AI’s Market Dynamics, Risks, and the Road Ahead
PART II
Artificial intelligence has moved beyond proof-of-concept into operating budgets, data centres, and national strategies. The commercial stack now spans SaaS applications, licensed models, hyperscale infrastructure, and data services, while policy and power markets shape costs and growth. The sections below describe how value is created, where constraints appear, and why geopolitics matters supported by recent sources.
How AI Generates Value: The Commercial Stack
1) Software-as-a-Service (SaaS)
Cloud-delivered AI software packages model capabilities into recurring subscriptions. Typical outcomes include faster deployment cycles and measurable telemetry for usage-based pricing. This layer expands as enterprises standardize on copilots, search/answering, contact-centre assistants, and document automation across departments.
2) Model Licensing and APIs
Foundation-model providers and specialty developers expose capabilities via APIs and enterprise licenses. In this model, intellectual property (model weights, data, safety tooling) and access guarantees form the moat. The approach is common for LLMs and image/audio generation, where customers embed inference into their workflows and pay per token, request, or seat.
3) Infrastructure and Chipmaking
Accelerators and interconnects remain the physical engine of training and inference. Reports consistently place Nvidia as the dominant accelerator vendor, often estimated near ~80% share of AI accelerators, with AMD and Intel expanding supply in inference and specialized form factors.
Risks and Constraints: The Other Side of Growth
1. Ethics and Bias
Models trained on historic data can reproduce inequities. Financial decisions, hiring, and healthcare triage highlight sensitivity to false positives/negatives. As model use scales, demand rises for auditable datasets, documented evaluation sets, and post-deployment monitoring that detects drift and disparate impact.
2. Regulatory Uncertainty
Policy is catching up. The EU AI Act sets the most comprehensive template so far, with staged obligations for governance, transparency, and safety testing (including for GPAI). Several other jurisdictions continue to issue vertical guidance, which creates a patchwork that enterprises map into internal controls and reporting.
Geopolitics and Competitive Landscape
1. United States vs. China
The U.S. emphasizes private-sector dynamism; China pursues state-led self- reliance. Export controls introduced in Oct 2022 and tightened in Oct 2023, and Dec 2024 restrict advanced AI chips and tooling to Chinese entities; allied alignment and additional actions in 2025 further limited transfers, shaping supply, costs, and model performance ceilings in restricted markets.
2. European Union
The EU’s strategy centres on “trustworthy AI” via the AI Act prioritizing safety, transparency, and governance. Critics argue that compliance overhead may slow iteration; supporters point to legal certainty and risk reduction for high impact use cases.
3. The Global Divide
World Bank commentary highlights the risk of an “AI divide,” were infrastructure, skills, and capital determine productivity gains, potentially widening gaps across developing economies. Case studies also note productivity uplifts in targeted pilots (e.g., 14% gains for
call-centre agents using gen-AI assistance), suggesting uneven but real benefits when capability and context align.
Capital Flows and Enterprise Integration
Private investment remains elevated. The Stanford HAI 2025 AI Index reports $109.1B in U.S. private AI investment in 2024, far outpacing other regions; generative AI alone attracted $33.9B globally, up 18.7% year over year. Business usage rose to 78% in 2024, from 55% the prior year, indicating rapid mainstreaming. Budget-modelling work at Wharton suggests AI could lift productivity and reduce deficits over multi-decade windows as integration deepens and costs normalize.
Where the Cycle Stands
Analysts characterize 2025 as a pivot from exuberance to operational focus. Gartner’s 2025 Hype Cycle notes generative AI entering the “Trough of Disillusionment,” where expectations reset and foundations (AI-ready data, agents, guardrails) receive emphasis. This typically coincides with sharper ROI scrutiny and clearer program governance.
What Financial Observers Monitor
- Unit Economics: token costs, context-window efficiency, caching/quantization, acceptance rates, and human-in-the-loop labor per
task. - Utilization & Mix: share of spend on training vs. inference; on-prem vs. cloud; regional cloud mix and sovereign requirements.
- Regulatory Milestones: EU AI Act checkpoints (2025–2027), sectoral guidance in finance/health, and cross-border data rules
- Supply Signals: accelerator availability, memory bandwidth roadmaps, and foundry capacity; export-control updates affecting vendor assortments.
- Energy & Location: PUE trends, grid mix, and siting near low-carbon baseload or curtailed renewables; colocation near fibre routes to reduce latency.
- Adoption Indicators: share of employees with AI seats, percent of workflows grounded on enterprise data, and evaluation dashboards tracking factuality, bias, and safety incidents.
Conclusion
AI has matured into a layered commercial system: SaaS products that users touch, licensed models that power them, infrastructure that supplies compute at scale, and data-governance services that make usage viable. The same factors driving AI growth big investments, new rules, and the race for computing power also create risks like privacy issues, bias, high energy use, and market concentration. Near-term sentiment reflects a shift from hype to durable value creation, consistent with cycle theory and with measured, rule-aligned deployment. Over the next decade, outcomes will hinge on three variables: affordable compute, trustworthy governance, and the breadth of real workflows that AI can reliably improve trends that current investment, regulation, and energy data already illuminate.