Frontier AI companies looking to raise public funds have left lots of investors puzzled. While private funders assign them eye-popping valuations, scepticism has risen over the vast sums of capital going into the American AI industry, especially amid a web of circular deals.
Anthropic’s valuation, for instance, is projected by some big investors at around $2 trillion by 2028, based on its forecast of $190-200 billion in revenue. That would be more than double its current valuation of about $965 billion, with its annualized revenue at $47 billion in May.
Its rival OpenAI is privately valued at $852 billion, with its annual topline put at just over $25 billion, as extrapolated from its monthly data. Can the equity of these companies be worth so much?
It can only be justified by exponential growth. Which is what the industry is betting on with mega investments in digital infrastructure. Together, US-based Alphabet, Amazon, Microsoft, Meta Platforms and Oracle are expected to spend about $720 billion on capital projects in 2026, as per LSEG estimates.
AI services are costly to offer but expected to be highly profitable too. This paradox may reflect an investor perception that the industry addresses not one market, but two.
The first is a mature one for Predictive AI, where machine learning has already automated fraud detection, credit scoring, phone feeds, advertising, logistics and industrial forecasts, with gains visible across businesses and handsets. The second is a frontier market for Generative and Agentic AI created by tech advances.
Players are putting billions into models, chips, data centres, networks, power generators and talent in anticipation of lower unit costs and potentially boundless usage (one agent can hire another).
Yet, these systems still hallucinate and make errors, while agents go rogue too often. Grumbles have arisen that productivity wins do not justify AI token bills. Some enterprise users say it turns out costlier to let bots take over human roles, especially if the unseen cost of layoffs is taken into account. The web of capital, meanwhile, has grown complex.
While AI majors invest in a huge hardware build-up, chipmakers like Nvidia and AMD are recycling some of their AI windfalls into this ecosystem through equity and other deals that link chip supplies with future compute capacity.
The Bank for International Settlements has flagged risks of AI investment being financed via arrangements that resemble debt but stay off balance sheets. Nvidia’s plan with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR for AI-funding platforms would suggest a large appetite for a piece of the action.
The trouble with closely interlinked deals is that a single node of failure might trigger a big blowout.
That said, if inference and token costs drop sharply and AI models achieve efficiency and stability, it may be easier to count on future returns that justify sky-high valuations. Yet, so long as variable costs exist and differentiation finds customers, this is unlikely to end up as a winner-takes-all market; in other words, rivalry will leave no space for monopoly profits.
Predictive AI has proven itself, but AI bulls are yet to convince sceptics of a pot of gains at the end of the generative and agentic spectrum. While blue-sky gazing has a stellar record in the world of digital technology, with virtually no limit on what can be dreamt into being, hardnosed investors prefer to go by the facts in evidence.
What works well for innovation needn’t suit financial calls, even if AI bulls foresee big wins.