The artificial intelligence industry has crossed a critical threshold. The debate is no longer about whether AI works, but whether it can make money at scale. After years of massive investment in data centres, chips and foundation models, the centre of gravity is shifting decisively toward applications — tools that solve real problems, deliver measurable value, and generate sustainable revenue.
From infrastructure obsession to a profitability question
Over the last few years, AI growth has been driven by infrastructure. In 2025 alone, companies globally spent about $320 billion on data centres, GPUs and large models. Yet this spending has not translated into strong margins for foundation model providers.
High inference costs, relentless competition and falling prices have squeezed profits. Even leading players like OpenAI reached impressive revenue milestones while remaining deeply loss-making. This model — sustained by venture capital and strategic corporate funding — is proving fragile and difficult to sustain over the long term.
Why AI applications tell a different story
The application layer paints a far more encouraging picture. In 2025, businesses spent roughly $19 billion on AI applications, accounting for more than half of all generative AI spending. This is a remarkable shift just three years after ChatGPT’s launch.
Crucially, this spending reflects real demand rather than experimentation. Companies are no longer “piloting” AI — they are embedding it into everyday workflows. At least ten AI products now earn over $1 billion in annual recurring revenue, while dozens more have crossed the $100 million mark. This is where AI is beginning to look like a normal, profitable software market.
What investors are signalling through deals
Investor behaviour confirms this transition. Private equity and strategic buyers are increasingly targeting AI application companies rather than pure infrastructure plays. By the third quarter of 2025, deals involving AI applications surged, with most acquisitions aimed at strengthening existing product portfolios.
High-profile examples underline the point. When Meta acquired Manus, or when Nvidia bought smaller AI startups, the motivation was not access to more compute. It was about acquiring products that already deliver tangible business outcomes for customers.
Departmental AI: where real value is emerging
The strongest traction is visible in “departmental AI” — tools designed for specific functions such as software development, customer support or operations. In 2025, AI coding tools alone accounted for more than half of this segment’s revenue.
Adoption is deep rather than superficial. Around half of all developers now use AI coding assistants daily, rising to nearly two-thirds in top-performing firms. These tools save time, reduce errors and directly improve productivity, making them easy to justify in corporate budgets.
Applications reshaping the foundation model race
Even competition among large language models now reflects the primacy of applications. has rapidly increased its share of enterprise AI spending by focusing on coding and enterprise use cases, while OpenAI’s relative dominance has declined.
This shift illustrates a broader truth: applications drive demand for models and infrastructure, not the other way around. Models that excel in specific, high-value use cases gain traction, while general-purpose offerings face commoditisation.
When AI finally turns profitable
According to , generative AI reached its first profitable year in 2025, with contribution margins improving sharply as infrastructure costs fell and efficiency rose. However, most of these gains are accruing to firms selling end-to-end solutions, not raw compute or model access.
This distinction matters for investors. Simply adding a thin interface on top of a large model is unlikely to create durable value. Profitable AI businesses are those embedded deeply into workflows, built on unique data, and hard to replace — especially in verticals like healthcare, law, finance and manufacturing.
Circular financing and why applications break it
One concern in today’s AI economy is circular financing. For example, some reported AI revenues effectively flow from one tech giant to another, with discounts masking thin or nonexistent margins. This obscures real demand.
Applications break this loop. They earn revenue from customers who pay because the product delivers value, not because it advances a strategic narrative. As fundamentals such as customer retention, growth rates and profitability regain importance, applications stand out as the strongest signals of genuine market demand.
Policy challenges in the next AI phase
For governments, the rise of applications raises new regulatory questions. Competition policy becomes critical as foundation model providers begin building their own applications, potentially squeezing independent developers. Issues of copyright and data provenance are intensifying, while privacy rules must adapt to AI agents that handle sensitive information.
At the same time, premature over-regulation could stifle innovation. The application layer still needs space to experiment, fail and refine products. A balanced approach — encouraging competition, scrutinising acquisitions, and discouraging predatory acqui-hires — will be essential.
What to note for Prelims?
- Difference between AI infrastructure and AI applications.
- Rising share of AI spending going toward applications.
- Concept of departmental AI tools.
What to note for Mains?
- Why AI profitability depends more on applications than models.
- Implications of market concentration and competition in AI.
- Regulatory challenges at the application layer.