The artificial intelligence (AI) industry has reached a crossroads. For years, companies have poured money into infrastructure such as data centers, chips and underlying models. Now, the big question is not whether AI works, but whether AI can generate profits. The solution lies in concrete AI applications, and goes beyond just increasing the amount of training or larger GPU clusters.
AI infrastructure and AI applications
In 2025, businesses spent approximately $320 billion on AI infrastructure. Despite this huge investment, the profit margin of the basic model business is thin. High inference costs reduce revenue, and competition drives down prices. For example, OpenAI reached $13 billion in annual revenue by August 2025, but is still losing $5 billion in 2024. This approach is not sustainable. It is a short-term solution supported by venture capital and corporate funding in the hope of better returns.
For AI applications, it’s a different story. In 2025, businesses spent $19 billion on AI applications, accounting for more than half of all generated AI spending. This represents more than 6% of the total software market and was reached in just three years after ChatGPT’s launch. More importantly, this spending demonstrates real market demand. Companies are no longer just testing AI; They use it extensively. Currently, at least 10 AI products generate more than $1 billion in annual recurring revenue, and 50 products generate more than $100 million.
Meta’s $2 billion acquisition of Manas in December 2025 is indicative of this change. Singaporean startup Manus launched its AI agent just nine months ago and quickly reached $125 million in annual revenue, proving both its technical prowess and its business success by offering a simple and effective AI product that not only talks about tasks, but also performs them.
Investors are looking for companies with real customers, not just technology. Through the third quarter of 2025, there were 265 private equity deals related to AI applications, an increase of 65% year over year, and 78% were add-on acquisitions of existing portfolio companies. Strategic mergers and acquisitions in the AI space hit an all-time high through Q3, with deal value increasing 242% year-over-year.
where is the real value?
The real value is emerging in the sectoral AI segment. In 2025, coding tools accounted for $4 billion of the $7.3 billion sectoral AI market, making it the largest segment. Half of all developers now use AI coding tools every day, and this number rises to 65% at top-performing companies. When ServiceNow acquired Moveworks or Nvidia acquired some AI startups, these weren’t infrastructure deals. These were investments in companies that help customers achieve real business outcomes with AI.
The underlying model landscape itself tells the story of the application. Anthropic now accounts for 40% of enterprise LLM spending, up from 24% last year and 12% in 2023, while OpenAI’s enterprise share fell from 50% to 27% in 2023. Anthropic has achieved this by dominating coding applications, holding 54% market share compared to OpenAI’s 21%. Applications drive infrastructure deployment and underlying models, not the other way around. According to a report by Morgan Stanley, the contribution of generative AI could reach 34% in 2025, its first revenue year, and rise to 67% by 2028 as infrastructure costs fall and efficiency increases. However, most of these profits go to companies that sell complete solutions, not just raw computing power.
Retail investors now have to decide which use cases will create the next wave of real value, rather than just adding a simple interface with no real value to ChatGPT. But solutions built for specific verticals like healthcare, legal, finance, and manufacturing—deeply integrated into workflows, using proprietary data, and integral to operations—are businesses worth serious investment in.
When investors look beyond the narrative and focus on fundamentals, the free market allocates resources better. Revenue, customer retention, growth rates, and the path to profitability are once again important. Currently, circular finance is obscuring true demand. For example, a significant portion of Microsoft’s reported Azure AI revenue comes from spending on OpenAI compute at deeply discounted rates that essentially cover only Microsoft’s costs. The application breaks this pattern as it generates revenue from outside the circular lending loop.
core issue
For governments, the next stage will pose tough competition questions, especially as underlying model providers begin to build their own applications. When OpenAI launches coding tools or Anthropic develops enterprise solutions, it puts pressure on independent application builders who don’t have the same infrastructure benefits. Copyright issues are also becoming increasingly important, as the source of training data becomes an important legal issue. Privacy rules will need to adapt to deal with AI agents accessing large amounts of personal and business information.
Policymakers should not rush to impose strict regulations. The application layer needs room to experiment, fail, and improve until you find product-market fit. But it’s still important to review competition rules, especially takeover rules that prevent large companies from buying or shutting down potential rivals. The trend toward acquisitions (primarily buying startups for their employees and then shutting them down) often leaves employees stranded and can hurt the energy and innovation the industry needs.
The Internet didn’t make money by selling bandwidth. Monetized by building applications where bandwidth is at a premium. AI will follow the same trajectory.
Arindam Goswami is a research analyst in the High Tech Geopolitics Program at Takshashila Institute, Bangalore.
issued – February 4, 2026 12:08am IST
