AI capital expenditure expectations near a peak as thematic investing shifts toward relative value

Why has the previous approach to thematic investing become more difficult?

Citrini Research says that when it began researching thematic investing in 2023, high-quality companies were valued cheaply, and the market was still underestimating the scale of the coming AI investment. As the market priced in higher growth expectations, opportunities gradually shifted toward speculative, not-yet-profitable companies whose cases depended on potential market size.
The firm therefore rebuilt the Citrindex model portfolio from the bottom up: it retained most of its core themes but changed how it expressed those investments, focusing more on reversals, pairs trades and new ideas.

Citrini estimates that annual recurring revenue across the AI ecosystem is at least 200 billion dollars; meanwhile, hyperscalers’ annualized capital expenditure is approaching 1 trillion dollars, and the latter figure does not include all AI spending.
The firm believes that narrowing the gap between revenue and capital expenditure will require a substantial increase in revenue, a slowdown in capital expenditure, or both. In its view, doubling capital expenditure from this level would require trillion-dollar-scale revenue, so the pace of capital expenditure growth is already nearing a peak.
Price competition is also affecting the quality of revenue growth: the report says OpenAI recaptured market share after sharply cutting token prices on 30 July, showing that customers will switch services in response to price. However, Citrini also believes token prices are stabilizing for now and that the two leading labs may still have long-term value.

Citrini believes hyperscalers can no longer finance projects solely from their own cash flows, making debt and equity issuance more important; it also observes that demand for these issuances is beginning to weaken.
Power, interconnection, labor and equipment supply are slowing data-center construction. If delays and cost increases persist, project completion dates could become misaligned with debt maturities, which could in turn slow spending.
The report gives the example of a data-center project backed by Anthropic that stalled because of a Texas data-center moratorium, leaving 15 billion dollars in bridge loans held by banks tied up. Citrini expects stricter local requirements and slower permitting to push up construction costs as well.

Citrini observes that overall job openings have increased, employment in cyclical manufacturing has rebounded, and survey readings in services and manufacturing are at multi-year highs; it also lists tax cuts, tariff relief and the reshoring of industry as other economic drivers.
The firm also believes that a genuine AI bubble bursting could hurt the economy. However, projects already under way are still years from completion, and labor bottlenecks may persist, so there could be a long lag between financial markets repricing and construction activity stopping.
Asset prices have already shown more cautious signals: of the 14 asset groups Citrini tracks, 9 show a stagflationary tilt; cyclical stocks have fallen relative to defensive stocks over the past one, three and six months by 2.4%, 2.9% and 6.9%, respectively. The defensive rotation has continued for several months, and valuations for some consumer staples stocks are already high.