Executive Summary
In June 2026, global risk assets stand at an extremely delicate inflection point. Over the past three years, artificial intelligence has become the most important theme in global capital markets. It has not only driven valuation expansion for hardware and semiconductor assets such as Nvidia, TSMC, Broadcom, Astera Labs, and Micron, but also pushed private or semi-public AI assets such as OpenAI, Anthropic, SpaceX, xAI, Cerebras, and Cursor into a new valuation regime. At the same time, AI infrastructure construction has evolved from a simple software narrative into a supercycle centered on GPUs, HBM, advanced packaging, wafer capacity, electricity, natural gas, data centers, private credit, and capital-market financing.
This cycle is not exactly the same as the 2000 internet bubble. AI is not a pure concept bubble with no product, no demand, and no revenue. Large models have genuinely improved efficiency in coding, content creation, customer service, data analysis, search, and enterprise workflow automation; cloud providers, model companies, and hardware supply chains also generate real revenue. The problem is that a genuine technological revolution does not mean asset prices are always reasonable. Historically, the most dangerous bubbles are often not built on fake technologies, but on real technologies after they have been over-financialized, over-capitalized, and priced with excessive optimism.
The core risk in the current AI market has shifted from “Is the technology real?” to “Are capital returns sufficient to support the current level of investment?” After 2024, global AI data-center capital expenditure accelerated. From 2025 to 2026, hyperscale cloud providers, model companies, data-center developers, private-credit funds, and bond markets jointly pushed the scale of AI infrastructure financing sharply higher. AI is turning from a growth story in the equity market into balance-sheet engineering jointly supported by the bond market, the private-credit market, and the energy market.
Meanwhile, the crypto market has not been the biggest beneficiary of this round of global liquidity expansion. Under the traditional framework, M2 expansion, fiscal deficits, expectations of monetary easing, and rising risk appetite should usually support Bitcoin and crypto assets. Yet from late 2024 to mid-2026, a considerable share of newly created dollar liquidity was absorbed by the AI value chain. AI CapEx has become a “liquidity black hole” for global risk capital: equity investors buy AI equities, bond investors buy AI-related credit assets, private funds finance data centers, and banks as well as non-bank institutions lend to large technology companies and data-center projects. By contrast, although Bitcoin remains a high-beta asset in the global liquidity cycle, its marginal capital-attraction power has been suppressed by the AI trade.
The core judgment of this paper is that AI is neither a simple scam nor a super-miracle that can be extrapolated linearly without constraint. It is more like a complex cycle in which a real technological revolution and a financial bubble coexist. The bubble component is not that AI has no value, but that the market has pulled too much future cash flow, too much productivity improvement, too much social acceptance, and too much low-cost energy and low-cost capital into current valuations ahead of time.
For the crypto market, the meaning of the AI bubble is not merely whether U.S. equities will fall. More importantly, it is changing the correlation structure between Bitcoin and global liquidity. Over the past decade, Bitcoin has mainly been correlated with dollar liquidity, real interest rates, risk appetite, and regulatory cycles. In 2026, however, Bitcoin must face a new variable: whether AI continues to absorb global marginal risk capital. If AI keeps expanding, the crypto market may remain for a long time in a state of “liquidity without control over incremental capital”. If AI valuations compress, crypto assets may fall alongside risk assets in the short term, but after credit clears and monetary easing resumes, Bitcoin may also become one of the first assets to rebound in the next liquidity-repair cycle.
Therefore, from the second half of 2026 through 2027, investors need to monitor three lines simultaneously: first, whether AI CapEx shifts from accelerating expansion to marginal deceleration; second, whether AI financing spreads from equity markets into credit markets and creates systemic leverage; third, whether the crypto market regains independent capital inflows rather than continuing to be pulled by the risk appetite of U.S. technology stocks.
HTX Research, the research arm of HTX, will continue to track the interplay between the AI capital cycle and crypto-market liquidity, providing data-driven assessments to help market participants navigate this repricing process.
1.1 The June 2026 Macro Backdrop: Risk-Asset Repricing under Energy Shock, Inflation Expectations, and High Interest Rates
The macro environment in June 2026 is neither a typical easing cycle nor a typical recession cycle. It is a complex environment shaped by energy shocks, geopolitics, sticky inflation, high interest rates, and AI investment impulses.
On the surface, the market has gradually become accustomed to high rates. The Federal Reserve’s policy rate remains in the 3.50%-3.75% range. Although this is below the extremes of the 2022-2023 tightening cycle, it is still materially higher than the zero-rate environment of the 2010s. More importantly, the market is no longer convinced that the Fed will cut rates quickly. Energy prices, food prices, and tariff pass-through have revived upside inflation risks, making it harder for the Fed to send a clear easing signal in mid-2026.
This matters greatly for risk-asset pricing. Over the past decade-plus, the core valuation pillar for global risk assets has been low interest rates and abundant liquidity. Technology stocks, growth stocks, crypto assets, and high-valuation private-market projects all benefited from an environment in which future cash flows could be discounted at low rates. Once discount rates remain high for an extended period, asset valuations must be supported by real cash flows rather than distant narratives alone.
What makes the 2026 macro environment distinctive is that inflation pressure is not coming purely from overheated demand; it is increasingly coming from supply-side constraints. U.S.-Iran tensions, disruptions to the Strait of Hormuz, and higher risks in oil and gas transportation continue to disturb energy prices and inflation expectations. Even if oil prices fall in the short term because of a ceasefire or temporary agreement, the market will find it difficult to immediately believe that energy supply chains have fully recovered. As long as shipping, insurance, settlement, geopolitical security, and capacity scheduling remain uncertain, energy prices will continue to pose upside risks to inflation expectations.
For U.S. politics, energy prices and inflation are not only economic issues, but also electoral issues. Gasoline, food, electricity bills, and rent are the living-cost variables that ordinary voters feel most directly. Even if the stock market rises, if median households cannot benefit enough from higher equity prices and instead face higher electricity bills and living costs, “growth” in macro data will have difficulty translating into political satisfaction.
