Jefferies warns that the AI capital-expenditure boom is entering a return-validation phase, while North Asian memory and gold remain key beneficiaries

Overview

In “Asia Strategy 3Q/26: Asia Maxima - AI crescendo,” Jefferies discusses global and Asian asset allocation, focusing on the sustainability of the US AI capital-expenditure boom, the extent to which North Asian semiconductor and memory industries benefit, US interest-rate and inflation risks, and relative opportunities across gold and major Asian markets. The report remains cautious on the US AI capital-expenditure chain, believing that the long-term risk of capital destruction is rising; however, it is more positive on upstream hardware, DRAM memory, and selected North Asian equity markets that receive capital-expenditure spending.

Core views

The report's core view is that the AI capital-expenditure arms race is still accelerating, but monetization capability has not yet been sufficiently proven, and capital expenditure increasingly depends on debt financing. If investors suddenly realize that hyperscalers, OpenAI, Anthropic, and similar entities cannot generate adequate returns from AI investment, their willingness to finance could reverse rapidly and be amplified by circular arrangements involving supplier-financing customers. The long-term base case remains substantial capital destruction from US AI investment, while cheaper Chinese open-source large models could gain market share. At the same time, concentrated DRAM supply, the practical constraints of Moore's Law, and long-term supply agreements are strengthening memory makers' pricing power, allowing the North Asian picks and shovels trade to continue benefiting.

Analysis framework

The report uses macro asset allocation, cross-regional equity comparison, earnings-estimate revisions, capital-expenditure and cash-flow pressure analysis, bond-yield and inflation monitoring, and industry supply-constraint analysis. It cross-validates AI capital expenditure through company earnings, macroeconomic GDP contributions, import structures, funding sources, and profit distribution across the supply chain.

Methodology notes

• Macro strategyCapital-expenditure cycle analysis

  AI capital expenditure as a percentage of GDP, fixed investment, and corporate cash flow

  By comparing AI-related capital expenditure with US GDP, nonresidential fixed investment, pretax profits of nonfinancial corporations, and hyperscalers' operating cash flow, the report assesses whether the current cycle is approaching a peak and evaluates the potential risk of capital destruction.

• Industry-chain analysisPicks and shovels framework

  Separation between beneficiaries and funders of capital expenditure

  The report distinguishes between hyperscalers that pay for AI capital expenditure and Nvidia, DRAM, and semiconductor hardware companies that receive it, arguing that the market primarily rewards the latter.

• Economic conceptJevons Paradox

  Efficiency gains can lead to higher total demand

  The report uses Jevons Paradox to explain how falling token costs can instead drive higher AI usage and demand for computing power.

• Technology cycleAmara's Law

  Overestimating the impact of technology in the short term and underestimating it in the long term

  The report compares the AI cycle with the Dot-com and railroad cycles, arguing that AI will be transformative, although excessive capital allocation may still cause cyclical losses.

Asset mapping & comparison

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

• US hyperscalers

  Funders of AI capital expenditure

  Strengths

  Still supported by earnings momentum generated by AI capital expenditure and demand for cloud services.

  Weaknesses

  Capital expenditure as a percentage of cash flow has risen sharply, debt financing has increased, and AI monetization remains insufficiently proven.

  Comparison

  Significantly underperforming memory makers such as Micron, Hynix, and Samsung.

  Risks

  If investors question the returns on AI investment, willingness to finance could fall suddenly, putting valuations and share prices under pressure.

• DRAM memory makers

  Beneficiaries of AI capital expenditure

  Strengths

  Highly concentrated global supply, increasing long-term sales agreements, and stronger supply constraints from Moore's Law are supporting sharply higher contract prices.

  Weaknesses

  The cyclical industry may still be affected by slowing demand or a peak in capital expenditure.

  Comparison

  The report shows that the three major memory makers have significantly outperformed the four major hyperscalers year to date and since 2023.

  Risks

  If the AI capital-expenditure cycle reverses, memory prices and earnings expectations could decline.

• North Asia equities

  Beneficiary region of the AI hardware supply chain

  Strengths

  Taiwan and Korea are benefiting from semiconductor, memory, and AI hardware demand, leading to higher market-cap rankings.

  Weaknesses

  Sensitive to global AI capital expenditure and the semiconductor cycle.

  Comparison

  North Asia has more direct exposure to AI hardware momentum than Asian markets that rely more heavily on domestic demand or foreign capital flows.

  Risks

  Disputes over AI capital-expenditure returns, a correction in US technology stocks, or a reversal in memory prices.

• Gold

  Beneficiary of dollar debasement and official-reserve allocation

  Strengths

  Central banks hold more gold than Treasuries, the dollar's share of official reserves is declining, and worsening US fiscal conditions support the long-term thesis.

  Weaknesses

  Entering a short-term consolidation phase after a historic rally.

  Comparison

  Compared with dollar assets, gold is supported by dedollarization and fiscal concerns.

  Risks

  If the dollar rebounds temporarily or real interest rates rise, gold could remain in consolidation.

