HTX Research, the research arm of cryptocurrency exchange HTX, has released a new report examining the current cycle positioning of U.S. artificial intelligence stocks. The report’s core thesis is that the AI industry and AI stocks are in desynchronized cycles—technology diffusion remains in its early stages, but capital expenditure, valuation levels, and investor sentiment have surged far ahead, creating a mismatch where “the industry is not in a bubble, but the financial architecture already shows speculation.”
The report, titled “The Industrialization of Intelligence and Bubble Cycles: Token Economics, Capex, and the Repricing of Risk-Reward in U.S. AI Stocks,” was published on August 23, 2026. It argues that the market’s pricing logic for AI stocks is undergoing a fundamental transformation.
Pricing Variables Shift from Scarcity to Cash Flow
The report notes that the market initially priced in the scarcity of GPUs, high-bandwidth memory, servers, and data center capacity, and subsequently reflected the capability gains delivered by frontier models and coding agents. However, entering 2026, the variables driving 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 whether massive AI investments can generate sustainable free cash flow.
Pricing LogicOld Signals (Early Industry Expansion)New Signals (From 2026)HardwareScarcity of GPUs, HBM, servers, and data center capacityToken production costsCapabilityCapability gains from frontier models and coding agentsTask completion reliability, enterprise workflow penetrationFinancialModel parameter scale, capex scale itselfWhether massive investment can generate sustainable free cash flow
Behind this shift is the dramatic expansion of capital expenditure. J.P. Morgan Asset Management estimates that five U.S. hyperscale cloud providers will spend approximately $697 billion in capital expenditure in 2026 (approximately NT$22.2 trillion), with capex as a share of operating cash flow climbing from roughly 33% in 2023 to a projected 93%. The report emphasizes that once capex consumes the vast majority of operating cash flow, market attention inevitably shifts from revenue growth to capital returns.
▲ Capex as a share of operating cash flow for five U.S. hyperscale cloud providers is projected to climb from approximately 33% in 2023 to approximately 93% in 2026 (data cited above).
Note: According to data compiled by capex tracking firm Value Add VC from the earnings reports and official guidance of Microsoft, Google, Meta, and Amazon (as of Q2 2026), the four companies’ combined capex will grow from approximately $410 billion in 2025 to a projected $725 billion in 2026, a year-over-year increase of roughly 77%—consistent with the report’s thesis of sharply escalating capital expenditure.
The report draws a clear distinction between industry fundamentals and financial markets. Cloud revenue, coding agent adoption, semiconductor sales, and enterprise demand are all growing in real terms, indicating that AI technology itself is not a false narrative. However, capital expenditure, external financing, data center projects, private model valuations, and several high-multiple second-tier stocks increasingly exhibit speculative characteristics.
The report further argues that surface-level P/E ratios fail to reflect true valuation levels. Using Alphabet (GOOGL) as an example, its P/E ratio is distorted by investment income; Amazon’s (AMZN) current accounting profits also fail to reflect normalized valuation. What truly holds value are companies that achieve the tightest alignment among normalized valuation, competitive moats, cash flow, and AI optionality.
Based on current prices and cycle positioning, the report identifies Alphabet as offering the most attractive overall risk-reward asymmetry. The research team applied the same framework to Microsoft (MSFT), Meta (META), TSMC (TSM), NVIDIA (NVDA), Amazon, Oracle (ORCL), Micron (MU), AMD (AMD), Arista (ANET), and Vertiv (VRT), with the goal of distinguishing companies with high fundamental win rates from those whose valuations already demand near-perfect execution.
The eleven tracked stocks covered by the report’s framework are summarized below:
CompanyTickerIndustry PositioningAlphabetGOOGLCloud/Advertising; report’s top pickMicrosoftMSFTCloud computing/Enterprise softwareMetaMETASocial platforms/AI infrastructureTSMCTSMSemiconductor foundryNVIDIANVDAAI accelerator chipsAmazonAMZNCloud computing/E-commerceOracleORCLCloud databaseMicronMUMemory chipsAMDAMDData center processorsAristaANETData center networking equipmentVertivVRTData center cooling and power infrastructure
Shifts in Crypto Investor Allocation Behavior
The report also explores how AI themes are influencing crypto investors’ asset allocation behavior. 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—an increasing number of users are viewing crypto assets and U.S. equities as different allocation directions within a single global risk asset system.
HTX describes itself as one of the earliest cryptocurrency exchanges to systematically advance in this direction. According to data disclosed in August 2026, the platform’s TradFi perpetual contracts segment has surpassed $2.5 billion in cumulative trading volume (approximately NT$80 billion), supporting more than 170 TradFi-related assets spanning U.S. equities, ETFs, gold, silver, crude oil, AI semiconductors, memory, aerospace, and pre-IPO themes such as OpenAI and Anthropic.
The model’s operational foundation rests on the platform’s existing crypto user base. Users holding stablecoins such as USDT can directly trade TradFi assets within the same account, without needing to open a brokerage account or transfer funds into a separate financial system. When risk appetite declines, users can allocate to gold, ETFs, or large-cap tech stocks; when risk appetite recovers, they can increase exposure to crypto assets and high-beta AI assets.
The report concludes that the competitive frontier among trading platforms is shifting—from spot markets, derivatives, liquidity, and listing speed toward a broader competition encompassing multi-asset access, wealth management, and AI investment tools. For platforms with durable competitive advantages, the core capability will evolve from pure trade execution to global asset allocation capability.