This is where AI infrastructure expansion collides with macro politics. AI data-center construction requires enormous amounts of electricity, water, land, transmission networks, and gas-fired generation support. Against a backdrop in which energy prices have already risen because of geopolitical conflict, data centers further increase local power demand and can easily trigger resident backlash over higher electricity bills, environmental pollution, water stress, and land use. In other words, AI is no longer just a capital-market growth story; it is entering the intersection of local politics, energy policy, and household living costs.
This will change how markets price AI assets. In the past, investors mainly focused on AI model capability, user growth, token consumption, cloud revenue, GPU orders, and data-center delivery speed. After 2026, however, investors must also focus on a variable that had previously been underestimated: whether society is willing to pay for AI infrastructure expansion.
https://cryptohayes.substack.com/p/the-butterfly-touch
https://cryptohayes.substack.com/p/the-butterfly-touch
https://cryptohayes.substack.com/p/the-butterfly-touch
If the electricity, land, and water costs brought by AI data centers are ultimately borne by residents, AI will quickly become a political issue. Democrats and Republicans alike may build campaign narratives around whether ordinary people are subsidizing large technology companies. For capital markets, this means AI companies’ future freedom to expand is not unconditional. Once regulation, taxes, construction limits, environmental reviews, or power-cost-sharing mechanisms change, the long-term profit model of AI companies will have to be revalued.
In this environment, the Trump administration and the Republican camp face a dilemma. On the one hand, the United States needs to maintain its edge in the AI race, especially because AI infrastructure is viewed as a strategic asset in U.S.-China technology competition, military intelligence, automated production, and global data-sovereignty competition. On the other hand, energy prices, inflation pressure, and ordinary voters’ resistance to data centers may force politicians to send tougher regulatory or tax signals to the AI industry during election season.
Such political signals may not necessarily turn into long-term policy, but the market cannot ignore them. Asset prices do not wait for legislation to be enacted; they begin to re-discount the moment the policy narrative changes. If U.S. politicians begin using phrases such as “data centers are pushing up electricity bills”, “AI is taking jobs”, and “technology giants enjoy subsidies while ordinary people pay” as electoral language, the AI sector will move from a pure growth narrative to a high-valuation asset class carrying regulatory and populist risk.
For the crypto market, this macro environment has a dual effect. On the one hand, high inflation and high interest rates suppress global liquidity and weaken the valuation-expansion room for Bitcoin, Ethereum, and altcoins. On the other hand, if AI assets undergo valuation compression because of political and energy shocks, falling risk appetite will drag on crypto in the short term. Over a longer cycle, however, if an AI bubble burst leads to credit contraction, economic downturn, and policy rescue, renewed monetary easing may reopen upside potential for Bitcoin.
Therefore, the June 2026 macro environment should not be understood simply as “bullish” or “bearish”. It is more like a stress-test environment: energy shocks test inflation; inflation tests the Fed; electricity costs test AI expansion; AI valuations test U.S. equities; and risk appetite ultimately tests the crypto market.
1.2 The AI CapEx Supercycle: From Software Narrative to Energy, Credit, and Capital-Market Engineering
Over the past two years, the market’s understanding of AI has gone through three stages.
The first stage was the model-capability stage. After ChatGPT was released at the end of 2022, the market mainly focused on the capability frontier of large models, parameter scale, reasoning ability, coding ability, multimodal capability, and user growth. At this stage, the AI narrative was concentrated on model companies such as OpenAI, Anthropic, Google DeepMind, Meta, and xAI. Investors believed that whoever had the strongest model would control the future operating-system gateway.
The second stage was the compute-bottleneck stage. As model training and inference demand grew rapidly, the market began to realize that the truly scarce resource was not a single application, but compute, GPUs, HBM, advanced packaging, wafer capacity, and data-center delivery capability. Nvidia, TSMC, ASML, SK Hynix, Micron, Broadcom, and Astera Labs became the core beneficiaries of the AI value chain. The AI narrative expanded from “model companies” to “picks-and-shovels sellers”.
The third stage was the financialization of infrastructure. Entering 2025-2026, AI was no longer merely a technology-company R&D project, but a massive capital-expenditure cycle. Data-center construction requires land, energy, cooling, transmission, networking, GPUs, servers, leases, debt financing, and long-term customer commitments. AI infrastructure began to be packaged into joint ventures, lease agreements, private debt, asset-backed securities, and long-term revenue contracts. It evolved from a technology narrative into capital-market engineering.
Meta’s partnership with Blue Owl to develop the Hyperion data center is a typical example of this trend. Large technology companies are no longer simply building data centers with their own cash. Instead, through joint ventures, long-term leases, residual-value guarantees, and private-credit financing, they transform part of their capital expenditure into financial assets. On the surface, this can improve the apparent cash-flow profile of technology companies and reduce one-time CapEx pressure; in essence, however, it transfers AI infrastructure risk from corporate balance sheets to private-credit investors, insurers, pension funds, and bondholders.
This bears a certain structural resemblance to the 2008 subprime crisis. The 2008 problem was not that houses did not exist or that Americans did not need housing. It was that mortgages, ratings, securitization, and leverage were built on overly optimistic expectations for home prices. Today’s AI problem is also not that compute does not exist or that enterprises do not need intelligence. It is that data centers, GPUs, and long-term compute contracts are being financialized and built on an assumption: future AI revenue will be sufficient to cover today’s enormous capital expenditure.
That assumption has not yet been proven.
Large technology companies have strong cash flow, but AI CapEx is growing fast enough to change their capital-allocation logic. In the past, technology giants such as Apple, Microsoft, Alphabet, and Meta supported high valuations through high margins, share buybacks, and asset-light business models. Now AI is forcing them into a heavy-asset era. More capital is being invested in data centers, chips, electricity, and long-term infrastructure rather than buybacks, dividends, or high-margin software expansion.
