Capex at 93% of Cash Flow: HTX Says AI Tech Is 'Early', Markets 'Late'

For AI builders, enterprise adopters, and model developers, the report's central distinction cuts through the noise: the technology diffusion curve is still in its early stages even as the market's pricing cycle has run far ahead. HTX Research says 2026 returns will be driven not by model parameter counts or the scale of capex, but by token production costs, task-completion reliability, usage intensity, and how deeply AI penetrates enterprise workflows — a shift that rewards inference-time efficiency and operational cost discipline over raw training-scale ambition.

HTX Research, the research arm of the HTX crypto and Web3 platform, published a report on August 23, 2026 arguing a core distinction that will likely define the next leg of the AI investment cycle: the technology trajectory of artificial intelligence remains in its early stages, while capital expenditure, equity valuations, and investor sentiment have already entered the late cycle. The report, titled "The Industrialization of Intelligence and the Bubble Cycle: Token Economics, Capital Expenditure, and the Repricing of Risk-Reward Across U.S. AI Equities," was distributed via PR Newswire and frames a market that has repriced AI twice already — first on the scarcity of GPUs, high-bandwidth memory, servers, and data-center capacity, and later on the capability gains delivered by frontier models and coding agents. By 2026, according to HTX Research, the variables driving equity returns are shifting again, away from model parameter counts and raw capex scale toward token production costs, task-completion reliability, usage intensity, enterprise-workflow penetration, and above all the ability of enormous AI investments to generate durable free cash flow.

“

More revealing than the absolute figure is the trend: capex has risen from roughly 33% of these companies' operating cash flow in 2023 to an estimated 93% in 2026.

The most striking data point in the report is the sheer magnitude of hyperscaler spending. HTX Research cites J.P. Morgan Asset Management estimates that five U.S. hyperscalers will spend approximately $697 billion in capital expenditure in 2026. More revealing than the absolute figure is the trend: capex has risen from roughly 33% of these companies' operating cash flow in 2023 to an estimated 93% in 2026. When capital expenditure consumes the overwhelming majority of operating cash flow, the report argues, market attention necessarily migrates from revenue growth to return on invested capital. This is the late-cycle signal, and it matters because it unmoors the market's pricing logic from headline growth rates and forces a repricing of risk-reward across the entire AI equity complex.

The report is careful to reject the simplistic conclusion that AI is a false narrative. It explicitly notes that cloud revenue, coding-agent adoption, semiconductor sales, and enterprise demand are all growing in real terms. The technology, in other words, is real and diffusion is early. What displays increasingly speculative characteristics, according to HTX Research, are the financial layers wrapped around that technology: capital expenditure itself, external financing arrangements, data-center project pipelines, private-model valuations, and a cohort of high-multiple second-tier equities. Headline price-to-earnings ratios, the report warns, fail to capture true valuation levels — Alphabet's multiple, for instance, is distorted by investment income, and the report begins to make a similar point about Amazon before the source text is truncated.