HTX research examines U.S. AI stocks: Technology is still in the early stages, but capital expenditu

Intelligent industrialization and bubble cycles: Tokenology, capital expenditures, and risk-return repricing of U.S. artificial intelligence stocks

HTX Research, the research arm of HTX, has released a new report. The core argument is that the artificial intelligence industry and artificial intelligence stocks are not at the same stage of their respective cycles-technology diffusion is still early, and capital expenditures, valuations and investor sentiment are far ahead.

Variables driving stock prices

The market initially priced the scarcity of GPUs, high-bandwidth memory, servers and data center capacity, followed by the capabilities improvements brought by cutting-edge models and coding agents. By 2026, the variables driving stock returns are shifting from the number of model parameters and the size of capital expenditures to token production costs, task completion reliability, usage intensity, corporate workflow penetration, and the ability of huge AI investments to generate sustained free cash flow.

Behind this shift are changes in the scale of capital expenditures. JPMorgan Asset Management estimates that five U.S. hyperscale cloud service providers will spend approximately US$697 billion in 2026, and capital expenditures will account for an estimated 93% of their operating cash flow from approximately 33% in 2023. Once capital expenditures consume the vast majority of operating cash flow, market attention must shift from revenue growth to return on capital.

Bubbles exist in the financial architecture, not in the industry itself

Cloud revenue, coded agent adoption, semiconductor sales and corporate demand are all actually growing, which means that AI technology itself is not a false narrative. However, capital expenditures, external financing, data center projects, private model valuations and numerous high-multiple second-tier stocks are taking on increasingly obvious speculative characteristics.

The S & P P/E ratio also does not reflect the true valuation level: Alphabet's P/E ratio is distorted by investment income, and Amazon's current accounting profit does not represent normalized valuation. What is really valuable is the closest fit between normalized valuations, moats, cash flow and AI options.

At current price and cycle positions, the report believes that Alphabet provides the most attractive overall asymmetry, and applies the same framework to Microsoft, Meta, TSMC, Nvidia, Amazon, Oracle, Micron, AMD, Arista and Vertiv-distinguishing those companies with high fundamentals and those whose valuations already require near-perfect execution.

AI is reshaping the way cryptocurrency investors allocate capital

As a common theme in global capital markets, AI's influence has gone beyond U.S. stock pricing to the allocation behavior of cryptocurrency investors. As stocks such as Nvidia, Micron, TSMC, Broadcom, Meta, and Alphabet have entered the daily portfolio of cryptocurrency users along with gold, crude oil, ETFs and Pre-IPO assets, more and more users now regard cryptocurrency and U.S. stocks as different allocation directions within the same global risk asset system.

HTX is one of the first cryptocurrency exchanges to systematically develop in this direction. According to data disclosed in August 2026, the cumulative trading volume of the platform's TradFi perpetual contract segment has exceeded US$2.5 billion, supporting more than 170 TradFi-related assets, covering U.S. stocks, ETFs, gold, silver, crude oil, AI semiconductors, memory, aerospace, and Pre-IPO themes such as OpenAI and Anthropic.

The key to the establishment of this model lies in the platform's existing cryptocurrency user base-they have completed registration, verification and recharge. Users holding stablecoins such as USDT can trade TradFi assets in the same account without opening a brokerage account or transferring funds to another financial system-allocate gold, ETFs or large-cap technology stocks when risk appetite declines, and increase exposure to cryptocurrencies and high beta AI when risk appetite rebounds.

The competitive boundaries of trading platforms are shifting

Competition between trading platforms will transcend the spot market, derivatives, liquidity and currency placement speed to a broader competition-multi-asset access, wealth management and AI investment tools. For a platform with lasting competitiveness, its core capabilities will evolve from simple execution to global asset allocation.

This confirms a larger judgment in the report: AI is changing not only model capabilities and computing power needs, but also capital flows, allocation behaviors, and the way financial products are organized. HTX's early deployment in the TradFi field echoes HTX Research's continuous tracking of AI themes and cross-market capital flows-identifying cycle locations, interpreting capital flows, and understanding linkages between assets are both the core of research and business decisions. The source of first-mover advantage in decision-making. As AI pushes global markets into a new stage of integration, institutions that can understand both industrial cycles and capital flows will be in a better position in the next round of competition.