This means the core basis for valuing technology stocks has changed. In the past, the market awarded large technology companies high valuations because they had asset-light models, high gross margins, high free cash flow, network effects, and global monopoly advantages. Now AI is pushing them toward models that are capital-intensive, energy-intensive, depreciation-intensive, and financing-intensive. Capital markets must answer one question: are these companies still software platforms, or are they becoming the power-and-compute utilities of a new era?
If they are still software platforms, current valuations may be explainable by high margins and high growth. But if they gradually become heavy-asset infrastructure operators, valuation multiples must be reduced. Traditional utilities, telecom operators, and data-center REITs find it difficult to enjoy software-company multiples over the long term. The danger of AI is that it may cause the market to buy a capital-expenditure model that is gradually becoming utility-like while applying software-company valuations.
Of course, the bullish case for the AI supercycle also has merit. AI inference demand may be far greater than training demand. Enterprise agents, code generation, robotics, office automation, financial analysis, drug discovery, industrial simulation, and personal AI agents could all generate enormous token consumption. If each watt of electricity can produce more tokens, and if inference chips, memory, networking interconnects, and model efficiency continue to improve, demand for AI infrastructure may be far from peaking.
This is also the core logic emphasized by AI bulls such as Gavin Baker: AI is not a SaaS bubble, but a physical-bottleneck supercycle composed of electricity, wafers, HBM, advanced packaging, and inference chips. As long as TSMC, ASML, HBM, and the power grid cannot rapidly become oversupplied, AI CapEx will not instantly spin out of control like the internet bubble. The true excess returns are not in chatbots themselves, but in the physical infrastructure that can reduce token costs, improve performance per watt, and solve inference bottlenecks.
This logic is not unreasonable. It reminds us that AI cannot be simplistically equated with the 2000 internet bubble. In 2000, many internet companies had no revenue, no moat, and no cash flow. Today, the main buyers of AI are the technology giants with the strongest cash flows in the world. Nvidia’s profits are real, cloud providers’ revenue is real, and enterprise demand for automation is real.
The problem is that real demand does not guarantee that all asset prices are reasonable. More importantly, returns are distributed very unevenly within the AI value chain. Picks-and-shovels sellers may make money while model companies burn cash; chip companies may earn high gross margins while cloud providers shoulder low-return capital expenditure; data-center developers may lock in leases while the ultimate risk is borne by the large technology companies that sign long-term commitments. Investors cannot focus only on the expansion of total AI demand while ignoring differences in capital returns across segments.
From the perspective of bubble identification, the key question for the current AI market is not “is AI useful?” but the following three questions:
First, can AI revenue growth cover CapEx growth? If cloud providers increase annual capital expenditure by hundreds of billions of dollars but incremental AI revenue does not grow at the same speed, free cash flow will face persistent pressure.
Second, will the financing cost of AI infrastructure continue to rise? When project financing shifts from cash flow to debt and private credit, interest rates, credit spreads, and investor risk appetite become critical variables.
Third, will the social cost of AI be politically repriced? If data-center power demand raises residential electricity bills, triggers local-government backlash, and tightens environmental reviews, AI expansion will be constrained by non-market factors.
Therefore, the core contradiction of the AI CapEx supercycle is this: technological demand is real, but capital markets may over-extrapolate that demand; infrastructure bottlenecks are real, but bottleneck-asset valuations may already reflect many years of growth; AI’s long-term value is real, but short-term capital returns may be eroded by energy, financing, and regulatory costs.
1.3 The Essence of the AI Bubble: Not a Technology Bubble, but a Mismatch among Cash Flow, Energy, and Social Carrying Capacity
To judge whether AI is in a bubble, one should not look only at share-price gains or technological progress. A more reasonable framework is to distinguish between a “technological revolution” and a “financial bubble”.
A technological revolution asks: will this technology change the way production is organized?
A financial bubble asks: have asset prices already pulled forward too much future return?
AI is clearly a technological revolution. Large models have already demonstrated real capability in code, text, images, video, voice, search, customer service, investment research, contract review, and enterprise workflow automation. As inference costs fall, model capabilities improve, enterprise data becomes integrated, and Agent workflows mature, AI’s impact on productivity is likely to continue expanding.
But AI may also contain a financial bubble. The bubble is not caused by fake technology, but by capital markets paying overly high valuations for very long-dated cash flows while underestimating financing costs, competition intensity, policy resistance, and demand uncertainty along the path.
The current AI bubble is mainly reflected in five areas.
First, valuations extremely extrapolate future growth. Market pricing for AI companies implies very high revenue growth, very high profit margins, and very long periods of competitive advantage. Valuation expectations for companies such as OpenAI, Anthropic, and SpaceX in private or IPO markets approach or even exceed the trillion-dollar level, meaning investors believe they can sustain high growth for many years and ultimately earn extremely high profits. But competition among model companies is intensifying, inference prices are falling quickly, open-source models are catching up, and the true elasticity of enterprise willingness to pay for AI is still being tested. High revenue growth does not necessarily equal high profit growth.
Second, CapEx and revenue are mismatched. AI data centers must be built in advance, while revenue is released gradually in the future. This model naturally carries cyclical risk. If demand continues to exceed expectations, early construction can produce huge returns; if demand falls short, data centers, GPUs, servers, and power contracts become sunk costs. Compared with software businesses, mistaken investment in AI infrastructure is harder to adjust quickly.
Third, financing structures are becoming more complex. AI infrastructure financing is expanding from companies’ own cash flow into bonds, private credit, joint ventures, leases, guarantees, and securitization. This increases construction speed, but also expands the scope of risk transmission. Once AI revenue expectations are revised down, not only equities but also credit assets may be repriced. At that point, risk will no longer be limited to technology-stock investors; it will spread to banks, private-credit funds, insurers, and pension funds.
Fourth, energy costs are uncertain. AI essentially converts energy into intelligence. Training and inference both require electricity, and data centers require stable power supply and cooling systems. Rising energy prices directly increase the cost of AI services; grid bottlenecks restrict data-center delivery speed; rising natural-gas prices affect the cost of new power generation. If geopolitical conflict continues to disturb oil and gas prices, AI companies’ margins and expansion speed will both be affected.
