AI trades require rotation, and semiconductors’ relative strength versus Hyperscalers is unlikely to persist over the long term

Overview

This issue of Flows & Liquidity focuses on internal rotation within AI trades, U.S. money creation, MicroStrategy’s impact on the Bitcoin market, and cross-asset fund flow and positioning indicators. The core question is whether the sustained strength of semiconductor stocks relative to Hyperscalers is sustainable, and whether expanding U.S. liquidity can continue to support financial assets.

Core views

The report argues that valuation and earnings divergence within the AI value chain needs to converge. JPMorgan’s base case leans toward the constructive scenario, in which Hyperscalers and AI model providers catch up with semiconductors through better AI monetization, revenue, and earnings; however, the market’s key concern is the negative scenario in which capital spending slows after 2027, dragging on semiconductor demand and the broader AI trade. At the same time, U.S. M2 proxy indicators continue to expand, supporting financial assets such as U.S. equities, but low cash allocations leave asset prices more vulnerable to negative shocks.

Analysis framework

The report uses relative performance, consensus capital spending expectations, credit spreads, short interest, AI compute pricing, LLM token pricing, proxy money supply indicators, ETF flows, option skew, and cross-asset positioning monitoring to track AI trades, the liquidity environment, and risk appetite from multiple dimensions.

Methodology notes

• Relative performance and valuationAI value chain relative performance tracking

  Compare the relative performance of U.S.-listed Hyperscalers and semiconductor stocks, and assess whether the divergence is sustainable by combining market cap, sales mix, stock prices, and credit spreads.

  If semiconductor outperformance comes from broad expansion across the AI value chain, it may lead to a constructive rotation through earnings catch-up by Hyperscalers; if it comes from customer profits being squeezed, it could weaken future capex and feed back negatively into semiconductor demand.

• Capital spending expectationsHyperscalers consensus capex expectations

  Track the growth path of Hyperscalers’ capital expenditures using Bloomberg-aggregated bottom-up analyst consensus forecasts.

  The report notes that capex growth is close to 100% year over year this year, but consensus expectations indicate a material slowdown after 2027, which is a key basis for market concern over the negative AI trade scenario.

• Market sentimentShort interest monitoring

  Observe short interest as a percentage of free float for baskets of Hyperscalers, semiconductors, AI beneficiaries, and AI-vulnerable companies.

  Short interest in Hyperscalers has risen since last October and accelerated further in May and June, indicating growing investor concern about capex and earnings pressure.

• AI commercializationAI compute price and LLM token price tracking

  Track NVIDIA Hopper GPU rental prices and the LLM token spending index to assess the monetization ability of Hyperscalers and model providers.

  The higher compute prices are, the stronger the ability of Hyperscalers to maintain or improve margins; the report says there were signs of improvement in April and May, but some pullback occurred in June.

• LiquidityU.S. M2 proxy indicator

  Use the sum of U.S. commercial bank deposits and U.S. money market fund assets under management as a proxy for U.S. M2 money supply.

  This indicator shows that the pace of U.S. money creation continues to strengthen in 2025 and 2026, mainly supported by bank balance sheet expansion and the Federal Reserve balance sheet shifting to modest expansion.

• Cross-asset allocationCross-asset fund flow and positioning monitoring

  Combine fund flows, trading activity, volatility, skew, and speculative positioning indicators across global equities, bonds, ETFs, commodities, currencies, and options markets.

  This framework is used to assess capital preference for risk assets relative to bonds and cash, as well as crowding and potential reversal risk across different asset classes.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Semiconductor and AI chip stocks

  Currently the main beneficiaries of AI trades, but exposed to rotation and capex slowdown risk.

  Strengths

  They have continued to outperform over the past several months, supported by upward revisions to Hyperscalers’ capex and demand for AI chips.

  Weaknesses

  If customers cut capex due to stock price pressure, wider credit spreads, or earnings pressure, semiconductor demand could be hit.

  Comparison

  Performance versus Hyperscalers has been too strong, and the report believes the current divergence will be difficult to sustain over the long term.

  Risks

  Capex slowdown after 2027, crowded AI trades, and declining expectations for memory and chip demand.

• Hyperscalers

  Potential beneficiaries of rotation and also key customers driving semiconductor demand.

  Strengths

  If AI monetization, revenue, and earnings improve, they could capture a larger share of the AI value chain.

  Weaknesses

  Their stock prices have been broadly flat over the past year, valuations have de-rated, and credit spreads have widened relative to semiconductors.

  Comparison

  They have lagged semiconductors, but there is room for catch-up in the constructive scenario.

  Risks

  Capex plans may be compressed by market pressure, and rising short interest reflects investor concerns.

• AI application beneficiaries

  May benefit as gains spread across the AI value chain.

  Strengths

  Short interest in the JPAIADPT Index has been relatively contained since early April and is below the level at the start of the year.

  Weaknesses

  Realization of gains depends on AI application rollout and enterprise customers’ willingness to pay.

  Comparison

  Compared with Hyperscalers and some semiconductor names, market short pressure is lower.

  Risks

  AI commercialization falls short of expectations, token prices decline, and end demand remains weak.

