The financial world is abuzz with the prospect of monumental initial public offerings from frontier AI labs, with Anthropic having confidentially filed and OpenAI reportedly preparing its own draft. Valuations are reaching dizzying heights, with Anthropic eyeing a nearly trillion-dollar market cap and OpenAI not far behind. Adding to this speculative fervor, SpaceX is pursuing an even more ambitious listing, aiming for $1.75 trillion. This concentration of capital formation harks back to the dot-com boom, suggesting a period of intense investor interest. However, a closer look at the revenue streams these AI darlings are projecting reveals a potential disconnect between market expectations and current realities.
These leading AI laboratories primarily target the upper echelon of the global AI market: large enterprises equipped with robust networks, significant talent pools, and substantial computing budgets. It is within this segment that their frontier models and copilots demonstrate their most impressive capabilities. Yet, even in this seemingly ideal environment, concerns are emerging. Sam Altman, CEO of OpenAI, recently acknowledged the “fair criticism” regarding the excessive costs associated with corporate AI adoption. Businesses are struggling to demonstrate a clear return on investment, especially when more cost-effective open-source alternatives often perform comparably. The revenue projections these AI companies are banking on do not yet align with the tangible returns corporate buyers are experiencing.
The history of business suggests that the most substantial and sustainable profits often arise from addressing unmet needs, particularly in less glamorous sectors that typically escape the attention of frontier labs and mainstream investors. The Digital Planet team’s 2026 Digital Evolution Index, which assesses 125 economies across 185 indicators, identifies significant unmet demand in both developed and developing regions where AI could genuinely scale. For instance, highly digitally evolved economies like the U.S. and Europe grapple with outdated infrastructure. A striking 43% of core banking systems and 95% of ATM transactions still rely on COBOL, a programming language older than The Beatles. The potential for AI to modernize these systems is immense, as evidenced by the market reaction when Anthropic suggested Claude could automate such modernization, causing IBM’s stock to fall sharply.
Beyond the developed world, 51 “Break Out” economies, including India, Brazil, Indonesia, Kenya, and Vietnam, are experiencing rapid digital acceleration. These nations present an even more compelling opportunity. Millions of users in these regions actively use mobile wallets and possess rich transaction histories but lack access to formal credit. AI-powered credit scoring, leveraging payment data, identity authentication, and fraud detection, could unlock enormous value. These payment systems are already operating at massive scales; India’s UPI, for example, processed 22.6 billion transactions in March 2026 alone, and mobile money moved over $2 trillion globally in 2025. This represents a vast, already monetizing market largely overlooked in the current AI revenue narrative.
Further afield, in “Watch Out” economies, primarily in Sub-Saharan Africa and South Asia, even more fundamental needs exist. Research indicates that a single application—AI crop-disease detection across just seven African countries—could generate $6.1 billion for 14 million smallholder farmers. Intriguingly, populations in these regions report the highest trust in AI, surpassing even Silicon Valley executives whose enthusiasm often drives market valuations. These applications, while not dominating daily headlines, represent tangible, high-impact opportunities for AI.
Looking back at previous technological shifts offers a valuable perspective. During the dot-com era, while companies like Pets.com and Webvan garnered significant attention, the enduring winners were those that provided foundational infrastructure: Cisco with its routers, Akamai delivering content, and eventually Amazon Web Services. The mobile revolution followed a similar pattern, with long-term value accruing to infrastructure providers like American Tower and Crown Castle, rather than solely to handset manufacturers. The more transformative a technology, the more value tends to migrate to the underlying infrastructure that every participant must utilize.
Strategic acquirers appear to recognize this pattern. Despite a depressed deal market in 2025, data infrastructure remained a hot sector. IBM acquired DataStax, ServiceNow purchased Data.world, and Salesforce spent $8 billion on Informatica. These acquisitions suggest a focus not on which AI model will ultimately dominate, but on the essential data pipelines and infrastructure that every AI-driven enterprise will perpetually require. The economics of the AI buildout are also sobering, with Bain & Company warning of an $800 billion shortfall needed to justify compute spending by 2030. Oracle’s disclosure of $248 billion in data-center leases over 15 to 19 years, against shorter customer contracts, highlights the significant fixed costs involved. Furthermore, open-weight models are driving down inference prices by 30% to 50% annually, compressing potential margins for model layer providers.
None of this inherently dooms the upcoming mega-IPOs. OpenAI might yet meet its revenue targets, Anthropic could secure enough enterprise clients, and SpaceX’s launch economics may indeed justify its valuation. However, the current race to market often involves selling a vision of AI seamlessly integrating into a global economy of augmented knowledge workers, a reality that the current return on investment data does not yet fully support. Investors who achieved generational wealth in past cycles did not chase the most exciting narratives at the IPO moment. Instead, they asked a fundamental question: where is the undeniable need, and what essential component will every participant in this new economy invariably pay for, time and again? That question, today, points toward COBOL modernization in Stuttgart, fraud detection in São Paulo, and crop-disease models in Addis Ababa, offering a more grounded, albeit less glamorous, investment thesis.