Fifth, politics and social carrying capacity matter. AI’s impact on employment, wages, electricity prices, land, water resources, and local environments is becoming an election issue. Ordinary voters may not hold large amounts of AI stocks, but they may bear higher electricity bills and greater job uncertainty. Capital markets previously assumed AI expansion could proceed without limit, but real society may not accept this. Once politicians begin using anti-AI, anti-data-center, and anti-technology-giant windfall-tax narratives to win votes, valuation risk for AI assets will rise materially.
This bubble structure resembles both the internet bubble and the subprime crisis, but it also differs from them in important ways.
The similarity with the internet bubble is that markets are paying high valuations in advance for a real technological revolution. The internet did change the world, but investors who bought large amounts of internet stocks in 2000 still suffered severe losses. Being right on the technological direction does not mean being right on the purchase price.
The similarity with the subprime crisis is that AI infrastructure is being financialized. Data centers, leases, guarantees, debt, private credit, and long-term contracts form a new asset-packaging chain. If the underlying cash flows are overestimated, the financial structure will amplify losses.
The difference from both is that the buyers of AI are mainly cash-rich technology giants rather than a large number of revenue-less startups or low-credit homebuyers. This means the AI bubble does not necessarily collapse as quickly as 2000 or 2008. It is more likely to follow a path of “local failures first, credit transmission later, and policy response afterward”.
In the first stage, the market will revalue the AI assets with the highest valuations, the lowest liquidity, and the most uncertain profitability. Examples include extremely expensive IPOs, model companies that have not yet reached profitability, small and mid-sized hardware companies dependent on long-term orders, and overhyped AI application stocks.
In the second stage, pressure will transmit to semiconductors, data centers, power equipment, and cloud providers. Investors will reassess order visibility, gross margins, inventory, customer concentration, and CapEx sustainability.
In the third stage, credit markets begin to react. Spreads on AI-data-center-related debt, private credit, lease guarantees, and project finance widen. Banks and non-bank institutions tighten lending, and companies reduce or delay capital expenditure.
In the fourth stage, the macroeconomy is affected. Because AI CapEx has already become an important component of U.S. economic growth, a slowdown in technology capital expenditure would pressure U.S. GDP, employment, industrial orders, and corporate earnings.
In the fifth stage, policy turns easier again. If markets fall too quickly, credit risk expands, and the economy enters recession, the Federal Reserve and fiscal authorities may again provide liquidity support. At that point, Bitcoin and high-quality crypto assets may see a new cyclical opportunity.
Therefore, the most dangerous aspect of the AI bubble is not whether it will burst tomorrow, but that it has become the core narrative on which global risk assets collectively depend. As long as AI keeps rising, U.S. equity indexes, semiconductors, data centers, power equipment, private valuations, risk appetite, and some crypto assets can all benefit. Once AI expectations reverse, the impact will not remain within technology alone; it will spread into the global liquidity structure.
2.1 Why the Crypto Market Has Not Captured All the Liquidity: AI as the “Black Hole” of Incremental Dollar Capital
Over the past decade, the crypto market developed a relatively clear macro trading framework: when global liquidity expands, U.S. dollar real rates fall, and risk appetite rises, Bitcoin usually performs well; when liquidity tightens, the dollar strengthens, and real rates rise, Bitcoin and altcoins usually come under pressure.
This framework was very effective in 2020-2021. Post-pandemic fiscal stimulus, monetary easing, zero interest rates, and abundant liquidity drove large amounts of capital into Bitcoin, Ethereum, DeFi, NFTs, GameFi, and risk assets. Bitcoin was viewed as “digital gold” and also as a hedge against dollar depreciation in an era of high liquidity.
But from late 2024 to mid-2026, this framework began to deviate. Dollar liquidity did not completely dry up: M2 expanded again, fiscal deficits remained high, and capital-market risk appetite did not disappear entirely. Yet Bitcoin significantly underperformed AI stocks. Nvidia, AI semiconductors, data centers, power equipment, and some private AI assets absorbed more incremental capital. Although the crypto market still experienced cyclical rebounds, it did not become the largest beneficiary of global liquidity expansion.
The reason is that AI replaced crypto as the most consensus-driven growth narrative in this round of global capital markets.
Capital chases expected marginal returns. After 2023, AI offered investors a story that traditional finance could understand more easily than crypto: real companies, real revenue, real orders, real capital expenditure, real IPO paths, and real index weights. By contrast, although the crypto market has made progress in Bitcoin ETFs, stablecoins, RWA, and on-chain trading infrastructure, many altcoins still lack cash flows, narratives rotate quickly, regulatory risk remains high, and investor trust has not fully recovered.
AI also attracts capital through broader channels than crypto.
Equity investors can buy listed companies such as Nvidia, Microsoft, Meta, Google, Amazon, Broadcom, TSMC, Micron, and Astera Labs.
Private-market investors can buy unlisted assets such as OpenAI, Anthropic, xAI, Cursor, Cerebras, and Databricks.
Bond investors can buy large technology-company bonds, data-center project bonds, private-credit products, and asset-backed securities.
Commodity investors can allocate to related assets such as natural gas, power equipment, copper, uranium, rare earths, and the HBM supply chain.
Local governments and sovereign funds can participate in the AI race through industrial funds, tax incentives, and infrastructure investment.
By contrast, although the capital-entry points for the crypto market are expanding, they remain concentrated mainly in BTC ETFs, ETH ETFs, exchanges, stablecoins, on-chain funds, and a small number of listed-company equities. AI has stronger capacity to absorb capital, and its narrative is easier to integrate with traditional financial balance sheets, so it is more able to absorb incremental dollar liquidity.
This is why the crypto market in 2026 can show a seemingly contradictory phenomenon: macro liquidity has not completely disappeared, but the crypto market still lacks a sustained primary uptrend. Liquidity is not distributed evenly. It flows toward the direction with the most imagination, the easiest institutional allocation, and the easiest financial-model explanation. In 2024-2026, that direction is AI, not crypto.
