The past few quarters have underscored a clear transition: artificial intelligence is moving from sprint to scale. Model capability continues to set new standards, usage is broadening and the capital expenditure behind infrastructure is rising faster than forecasts made even six months ago.
AI is also shifting from headline launches to operational integration. Enterprises are beginning to show early productivity and product cycle gains in their reported numbers, widening the gap between enablers and early adopters and those firms facing growing disruption risk. A notable advance has been the emergence of longer-horizon, deeper research modes, giving systems more tokens, tools and time which materially improve accuracy on complex, multi-step tasks.
In parallel, agentic capabilities are accelerating into systems that plan, browse, write and run code then act against objectives. Industry leaders expect the next demand wave to be dominated by inference at a vastly higher scale as agents proliferate across workflows and users, a shift that further tightens supply and makes power and networking the new binding constraints.
AI in everyday workflows
Usage is compounding on two fronts: more users and higher consumption per user as AI embeds into production systems. Leading assistants have added substantial users year-to-date, while tokens per use are rising as reasoning-heavy products roll out. In software development, AI code editors and copilots are now responsible for a meaningful share of net new code at major vendors, compressing release cycles. On consumer platforms, AI-powered ranking and creative tools are lifting engagement and conversion.
Crucially, recent earnings commentary points to tangible operating benefits: higher first-line resolution in customer support with AI agents, shorter sales timelines where AI prospecting and drafting are used and back office automation delivering opex leverage ahead of revenue from AI-native products. These signals are appearing first through cost and cycle-time improvements rather than headline revenue, especially outside technology.
Behind this, the investment cycle remains in full expansion. Core US hyperscalers collectively spent well over $100bn in Q2, up sharply year-on-year, and 2025 plans have been revised higher again. Street estimates now point to 2026 AI-related spend approaching $0.5trn when including data centres, power and ancillary systems – reflecting sustained demand and elongated build times. Management commentary across Microsoft, Alphabet, Amazon and Meta Platforms remains consistent: demand exceeds capacity and power availability is a gating factor for deployments. Sovereign programmes add another layer, with national AI initiatives now accounting for tens of billions of dollars in annualised spend across Europe, the Middle East, India and parts of Asia. Together, hyperscaler, enterprise and sovereign demand support a multi-year capex cycle, not a 2025 peak.
AI is also shifting from headline launches to operational integration…widening the gap between enablers and early adopters and those firms facing growing disruption risk
As with prior platform shifts such as smartphones and cloud computing, cheaper and better AI is creating ‘invisible opportunities’ that are hard to size from current forecasts but may prove much larger than the initial use-case set. We see this in the emergence of AI-native tools that are rapidly scaling revenue and market share, while platforms monetise usage through cloud computing and advertising uplift.
Conversely, disruption is intensifying where AI reduces competitive advantages or disintermediates workflows. Creative services and parts of information services are encountering fading growth outlooks and pricing pressure; many software offerings are facing questions on their terminal value as AI offers the opportunity of custom-built tools for specific tasks. More fundamentally, we expect AI to transition application software from being a tool that helps knowledge workers be more productive to one that completes tasks that need to be done without human involvement.
The next wave of investment
From an infrastructure standpoint, the bottlenecks are shifting – power, networking and physical plants are now major constraints. The investable value chain therefore extends beyond compute to high-bandwidth memory and advanced packaging, optical interconnect, liquid cooling, energy procurement, grid interconnects and new forms of power generation. AI output improves most when systems are allowed more time and computing resource and we expect a broader adoption of reasoning modes for high-stakes tasks and a rise in background agents – both support sustained inference intensity and therefore the further buildout of AI infrastructure.
The Polar Capital Artificial Intelligence Fund’s positioning is consistent with these dynamics. We maintain core exposure to infrastructure leaders, technology enablers and hyperscalers benefiting from the elevated capex cycle. This is complemented by application and end-market holdings where AI is directly improving productivity and product differentiation.
We remain cautious on business models where AI undermines legacy pricing or add-on monetisation and continue to recycle capital from AI-adjacent names into companies demonstrating measurable AI-driven economics. The outlook is constructive but selective: the opportunity set is expanding, dispersion is rising and stock selection grounded in usage data, supply chain signals and customer adoption will matter more than ever.
