AI's Next Investment Cycle: The Focus on Real Applications

The shift from AI infrastructure to AI applications signifies a maturation of the artificial intelligence ecosystem. In the early phase, enormous investments flowed into foundational layers—data centres, GPUs, cloud infrastructure and large language models—under the assumption that scale itself would generate value. While this phase proved that AI works technically, it also exposed a key weakness: infrastructure-heavy AI businesses struggle to achieve profitability due to high inference costs and intense competition that suppress pricing power.

AI applications, by contrast, sit closer to end-users and business workflows. They translate abstract AI capabilities into tangible productivity gains—such as faster coding, automated customer support or clinical decision assistance. The fact that businesses spent $19 billion on AI applications in 2025, accounting for more than half of generative AI spending, demonstrates that demand has moved beyond experimentation into routine deployment. This mirrors earlier technological revolutions, where value ultimately accrued not to infrastructure providers but to application builders.

Conceptually, this shift highlights an important economic principle relevant for policy and governance: technology adoption becomes sustainable only when it solves real problems. Just as the internet was monetised through applications like e-commerce and search rather than bandwidth alone, AI’s long-term viability depends on use cases that deliver measurable outcomes, not merely computational scale.

Foundation model providers face a structural profitability challenge because their cost base scales almost linearly with usage. Training and running large models requires continuous spending on GPUs, energy and cloud infrastructure. Even when revenues grow rapidly—as seen with OpenAI’s $13 billion annualised revenue—the high cost of inference and ongoing model upgrades erode margins, leading to sustained losses.

Competition further aggravates this problem. As more firms release comparable models, prices for API access are driven down, limiting pricing power. In addition, much of the reported revenue in the AI ecosystem comes from circular financing, where companies buy compute from cloud providers who are also their investors or partners. This obscures true market demand and delays financial discipline.

From a broader economic perspective, this illustrates why infrastructure-led growth alone is unsustainable. Without downstream applications generating independent demand, foundation models risk becoming commoditised utilities. This dynamic reinforces the argument that profitability and long-term innovation are more likely to emerge from application layers rather than from foundational technology alone.

Meta’s $2 billion acquisition of Manus illustrates how value in the AI sector is increasingly being captured by companies that deliver <strong>focused, outcome-oriented applications</strong>. Manus did not build a new foundational model; instead, it developed an AI agent that performed concrete tasks efficiently. Within nine months, it achieved$125 million in annual revenue, proving both product-market fit and scalability.

This acquisition signals a shift in strategic priorities among large technology firms. Rather than investing only in infrastructure or models, they are seeking to acquire applications that already have paying customers and embedded workflows. Similar logic underpinned acquisitions like ServiceNow–Moveworks and Nvidia’s purchases of AI startups, all of which emphasised customer value rather than raw computing power.

For policymakers and aspirants, this example underscores a critical insight: innovation ecosystems thrive when startups solve narrow but impactful problems. The Manus case also offers lessons for countries like India, where building applied AI solutions tailored to local sectors—healthcare, agriculture or governance—may yield greater economic returns than attempting to compete at the foundational model level.

Departmental AI refers to AI tools designed for specific functions such as coding, customer service, legal research or human resources. Its rapid growth—coding tools alone accounted for $4 billion of a$7.3 billion market in 2025—reflects strong demand for targeted solutions that integrate seamlessly into daily workflows. By automating routine tasks, departmental AI significantly boosts productivity and reduces error rates.

However, this shift also raises concerns. On the positive side, such tools augment human capabilities rather than fully replacing workers, enabling professionals to focus on higher-value tasks. On the negative side, uneven access to AI tools could widen productivity gaps between firms and countries, potentially reinforcing existing inequalities in the global economy.

From a policy standpoint, the challenge lies in managing this transition responsibly. Governments must encourage skill upgradation and digital literacy while ensuring that AI adoption does not lead to excessive job displacement. The rise of departmental AI thus represents both an opportunity for inclusive growth and a test of adaptive governance.

Investor confidence has shifted because applications provide clearer visibility into revenue streams, customer retention and profitability. Unlike infrastructure-heavy firms, application companies can demonstrate value through subscription models, recurring revenues and measurable productivity gains for clients. This explains the sharp rise in private equity and merger-and-acquisition activity centred on AI applications in 2025.

Another reason is the declining marginal returns from infrastructure investments. As foundational models become widely available, differentiation increasingly depends on how effectively AI is embedded into business processes. Applications break the cycle of circular financing by generating revenue from external customers rather than from within the AI ecosystem itself.

This shift also reflects a broader market correction where fundamentals—cash flow, growth rates and sustainability—matter more than narratives. In this sense, AI is transitioning from a speculative phase to a more disciplined, market-driven phase of development.

Governments face a delicate balancing act. On one hand, excessive regulation at an early stage could stifle innovation in the application layer, which still requires experimentation and failure to discover viable business models. On the other hand, unchecked consolidation—where large firms acquire or neutralise potential rivals—can harm competition and long-term innovation.

Competition policy therefore becomes central. Regulators must closely scrutinise mergers, especially acqui-hires that eliminate emerging competitors without delivering consumer benefits. At the same time, evolving issues such as copyright in training data and privacy in AI agents require adaptive regulatory frameworks rather than rigid rules.

For India and other developing economies, the priority should be to encourage domestic AI applications aligned with national needs while ensuring fair competition. The broader lesson is that AI governance must evolve alongside markets, ensuring innovation thrives without undermining economic fairness or democratic accountability.