The deeper issue is that Bitcoin’s marginal buyer structure has changed.
Early Bitcoin buyers were mainly crypto-native users, miners, traders, and some macro investors. After 2020, institutions, listed companies, funds, and ETF investors gradually entered. By 2026, Bitcoin increasingly looks like a high-beta liquidity asset in global macro portfolios rather than an alternative asset completely independent of traditional markets.
This institutionalization has both advantages and disadvantages. The advantage is that Bitcoin’s legitimacy, liquidity, and role in asset allocation have improved significantly. The disadvantage is that Bitcoin is more easily included in the same risk budget. When institutions simultaneously hold AI stocks, QQQ, Bitcoin ETFs, and high-yield bonds, once portfolio risk budgets decline, they tend to cut exposures together rather than retain Bitcoin separately.
This raises Bitcoin’s correlation with Nasdaq, AI stocks, and overall risk appetite. Bitcoin still has a digital-gold attribute, but at the short-term trading level it looks more like a “high-beta technology risk asset”. When the AI sector rises strongly, Bitcoin may lag because capital is being absorbed elsewhere; when the AI sector falls, Bitcoin may fall with it because risk appetite declines. This is the crypto market’s hardest current problem: it does not necessarily rise when others rise, but it is vulnerable to falling when others fall.
Stablecoin supply is also a key way to observe crypto liquidity. Stablecoins represent on-chain dollar purchasing power, the dry powder inside the crypto market. If stablecoin supply continues to grow, capital is still waiting on-chain for opportunities. If stablecoin supply stagnates or slows, even a short-term BTC rebound will struggle to support a broad altcoin market. Since 2026, the fundamentals of stablecoins as payment, cross-border-transfer, and on-chain-settlement tools have continued to improve, but trading capital has not rotated into altcoins on a large scale.
Therefore, the crypto market’s biggest current problem is not the absence of a macro narrative, but the lack of control over capital leadership. Bitcoin’s long-term narrative remains clear: hedge against currency debasement, sovereign asset, global settlement asset, and digital gold. But short-term market pricing power has been taken by AI. AI represents future productivity, while Bitcoin represents the future monetary system. In an environment of limited capital, still-high interest rates, and constrained risk budgets, investors are more willing to prioritize AI stocks that can bring visible revenue growth and index-weight increases rather than wait for Bitcoin’s long-term monetary attributes to be repriced.
This does not mean Bitcoin’s long-term logic has failed. On the contrary, it means Bitcoin is waiting for the next macro switch. When AI capital expenditure slows, technology-stock valuations compress, credit markets reprice, and policy turns toward easing again, Bitcoin may regain liquidity leadership. At that time, the crypto market’s opportunity may not come from AI continuing to rise, but from monetary easing after the AI bubble clears.
2.2 How AI Risk Transmits to the Crypto Market: Four Channels through Equities, Credit, IPOs, and On-Chain Liquidity
The impact of the AI bubble on the crypto market is not single or linear. It transmits through at least four channels: the equity risk-appetite channel, the credit-contraction channel, the IPO liquidity-drain channel, and the on-chain liquidity channel.
The first channel is the equity risk-appetite channel.
AI has already become the core index weight and sentiment anchor of U.S. equities. Nvidia, Microsoft, Meta, Google, Amazon, Broadcom, and TSMC affect not only the technology sector, but also global risk-asset pricing. When AI stocks rise, investor risk appetite strengthens, leveraged capital becomes active, volatility falls, and the crypto market usually receives some support. When AI stocks fall, risk appetite declines, institutions reduce exposure to high-volatility assets, and crypto tends to come under synchronized pressure.
This channel is especially important for Bitcoin. The launch of BTC ETFs has made Bitcoin more allocable in traditional portfolios, but it has also made Bitcoin easier to manage through risk budgets. When Nasdaq volatility rises, the AI sector pulls back, and funds need to reduce portfolio beta, Bitcoin ETFs may become liquid assets that are easy to sell. In other words, ETFs brought long-term incremental capital, but also stronger linkage with traditional markets.
The second channel is the credit-contraction channel.
AI infrastructure construction requires massive financing. If AI CapEx continues to expand, bond markets and private credit will take on more financing responsibility. Data-center project bonds, technology-company bonds, lease guarantees, private-credit funds, and bank loans will all become the credit foundation of AI expansion.
Once the market starts to doubt whether AI revenue can cover capital expenditure, credit markets will first demand higher risk premiums. After financing costs rise, data-center project returns fall, technology-company free cash flow comes under pressure, and some projects are delayed or canceled. Credit contraction will further lower expectations for AI supply-chain orders, forming a negative feedback loop.
The impact on the crypto market is very direct. Crypto assets are highly sensitive to liquidity and credit cycles. During credit expansion, leverage increases, stablecoin supply grows, exchange volumes are active, DeFi yields rise, and altcoins are more likely to rally. During credit contraction, leverage exits, market makers reduce risk budgets, trading depth thins, and altcoin liquidity dries up. If the AI credit cycle enters contraction, the crypto market will struggle to remain isolated.
The third channel is the IPO liquidity-drain channel.
In 2026, expectations that very large AI or AI-related companies such as SpaceX, OpenAI, and Anthropic will enter public markets become major events for global capital markets. Giant IPOs inherently absorb market liquidity. Active funds, passive indexes, sovereign funds, pension funds, and retail investors may all adjust portfolios to participate in these high-profile listings.
If these IPOs are priced extremely high, have limited free floats, and deliver only modest initial gains or even break issue price, the market will quickly reassess AI private-market valuations and secondary-market absorption capacity. For the crypto market, giant IPOs have two effects.
In the short term, they drain risk capital. Capital that might otherwise have been allocated to BTC, ETH, or crypto-related equities may rotate into AI IPOs. This effect is especially strong when AI IPOs are packaged as “the next Nvidia”, “the world’s largest intelligent-infrastructure company”, or a “trillion-dollar model platform”.
