AI valuations are tripling while unit Economics remain broken. Google, OpenAI and Big Tech are spending at historic scale on AI. The monetisation model has not caught up.

Key Highlights

  • Google's monthly token consumption has surged from 480 trillion to 3.2 quadrillion in one year, pressuring margins
  • Alphabet's Capital-expenditure/">Capital Expenditure is projected at up to $190 billion in 2025, six times the level of four years ago
  • OpenAI, Microsoft, Meta, and Anthropic face structurally similar cost-versus-Revenue mismatches
  • Monetisation remains unresolved, with subscriptions, Advertising, and API fees each carrying material limitations
  • The expansion of consumer-facing AI agents risks accelerating cost curves faster than revenue models can absorb

An Industry Betting Heavily on Demand That Has Not Yet Paid

The artificial intelligence sector is entering a phase where infrastructure spending is no longer theoretical. It is measurable, contractual, and in many cases, already committed. What remains uncertain is whether the demand side of the equation will scale at a comparable pace.

Google's annual developer conference this month offered a useful window into how the industry's largest integrated player is thinking about this challenge. Sundar Pichai unveiled a suite of consumer-facing AI agents powered by the Gemini 3.5 Flash model, designed to be embedded across both the Gemini app and Google Search, reaching a combined user base of over three billion. The strategic logic is clear: distribute AI broadly, capture usage data, and eventually convert that scale into durable revenue. The execution risk is equally clear: distributing AI broadly is extraordinarily expensive.

Alphabet (Nasdaq:GOOGL) now consumes 3.2 quadrillion tokens monthly, up from 480 trillion a year earlier. Each token requires compute, and compute requires capital. The company's projected capital expenditure for 2025 stands at up to $190 billion, a figure that would have been considered implausible even four years ago. Critically, that capital is also buying less than it once did. Energy costs, chip procurement, and data centre construction have all risen materially. Efficiency gains from model optimisation can partially offset these pressures, but they are unlikely to close the gap entirely at current growth trajectories.

A Structural Problem, Not a Google Problem

It would be a mistake to frame this as a challenge unique to Alphabet. The cost-versus-monetisation tension is industry-wide, and each major player is navigating it from a structurally different position.

OpenAI, the pioneer of consumer AI through ChatGPT, remains deeply unprofitable despite its commercial momentum. The company raised $40 billion in a SoftBank-led round in early 2025 at a $300 billion valuation, a figure that has since nearly tripled following a subsequent $122 billion raise that values the Business at approximately $852 billion as of early 2026. That the private valuation has almost tripled in under a year while the company continues to burn cash is itself a pointed illustration of the gap between capital market sentiment and underlying unit economics. Its strategic response has been to migrate toward higher-Margin enterprise products and to introduce premium consumer tiers, while its research pipeline has shifted meaningfully toward agentic coding tools. The risk is that repositioning upmarket cedes the consumer segment to better-capitalised rivals.

Microsoft (NASDAQ: MSFT) absorbed deep OpenAI integration into its Copilot product line at considerable cost, only to find early enterprise adoption underwhelming. The Investment/">Return on Investment for many business customers remained unclear, and the company was forced to revise some feature rollouts. The dynamic illustrates a broader problem: willingness to pay for AI features does not automatically follow from availability of those features.

Meta (NASDAQ: META) occupies a more defensible position in the near term. Its AI investments are underwritten by a highly profitable advertising business, and its deployment model, distributing Llama-based tools freely across its owned platforms, avoids the direct monetisation problem by treating AI as an engagement multiplier rather than a standalone revenue line. Whether this approach generates durable Competitive Advantage or simply defers the monetisation question remains to be seen.

Anthropic, the developer of the Claude model family, operates primarily through API licensing to enterprise and developer customers. Its consumer distribution is limited compared to Google or Meta, and its capital requirements are funded substantially through investment commitments from Amazon (NASDAQ:AMZN) and Alphabet. Its position is technically credible but commercially more vulnerable to shifts in enterprise procurement cycles.

Apple (NASDAQ:AAPL), whose own AI features are now partially powered by Google's Gemini models, sits in a distinct position: it has outsourced material parts of its AI infrastructure rather than bearing the full capital burden internally. Whether this Partnership arrangement remains commercially favourable as AI capabilities evolve is a question the market has not yet fully priced.

Three Levers, None of Them Sufficient Alone

The industry has identified three monetisation mechanisms, and all three are being deployed simultaneously: subscriptions, advertising integration, and API and enterprise licensing.

Subscriptions are the cleanest model structurally. They generate predictable revenue and create incentive alignment between usage caps and user behaviour. Google has quietly introduced tighter consumption limits for Gemini subscribers while maintaining higher thresholds than its free tier, a move that serves both cost management and conversion objectives. OpenAI's premium consumer tiers follow a similar logic. The limitation is that subscription penetration among casual AI users remains low, and the price sensitivity of the mass-Market Segment is not yet well understood.

Advertising integration is Google's most natural lever given its core business. AI-generated search responses now carry sponsored placements, and the company has signalled plans to embed product explainers alongside search advertising. The argument that more detailed AI queries carry richer commercial intent is structurally sound. The execution risk lies in user tolerance; if advertising density degrades the perceived quality of AI responses, it may accelerate migration to ad-free alternatives.

Enterprise API licensing is the highest-margin segment, but it is also the most cyclical and competitively contested. As model capabilities converge across providers, pricing power may compress.

The Agent Inflection Point

The launch of persistent, background-operating agents such as Google's Gemini Spark introduces a new variable into the cost equation. Unlike a query-response interaction, an agent that continues executing tasks after a user closes their device generates token consumption that is effectively decoupled from active user engagement. At scale, this is a materially different cost profile from the one operators have been managing to date.

Whether this accelerates the industry's monetisation problem or ultimately resolves it by demonstrating clear productivity value, and therefore pricing power, is the defining question for the next phase of the AI investment cycle.