• AI-vulnerable companies

  A segment for which the market continues to express concern.

  Strengths

  If concerns about AI substitution ease, there may be room for a sentiment recovery.

  Weaknesses

  Short interest in the JPAIVUL Index has risen steadily since the start of 2025 and is clearly higher than in other AI-related groups.

  Comparison

  Investor concern is more concentrated than for AI application beneficiaries and Hyperscalers.

  Risks

  AI-driven substitution, margin compression, and continued valuation pressure.

• U.S. equities

  Supported by U.S. money creation and expanding liquidity.

  Strengths

  Money creation is stronger than nominal GDP, providing liquidity support to financial assets, especially equities.

  Weaknesses

  Global non-bank investors hold relatively low cash allocations, leaving limited risk buffers.

  Comparison

  Relative to cash and bonds, equities may still benefit from liquidity transmission.

  Risks

  Negative shocks could trigger demand to rebuild cash, similar to the 2020 pandemic, the 2022 inflation shock, or geopolitical shocks.

• Bitcoin and MicroStrategy

  MicroStrategy’s policy shift introduces two-way flow risk into the Bitcoin market.

  Strengths

  Weak sentiment could become a contrarian bullish signal under certain conditions.

  Weaknesses

  Selling Bitcoin to pay dividends weakens its one-way buyer narrative and increases market uncertainty.

  Comparison

  Compared with a typical company, MicroStrategy has greater impact on market liquidity and sentiment because it holds about 4% of Bitcoin supply.

  Risks

  Insufficient U.S. dollar reserves, future Bitcoin sales, delays in U.S. crypto market structure legislation, and mutual amplification between the company’s valuation and Bitcoin prices.

Key data

• Hyperscalers capex growthAbout 100% year over yearThe report says Hyperscalers’ capex plans, after revisions, are close to historical peak growth rates this year.
• 2027 Hyperscalers capex expectation gapJPMorgan analysts about $1,150bn; consensus about $925bnJPMorgan’s internal forecast is above market consensus, and the gap reflects disagreement over the negative scenario.
• U.S. money creationAbout $1.6tr in 2025; tracking about $1.8tr in 2026M2 money supply is proxied by U.S. commercial bank deposits plus U.S. money market fund AUM.
• 2024 U.S. money supply expansionAbout $1.2trPreviously, cumulative growth from May 2023 to the end of 2024 was about $2.2tr, or about 9.6%.
• MicroStrategy U.S. dollar reserves$2.55bn, covering about 17 months of dividendsThe company targets coverage of at least 12 months of preferred dividends and interest expense; the report argues that 24-36 months of coverage would better reassure investors.
• MicroStrategy Bitcoin sale32 BTCThe company disclosed selling Bitcoin from May 26 to May 31, 2026 to pay preferred stock dividends.
• MicroStrategy year-to-date Bitcoin purchases$13.7bnThe report says its purchases account for about 70% of JPMorgan’s estimate of overall digital asset flows.
• Share of Bitcoin supply held by MicroStrategyAbout 4%As a large Bitcoin holder, its selling could have an amplified effect on market volatility and expectations.

Impact & implications

If the constructive scenario plays out, AI trades may rotate from semiconductors toward Hyperscalers, AI model providers, and AI application beneficiaries, easing the valuation pressure created by the current divergence. If the sharp capex slowdown implied by consensus expectations is confirmed, semiconductor demand, AI-related credit, and equity valuations could face a more persistent correction. Meanwhile, expanding U.S. liquidity still supports risk assets, but low cash allocations mean that if macroeconomic or geopolitical shocks emerge, investors rebuilding cash positions could amplify downside in asset prices.

Risks

• Hyperscalers’ capital spending slows materially after 2027, leading to downward revisions to semiconductor demand expectations.
• Semiconductor trades become overly crowded; if the relative performance gap does not converge, it could trigger a larger correction in the AI theme.
• AI compute prices and LLM token prices fell in June, and if weakness persists, it will undermine the AI commercialization and margin improvement thesis.
• Rising equity and debt financing costs for Hyperscalers may suppress willingness for future capital spending.
• Low cash allocations make financial assets more vulnerable to forced de-risking under macro, policy, or geopolitical shocks.
• MicroStrategy’s sale of Bitcoin creates expectations of two-way flows, potentially increasing volatility in Bitcoin and crypto assets.

What to watch

• Whether Hyperscalers’ 2027 and beyond capex guidance tracks closer to JPMorgan’s higher forecast or converges toward the consensus slowdown path.
• Whether the performance gap between semiconductors and Hyperscalers closes constructively through gains in Hyperscalers.
• Changes in short interest across baskets of Hyperscalers, semiconductors, AI application beneficiaries, and AI-vulnerable companies.
• Whether NVIDIA Hopper GPU rental prices and LLM token prices resume the improvement trend seen in April and May.
• The contribution of U.S. commercial bank asset growth, the Federal Reserve balance sheet, and money market fund AUM to the M2 proxy indicator.
• Whether MicroStrategy increases U.S. dollar reserve coverage to 24-36 months, as well as progress in U.S. crypto market structure legislation.