In the medium term, poor IPO performance can become a catalyst for the AI bubble to burst. The market would realize that high private-market valuations cannot necessarily be digested by public markets. Once this recognition forms, AI private valuations, technology-stock valuations, and risk-asset appetite will all be hit, and the crypto market will also experience linked selling.
The fourth channel is the on-chain liquidity channel.
When the AI sector is strong, capital can easily be diverted within the crypto market. Some crypto-native capital will rotate into U.S. equities, AI stocks, pre-IPO opportunities, on-chain U.S. equity contracts, and AI-related derivatives. Traders will find that AI stocks have deeper liquidity, stronger narratives, more institutional participation, and possibly better short-term returns than altcoins. This weakens capital rotation in the on-chain altcoin market.
In addition, AI has changed the narrative structure inside the crypto industry. From 2023 to 2025, many AI tokens tried to obtain valuations by riding the AI boom, but most projects lacked real revenue and core technical moats. As the market’s understanding of AI infrastructure deepens, investors will increasingly distinguish between assets that truly benefit from the scarcity of AI production inputs and tokens that merely attach an AI label. This is a repricing of the crypto AI sector.
Over the long term, crypto and AI are not only competitors. There are also areas of convergence, such as decentralized compute, model marketplaces, data ownership, Agent payments, on-chain identity, stablecoin settlement, automated smart-contract execution, and RWA-collateralized financing. All may become real scenarios where AI and crypto combine. But in mid-2026, the market has not yet proven that these directions can absorb capital on the same scale as traditional AI infrastructure. Therefore, the crypto AI narrative still needs to move from concept to cash flow.
From the perspective of risk transmission, the AI bubble affects different crypto assets differently.
Bitcoin is most affected by macro liquidity and risk appetite. If an AI selloff triggers market deleveraging, BTC will be under pressure in the short term. But if policy subsequently eases, BTC may be the first to recover.
Ethereum is affected by both on-chain activity and its risk-asset characteristics. If AI risk cools on-chain trading, DeFi, NFTs, and altcoin activity, ETH may underperform BTC. But if stablecoins, RWA, and on-chain finance continue to grow, ETH still has fundamental support.
High-beta altcoins are the most fragile. Under an AI risk shock, market makers will shrink inventories, traders will reduce leverage, and altcoin liquidity will dry up most easily. AI tokens that lack cash flow, real users, and are driven only by narrative may face severe valuation compression.
Exchange platform tokens and derivatives-related assets depend on trading volume. If an AI selloff triggers high market volatility, short-term trading volume may rise. But if the market enters a sustained bear phase, user risk appetite will decline and platform revenue will also come under pressure.
Stablecoins and on-chain Treasury-like assets may be relative beneficiaries. When risk assets fall, capital returns to stablecoins, money-market funds, short-duration yield assets, and on-chain dollar tools. Stablecoins’ “cash” attribute within the crypto market will become more important.
Therefore, the AI bubble is not a U.S. equity issue unrelated to crypto. It has become one of the external pricing cores of the crypto market. Future crypto research cannot look only at BTC ETFs, halvings, on-chain data, and regulation; it must also place AI CapEx, technology-stock valuations, data-center debt, energy prices, and giant IPOs on the same risk map.
2.3 Research Framework for the Second Half of 2026: How to Monitor the AI Bubble, Crypto Correlations, and the Next Opportunity
Facing the changing relationship between the AI bubble and the crypto market, investors cannot judge the market with a single indicator. A more reasonable method is to build a multidimensional monitoring framework that observes macro conditions, AI, credit, equities, crypto, and political risk together.
The first category is energy and inflation indicators.
Investors should focus on Brent, WTI, natural gas, spot electricity prices, regional electricity prices, grid loads in data-center-concentrated regions, as well as U.S. CPI, PCE, core services inflation, and inflation expectations. AI is built on electricity, and the marginal cost of electricity is affected by natural gas, transmission, land, and regulation. If energy prices continue to rise, AI inference and training costs will be repriced, pressuring model-company gross margins and cloud-provider CapEx returns.
At the same time, energy prices are an important variable for Federal Reserve policy. If rising oil prices push inflation higher, the Fed will find it difficult to cut rates quickly and risk-asset valuations will come under pressure. If oil prices fall while the economy slows, the Fed will have room to ease again. For the crypto market, the most favorable environment is usually not simply high inflation, but “falling inflation + loose liquidity + recovering risk appetite”. In mid-2026, the market has clearly not fully entered that environment.
The second category is AI CapEx and order indicators.
Investors need to monitor capital-expenditure guidance from Microsoft, Amazon, Google, Meta, Oracle, SpaceX, and others, especially whether management teams begin using phrases such as “moderate spending”, “optimize returns”, “discipline”, and “capacity digestion”. If large technology companies move from “accelerated buildout” to “controlled pacing”, the market will quickly revise down revenue expectations for the AI supply chain.
At the same time, investors should watch orders, inventory, gross margins, and customer concentration for Nvidia, TSMC, ASML, SK Hynix, Micron, Broadcom, Astera Labs, and other companies. If demand shifts from training to inference, the beneficiaries will change. If inference prices fall too quickly, model-company revenue growth may not cover compute costs. If data-center delivery is constrained, hardware orders may be delayed.
The third category is AI credit-financing indicators.
This may be one of the most important future indicators, but it is underestimated by ordinary investors. The bursting of the AI bubble may not first occur in stock prices; it may occur first in credit markets. Investors need to watch large technology-company bond issuance, credit spreads, data-center ABS/CMBS pricing, private-credit fundraising, bank lending standards, lease-guarantee terms, and project-finance failures.
If AI-related debt-financing costs rise, it means the market has begun demanding higher risk compensation. As long as credit markets shift from “willing to finance without limit” to “requiring proof of cash flow”, the pace of AI CapEx expansion will be restrained. This will directly affect technology-stock valuations and will also affect the crypto market through risk appetite.
The fourth category is giant IPOs and lock-up expiration schedules.
Whether companies such as SpaceX, OpenAI, and Anthropic list successfully, whether their listing valuations are accepted by public markets, whether their share prices can hold after listing, and whether insiders sell after lock-ups expire will all become important signals for the AI bubble.
