Opinion: The economics of AI

The concept of "free" services is used extensively by large entities to enter markets, often creating dependencies that can be monetized later. However, this designation is frequently a misleading market signal because service operators face significant underlying costs. The term "free" also functions as a barrier to competition, forcing rivals to match those terms to enter or remain in the market. Historical examples show that startups sometimes engage in price leading or dumping to gain market share, with subsequent funding often subsidizing rapid growth.
The current landscape of AI infrastructure is highly expensive due to the massive computational requirements for Large Language Models (LLMs). Building AI data centers requires extensive physical plant, including GPU clusters, high-speed interconnects, and substantial power infrastructure, demanding significant investment in real estate, cooling, and grid interconnection that exceeds typical utility models. This capital expenditure is being financed through complex arrangements involving hyperscalers, infrastructure funds, banks, and private equity, which creates layered risk due to dependencies on uncertain AI demand and technological evolution.
Financially, the total annual operating cost for AI services is estimated to be between USD 2 to 4 trillion, far exceeding the digital advertising market value. The sustainability of this model hinges on whether future advancements in silicon technology will lead to lower operational costs or if monetization strategies will shift away from consumer subsidies. Revenue generation may rely on shifting costs to employers or bundling AI capabilities into existing products, while premium subscription services remain dependent on consumers' broader economic well-being.

* Google used "free" services for search, Gmail, and Docs.
* Meta used "free" for its Facebook platform; Cloudflare used it for web caching.
* "Free" is presented as a method to expose products to the market, aiming to create later monetization dependency.
* Service operators incur real costs that must be covered by service operators.
* Uber posted a USD 5B operating loss in three months in 2019 and took thirteen years to achieve revenue positivity.
* DoorDash managed a USD 1.4B loss in 2022, with deliveries subsidized by financial backers.
* AI costs stem from the computational requirements for Large Language Models (LLMs).
* A modern AI data center requires approximately 60,000 GPUs, high-speed mesh connectivity, mass storage, and substantial power delivery (150kW to 200kW per rack).
* AI infrastructure investment is estimated at USD 750B in 2026 for capital expenditure.
* Aggregate capital expenditures by hyperscalers rose from USD 97B in 2020 to over USD 400B in 2025 and are projected to exceed USD 800B in 2026.
* AI investment involves data centre developers, infrastructure funds, private equity, banks, and bond markets.
* The operational cost of AI services is estimated to be between USD 2 to 4 trillion annually.

The narrative constructs a tension between the superficial promise of "free" access and the underlying reality of massive capital intensity and risk exposure in the AI sector. The analysis highlights that "free" services function less as altruistic offerings and more as strategic market entry tools designed to establish dependencies, echoing historical patterns seen in platform economics where initial subsidy fuels exponential growth necessary for later monetization.
The core mechanism of concern is how immense physical infrastructure requirements (data centers, power grids) are financed and how those costs—and associated risks from technological obsolescence and demand uncertainty—are distributed across complex financial instruments. The pattern observed is that leveraging long-duration debt against uncertain future AI demand transfers risk downstream to tenants, asset owners, and creditors. This suggests a systemic reliance on continuous growth and perfect technological progression to sustain the current capital structure, which introduces profound fragility if assumptions about Moore's Law or market demand shift.
The subsequent pivot toward funding—shedding labor costs via layoffs rather than achieving profitability directly through service monetization—reveals an institutional attempt to manage unsustainable capital demands by externalizing the financial burden onto the workforce and potentially societal structures. The juxtaposition of the potential for technological deflation against the risk of stagnation in silicon evolution suggests a critical inflection point: whether AI will trigger a sustainable economic shift or merely accelerate existing financial instabilities by increasing the scale of necessary, yet unproven, infrastructure investment. What is the true cost of this "acceleration" on long-term stability versus short-term market capture?