Independent research indicates that while AI technology remains fundamentally sound, corporate Capex is approaching cash flow constraints, revealing underlying financial speculation.
HTX Research, the research arm of cryptocurrency exchange HTX, has released a new report examining the current market cycle position of U.S. artificial intelligence stocks. Published on August 23, 2026, under the title “The Industrialization of Intelligence and Bubble Cycles: Token Economics, Capex, and the Repricing of Risk-Reward in U.S. AI Stocks,” the report argues that the AI industry and its equities are fundamentally out of sync. While technology diffusion remains in its early stages, capital expenditure, valuation levels, and investor sentiment have surged far ahead. This misalignment creates a dynamic where, as the report states, “the industry is not in a bubble, but the financial architecture already shows speculation.”
The market’s pricing logic for AI equities is undergoing a structural transformation. Initially, investors priced in the scarcity of GPUs, high-bandwidth memory, servers, and data center capacity. Subsequently, valuations reflected the performance gains delivered by frontier models and coding agents. Entering 2026, however, the primary drivers of stock returns are shifting away from model parameter counts and capex scale toward token production costs, task completion reliability, usage intensity, enterprise workflow penetration, and the ability of massive AI investments to generate sustainable free cash flow.
This shift is underpinned by a dramatic expansion in corporate spending. J.P. Morgan Asset Management estimates that five U.S. hyperscalers will spend approximately $697 billion (roughly NT$22.2 trillion) on capital expenditures in 2026. Consequently, capex as a share of operating cash flow is projected to climb from approximately 33% in 2023 to 93%. The report stresses that once capital expenditure consumes the vast majority of operating cash flow, market attention inevitably pivots from top-line revenue growth to capital returns.
HTX Research draws a clear distinction between underlying industry fundamentals and financial market behavior. Cloud revenue, coding agent adoption rates, semiconductor sales, and enterprise demand are all expanding in real terms, confirming that AI technology itself is not a false narrative. Conversely, capital expenditure, external financing, data center development, private model valuations, and several high-multiple second-tier equities increasingly exhibit speculative characteristics.
The report further contends that surface-level price-to-earnings ratios fail to capture true valuation levels. For instance, Alphabet’s (GOOGL) P/E ratio is distorted by investment income, while Amazon’s (AMZN) current accounting profits do not reflect a normalized valuation. Sustainable value, the authors argue, belongs to companies that achieve the tightest alignment among normalized valuation, competitive moats, cash flow generation, and AI optionality. Based on current pricing and cycle positioning, Alphabet is identified as offering the most attractive overall risk-reward asymmetry. The same analytical framework was applied to Microsoft (MSFT), Meta (META), TSMC (TSM), NVIDIA (NVDA), Amazon, Oracle (ORCL), Micron (MU), AMD (AMD), Arista (ANET), and Vertiv (VRT) to differentiate companies with high fundamental win rates from those whose valuations already demand near-perfect execution.
Beyond traditional equity analysis, the report examines how AI themes are reshaping asset allocation behavior among cryptocurrency investors. As companies such as NVIDIA, Micron, TSMC, Broadcom (AVGO), Meta, and Alphabet enter the everyday portfolios of crypto users—alongside gold, crude oil, ETFs, and pre-IPO assets—a growing number of participants are treating crypto assets and U.S. equities as distinct allocation vectors within a unified global risk-asset system.
HTX describes itself as one of the earliest cryptocurrency exchanges to systematically integrate this approach. According to data disclosed in August 2026, the platform’s TradFi perpetual contracts segment has surpassed $2.5 billion (approximately NT$80 billion) in cumulative trading volume. It currently supports more than 170 TradFi-related assets spanning U.S. stocks, ETFs, gold, silver, crude oil, AI semiconductors, memory chips, aerospace, and pre-IPO themes including OpenAI and Anthropic. The operational model leverages the platform’s existing user base: holders of stablecoins such as USDT can directly trade TradFi assets within the same account, eliminating the need for separate brokerage accounts or fund transfers. When risk appetite declines, users can rotate into gold, ETFs, or large-cap technology stocks; when risk appetite recovers, they can increase exposure to crypto assets and high-beta AI equities.
The report concludes that the competitive frontier among trading platforms is fundamentally shifting. Competition is moving beyond spot markets, derivatives, liquidity provision, and listing speed toward a broader contest encompassing multi-asset access, wealth management infrastructure, and AI-driven investment tools. For platforms with durable competitive advantages, the core capability will evolve from pure trade execution to comprehensive global asset allocation.