If giant IPOs surge on the first day and continue to rise, it means the market is still willing to pay a high premium for the AI narrative and risk appetite may continue. If IPOs open high and fade, lack sufficient trading absorption, and keep breaking their issue price after listing, it indicates that private valuations have failed to transfer into public markets and the AI narrative may enter a reflexive downward phase.
The implication for crypto is clear: successful giant AI IPOs will continue to absorb risk capital, and the crypto market may remain relatively weak. Failed giant AI IPOs will first cause risk assets to fall together, but in the medium term they may give crypto assets an opportunity to regain liquidity leadership.
The fifth category is politics and regulation.
Data centers, electricity prices, AI taxes, job displacement, model regulation, national security, data privacy, and antitrust will all become important issues in U.S. politics over the next two years. Investors need to monitor statements from the president, Congress, state governments, and local governments on AI data centers, especially policy changes in swing states and competitive districts.
If political language shifts from “AI is national competitiveness” to “AI makes ordinary people bear the cost”, the market will reprice regulatory risk in the AI sector. Even if policies are never ultimately implemented, campaign language itself can influence asset prices. What high-valuation assets fear most is not bad news itself, but a rise in uncertainty that raises discount rates.
The sixth category is internal liquidity in the crypto market.
Investors should focus on BTC ETF and ETH ETF net inflows, total stablecoin supply, exchange spot and futures volumes, BTC/ETH market depth, perpetual funding rates, on-chain active addresses, DeFi TVL, the share of altcoin market capitalization, and BTC dominance.
If BTC ETFs see persistent outflows when AI falls, stablecoin supply declines, and altcoin liquidity deteriorates, it indicates that the crypto market is still passively bearing external risk. If BTC stabilizes first after AI falls, stablecoin supply rises instead of falling, and ETF inflows resume, it indicates that the market is beginning to view Bitcoin as an alternative allocation after the AI bubble clears.
Based on the above framework, three scenarios can be set for the second half of 2026.
The first scenario is continued AI expansion and continued suppression of crypto. U.S.-Iran tensions ease, oil prices stabilize, the Fed keeps rates high but does not turn materially more hawkish, AI CapEx guidance keeps rising, and giant IPOs perform well. In this case, the U.S. equity AI chain remains the main line for global risk capital, and the crypto market may only have structural opportunities. BTC remains range-bound, ETH and altcoins lack a broad market, and AI-related on-chain assets need to prove real revenue to sustain valuation.
The second scenario is a moderate AI valuation correction followed by a short crypto selloff and then differentiation. Oil and electricity costs rise, some data-center projects are blocked, AI stocks consolidate at high levels, and giant IPOs underperform expectations without triggering a credit crisis. In this case, BTC may first fall with risk assets, but by less than high-beta altcoins. Stablecoins and on-chain dollar assets may benefit relatively. The market begins looking for the next main theme beyond AI. Crypto assets with real cash flow, trading revenue, stablecoin settlement, RWA, and infrastructure attributes may gain a relative advantage.
The third scenario is a rapid burst of the AI bubble that triggers credit contraction. Giant IPOs fail, AI stocks fall sharply, technology-company credit spreads widen, data-center financing is obstructed, banks and private-credit providers tighten, and the macroeconomy is hit. In this scenario, the crypto market will find it difficult to rise independently in the short term. BTC, ETH, and altcoins will all face pressure, especially illiquid and highly valued altcoins. But after policy rescue and expectations of monetary easing appear, BTC may bottom first and become a core asset in the next liquidity cycle.
A more accurate judgment is that AI is reshaping the flow of global liquidity, while the crypto market is in the process of searching for pricing power again.
In the past, crypto’s biggest narrative was “on-chain finance replacing traditional finance”. Now the biggest narrative in global capital markets is “AI improves productivity and restructures physical infrastructure”. These two narratives are not mutually exclusive, but they compete for capital. AI absorbs capital expenditure, while crypto absorbs crises of monetary trust. AI needs electricity and debt; Bitcoin needs liquidity and belief. AI’s risk lies in uncertain capital returns, while Bitcoin’s risk lies in the absence of a short-term cash-flow anchor.
The truly important intersection in the future may not be simple AI tokens, but the combination of “AI x crypto financial infrastructure”. For example:
First, stablecoins become the payment layer for AI Agents. When AI Agents can autonomously call services, buy data, pay for compute, and settle tasks, stablecoins may become the base currency of the machine economy.
Second, on-chain markets become the trading layer for AI assets. Pre-IPO assets, compute contracts, data-center revenue rights, GPU-rental income, and AI company equity derivatives may all be traded globally through on-chain markets or exchange derivatives.
Third, decentralized compute and data markets become complements to AI infrastructure. Although they will be hard-pressed to challenge centralized cloud providers in the short term, decentralized networks can provide more flexible long-tail compute, data labeling, model fine-tuning, and verification services in specific scenarios.
Fourth, exchanges become unified gateways for global risk assets. As user demand rises for 24/7 trading, USDC-denominated markets, high leverage, shorting, and cross-asset portfolios, crypto trading platforms have an opportunity to integrate AI stocks, on-chain assets, RWA, stablecoin yields, and derivatives into a single trading experience.
From this perspective, studying the AI bubble is not about simply judging whether AI will collapse. It is about understanding how global capital migrates among AI, U.S. equities, bonds, energy, and crypto. Opportunities in crypto often appear when old narratives are crowded and new narratives have not yet fully formed. In mid-2026, AI has already become a crowded narrative, while crypto is experiencing a phase of low expectations. Short-term risk still needs to be controlled, but medium- to long-term opportunities are also taking shape.
The final conclusion can be summarized in three sentences.
First, AI is a real technological revolution, but current capital-market pricing for AI infrastructure, model companies, and related stocks already contains obvious bubble components.
Second, from 2024 to 2026, AI absorbed a large amount of incremental global dollar capital, preventing the crypto market from fully benefiting from macro liquidity expansion and thereby increasing Bitcoin’s correlation with technology risk assets.
