Technology Sector: AI Infrastructure and the Next Wave of Growth

The major hyperscalers — the companies providing cloud compute, AI training infrastructure, and consumer AI applications — are committing to capital expenditure programmes that represent a step-change in the history of technology investment. Combined AI-related capital expenditure from the top five hyperscalers is expected to exceed $300 billion in 2025, with growth rates that suggest no near-term pullback. This level of sustained investment creates powerful and durable demand signals across multiple layers of the technology supply chain.

The buildout is proceeding in phases. The current phase is dominated by training infrastructure: the specialised compute clusters needed to develop frontier AI models. This phase is GPU-intensive and benefits primarily semiconductor designers and the high-bandwidth memory and networking infrastructure required to connect thousands of chips. The next phase — inference infrastructure — scales AI capabilities to billions of end users and requires a different, more distributed hardware architecture that opens opportunities for a broader set of semiconductor companies.

Data centre construction, power, and cooling infrastructure represent a multi-trillion dollar investment cycle that extends well beyond the technology sector. AI data centres require two to three times the power density of conventional compute facilities, driving unprecedented demand for electrical grid connections, power distribution equipment, cooling systems, and backup generation. The infrastructure required to power AI is creating investment opportunities across utilities, construction, and industrial equipment sectors.

Graphics processing units and AI accelerators sit at the apex of the AI infrastructure value chain. The dominant market position held by the leading GPU designer provides remarkable pricing power and margin potential, supported by the proprietary software ecosystem that creates switching costs for developers who have built workflows and models on their platforms. This combination of hardware performance leadership and software lock-in is rare and highly defensible.

Custom silicon — AI accelerators designed specifically by hyperscalers and large enterprises for their own workloads — represents a growing share of AI compute deployment. These custom designs optimise for specific inference or training tasks, offering cost and efficiency advantages over general-purpose GPUs for mature, stable workloads. The growth of custom silicon creates opportunities for chip design tool companies and specialised semiconductor IP providers, while creating competitive pressure for merchant silicon vendors over the medium term.

Networking infrastructure — the high-speed interconnects that link AI compute clusters — is a critical bottleneck in scaling AI training. The bandwidth requirements of large-scale AI training demand specialised networking solutions that are orders of magnitude more capable than conventional data centre networking. Companies with strong positions in high-performance networking for AI data centres are benefiting from high growth and strong pricing power.

While hardware companies have captured the largest share of AI value creation in the current phase, the software and applications layer represents the longer-term prize. Enterprise AI software — tools that help companies build, deploy, and manage AI applications — is growing rapidly as organisations move from AI experimentation to production deployment. The transition from proof-of-concept to enterprise-scale rollout is driving rapid growth in AI development platforms, MLOps tooling, and AI-enabled business applications.

Vertical AI applications — AI tools built for specific industries such as healthcare, legal, financial services, and manufacturing — represent large addressable markets with high barriers to entry from domain expertise and data requirements. Companies that combine deep domain knowledge with AI engineering capabilities are establishing durable competitive positions in these verticals. Healthcare AI, in particular, offers significant near-term commercialisation potential given the size of the market and the tangible productivity improvements AI can deliver.

The democratisation of AI capabilities through easy-to-use API-based platforms is enabling a new generation of AI-native startups and enabling incumbent software companies to embed AI into existing products. This creates both opportunity and competitive threat for established software vendors: those who successfully integrate AI to enhance product value and improve unit economics will strengthen their market positions, while those who fail to adapt risk disruption from AI-native competitors.

The AI infrastructure investment theme is real, durable, and large in scale — but not all companies will benefit equally, and valuations across the supply chain reflect meaningfully different risk-reward profiles. Companies closest to the AI capex spend (GPU designers, high-bandwidth memory providers, networking equipment companies) have the most direct and near-term earnings benefit but also carry the highest expectations and valuation risk.

The primary risk for AI infrastructure investment is a shift in capex prioritisation by the hyperscalers. Should AI monetisation fail to materialise at the pace required to justify current investment levels, the hyperscalers may moderate their capex programmes sharply, creating significant downside risk for hardware vendors whose revenue growth is highly dependent on this concentrated group of customers. Monitoring quarterly capex guidance and management commentary from hyperscalers is essential.

Regulatory risk is growing. Governments across the US, EU, and China are developing AI governance frameworks that may impose requirements on data handling, model transparency, and use-case restrictions. While regulation is unlikely to halt AI development, it may create compliance costs, restrict certain high-value applications, and advantage incumbents with regulatory affairs capabilities over smaller challengers.