From Impossible to Workable

Tasks once deemed impossible, such as mapping every AI mention in filings benchmarking strategy changes across five years of earnings calls, are far less complicated.

This shift requires a broader rethink within research organisations. CIOs and research leads must consider new workflows, resourcing strategies, and tech stacks. Automation is not replacing human analysts; it is removing the overhead so they can focus on what truly matters—judgment, synthesis, and communication.

Final Thoughts

AI is more than a trend. It is actively redrawing the investment landscape.

From semiconductors to advertising, from capital cycles to cash flow modelling, the assumptions underpinning the last decade of analysis are breaking. Investors who recognise this early and rewire their frameworks accordingly will have a significant advantage.

This is not just about new tools. It is about asking different questions, building different models, and seeing change clearly before it becomes consensus.