Third, if the AI bubble undergoes valuation compression, the crypto market will be hit in the short term. But after credit clears and policy eases, Bitcoin may again become one of the most elastic assets in the global liquidity-repair phase.
For investors, the most important task at the current stage is not to chase a single narrative, but to identify changes in capital flows. AI may still continue to rise, but it is no longer a risk-free growth story. The crypto market may continue to fluctuate, but it is also no longer an isolated high-volatility asset. In the second half of 2026, the real determinants of market direction will be whether AI CapEx can continue to be financed, whether energy prices can stabilize, whether politics reprices data-center costs, and whether Bitcoin can regain marginal pricing power over global liquidity.
3.0 HTX’s AI Business Practice: AINFT Extends the Platform from a Trading Gateway to an Intelligent Service Gateway
As AI becomes one of the core themes in global capital markets, crypto trading platforms are facing changes not only in asset prices and capital flows, but also in user behavior and entry points. In the past, users came to trading platforms mainly to deposit funds, trade spot assets, use derivatives, access wealth management products, conduct OTC transactions, and manage digital assets. However, as large language models become increasingly widespread, users are using AI tools more frequently for information processing, content generation, data analysis, strategy support, and intelligent productivity tasks. AI is gradually shifting from an external technology variable into a fundamental tool in users’ daily digital activities.
Against this backdrop, HTX is advancing its self-developed AINFT product within its ecosystem. AINFT aggregates mainstream large language model capabilities in the market and connects them with the Crypto payment system, forming a complete closed loop of “model capability + Web3 login + pay-as-you-go payment.” The key significance of this initiative is that HTX’s application of AI is no longer limited to employees using external models to improve R&D, customer service, or operational efficiency. Instead, HTX is beginning to build platform-level AI product capabilities for users. In other words, AI is not merely a back-end productivity tool; it is being incorporated into HTX’s product ecosystem and becoming a new extension of the platform’s service capabilities.
From a product logic perspective, AINFT first addresses the problem of fragmented access to different AI models. Today, mainstream AI services are distributed across platforms such as OpenAI, Anthropic, and Google. Ordinary users often need to register separate accounts, bind different payment methods, understand the capability differences between models, and switch back and forth between multiple products. This experience is not friendly to crypto users, nor does it align with the Web3 interaction model of “one wallet accessing multiple applications.” By aggregating mainstream large language model capabilities, AINFT integrates access to different models into a single product. Users no longer need to register across multiple AI platforms or switch between different products; instead, they can call different models through one unified entry point. This model does not attempt to retrain a closed model from scratch. Rather, it productizes and gateway-izes mature model capabilities already available in the market, converting them into a service layer better suited for crypto-native users. For HTX, this means adding a new AI service gateway beyond trading, market data, assets, and account systems, allowing the platform to extend from a single trading scenario into a higher-frequency intelligent service scenario.
The second key design of AINFT is Web3-native login and identity. Traditional Web2 AI products usually rely on email addresses, phone numbers, credit cards, and subscription accounts. Users need to bind personal identity information, payment methods, and usage records to centralized accounts. AINFT, by contrast, emphasizes login through wallet signatures such as TronLink, reducing registration friction and aligning more closely with crypto-native user habits. In this context, the wallet is not only a login tool, but also a Web3 identity gateway. Users can use their wallet to verify identity, authorize payments, and participate in future ecosystem interactions without relying entirely on traditional Web2 account systems. This design gives AI services a chain-native attribute from the very first point of entry, while also leaving room for future integration with points, token incentives, NFT rights, membership tiers, task systems, and on-chain behavior records. For a trading platform, this is particularly important because it allows AI services to move beyond being isolated tools and become product modules that can connect with the platform’s user system, asset system, and campaign system.
The third key design of AINFT is Crypto payment and a pay-as-you-go mechanism. Most traditional AI products adopt monthly subscription or fixed package models, where users need to pay a fixed fee in advance even if they only use the service occasionally. This model works for certain office users, but it does not necessarily fit the high-frequency, small-value, and flexible usage patterns of crypto users. AINFT adopts a pay-as-you-go model, allowing users to top up and call models based on actual usage needs rather than being locked into fixed subscription cycles. More importantly, once AI service consumption is connected with Crypto payments, the platform can link payment behavior, points rewards, token incentives, marketing campaigns, and user growth mechanisms together, creating a cycle of “using AI services — generating payment behavior — accumulating user activity — entering the platform ecosystem.” In other words, HTX’s promotion of AINFT is not simply about offering an AI tool. It is an exploration of how AI usage behavior can be coordinated with the platform’s payment logic, user growth, and ecosystem activities.
This product direction has clear synergy with HTX’s existing business system. HTX already has core capabilities in stablecoin liquidity, spot and derivatives trading, OTC services, wealth management, institutional services, and global user operations. AINFT can become a new high-frequency entry point beyond these existing businesses. In the past, users may have entered the platform only when trading, transferring funds, or using wealth management products. In the future, they may enter the HTX ecosystem more frequently for content generation, data analysis, information queries, market interpretation, office assistance, and intelligent services. For spot users, AI can improve the efficiency of information interpretation. For derivatives users, AI can help monitor market sentiment, news developments, and volatility risks. For new users, AI can lower the barrier to understanding crypto products and market information. For high-frequency users and professional traders, AI can serve as a tool for information aggregation, opportunity scanning, and decision support. As AI usage becomes more frequent, trading platforms that can embed AI capabilities into user accounts, wallets, payments, and asset systems may have the opportunity to evolve from “an entry point for trading assets” into “an entry point for using intelligent services,” while avoiding the AI bubble and building products with real application scenarios.
This report is published by HTX Research, the dedicated research arm of HTX. It examines how the AI investment supercycle is reshaping global liquidity flows and, in turn, the correlation structure between Bitcoin and the broader risk-asset complex.
References
https://cryptohayes.substack.com/p/the-butterfly-touch
https://wallstreetcn.com/articles/3774027#from=ios
The post first appeared on HTX Square.