• US Treasuries

  Global anchor for pricing risk assets

  Strengths

  The bond market remains relatively orderly for now.

  Weaknesses

  Persistent inflation, fiscal deficits, and a test involving new Fed leadership could push yields higher.

  Comparison

  A 10-year yield above 4.5% would signal danger for equities, while a break above 5% would be clearly negative.

  Risks

  A rapid rise in yields could trigger equity valuation compression and concerns about US fiscal financing.

Key data

• Expected capital expenditure by the four major hyperscalers this yearApproximately US$700bnThe four entities are Alphabet, Amazon, Meta, and Microsoft.
• Estimated AI capital expenditure next year including Oracle, Anthropic, OpenAI, and neo-cloudsMore than US$1tnEquivalent to approximately 3% of US GDP, 22% of US nonresidential fixed investment, and 33% of pretax profits of US nonfinancial corporations.
• Capital expenditure as a percentage of operating cash flow for the four major hyperscalers41% in 2023; expected to reach 92% in 2026Shows that AI capital expenditure is consuming significantly more cash flow.
• In-year bond issuance by the four major US hyperscalersUS$144bnSignificantly higher than US$83bn in 2025; Alphabet, Amazon, and Meta issued US$52bn, US$67bn, and US$25bn, respectively, in 1H26.
• Market share of the three largest DRAM suppliersEstimated at 89% in 1Q26Supply concentration is higher than before 2012, supporting memory makers' bargaining power.
• Increase in memory contract pricesMore than 200% year to date; more than 800% since early 2023Reflects tight DRAM supply and demand during the AI capital-expenditure cycle.
• Share-price performance of the four major hyperscalersDown 3.9% year to date on a market-cap-weighted basis; up 173% since early 2023Significantly lagging the three major memory makers.
• Share-price performance of Micron, Hynix, and SamsungUp 233% year to date on a weighted basis; up 960% since early 2023Shows that the picks and shovels trade has significantly outperformed the funders of capital expenditure.
• Contribution of US AI-related capital expenditure to actual GDP growth in 1Q261.13 percentage points, or 42% of the 2.68% year-over-year growthIndicates the US economy's high dependence on AI capital expenditure.
• US 10-year Treasury yield4.465% at quarter-end; recent high of 4.685% on May 19The report considers a break above 4.5% a warning signal for equities, while a break above 5% would be clearly negative.
• US CPI and fiscal deficitHeadline CPI at 4.2% year over year in May; fiscal deficit approximately 5.3% of GDPPersistent inflation and deteriorating fiscal conditions together create upside risks for interest rates.
• Information-processing equipment and software investment as a percentage of US nominal GDP4.88% in 1Q26Higher than the 4.46% peak during the fourth quarter of 2000's Dot-com era.

Impact & implications

For investors, the report suggests that the AI theme remains an important driver of global and Asian equities, but portfolios should more clearly distinguish between those funding capital expenditure and those benefiting from the supply chain. If returns on AI investment come under widespread scrutiny, hyperscalers and US growth stocks could face valuation and financing pressure; by contrast, DRAM, semiconductor hardware, and North Asian markets may continue to receive support from supply bottlenecks and earnings revisions. At the macro level, rising US yields, persistent inflation, and fiscal deficits may limit valuation expansion in risk assets, while gold remains supported by dedollarization and worsening US fiscal conditions.

Risks

• AI capital expenditure monetization falls short of expectations, causing hyperscalers and AI platforms to sharply reduce their willingness to finance.
• Circular arrangements involving supplier-financing customers could amplify a reversal in the AI capital-expenditure cycle.
• A break in the US 10-year Treasury yield above 4.5% or 5% could weigh on equity valuations and reignite fiscal concerns.
• Persistent US inflation, tariff pressure, and spillovers from the Middle East conflict could drive expectations for more hawkish monetary policy.
• DRAM and semiconductor stocks have already risen sharply; if AI capital expenditure peaks, earnings expectations and valuations could retreat.
• The contribution of US AI capital expenditure to GDP growth is excessively high; if related investment slows, macroeconomic growth momentum could weaken.

What to watch

• Whether the four major hyperscalers continue to weaken relative to the S&P 500 and in absolute terms.
• Whether AI capital expenditure continues to obtain financing from debt markets and how bond-issuance costs change.
• Changes in capital-expenditure guidance, operating cash flow, and depreciation expenses at Alphabet, Amazon, Meta, and Microsoft.
• DRAM contract prices, long-term sales agreements, and revisions to earnings expectations for Micron, Hynix, and Samsung.
• Whether the US 10-year Treasury yield breaks above 4.5% or moves further toward 5%.
• Changes in US CPI, PCE, corporate price surveys, and Fed policy expectations.
• Whether the cost advantage of Chinese open-source large models continues to challenge the commercialization of closed-source models such as those of OpenAI and Anthropic.
• Signals that gold's consolidation is ending amid a declining share of the dollar in reserves, central-bank gold purchases, and worsening US fiscal conditions.