Wave of AI Capital Expenditure Approaches, Projected to Reach $5.5 Trillion by 2030

Overview

J.P. Morgan released the AI Capex 2.0 research report, updating forecasts for artificial intelligence infrastructure investment and financing markets. The core view is that the AI capital expenditure wave is fully unfolding, with total investment projected to reach $5.5 trillion by 2030, with debt financing accounting for approximately $4.1 trillion. Power supply has become the most critical constraint for data center expansion, but developers are alleviating bottlenecks through innovative solutions such as behind-the-meter power supply and self-generated power. The report details the capital expenditure plans and financing strategies of hyperscaler cloud vendors (Google, Amazon, Meta, Microsoft, Oracle), as well as the role of various capital markets including investment-grade bonds, leveraged financing, securitization, and private credit in AI financing.

Core views

AI capital expenditure forecast increased. The report raised the 2030 AI capital expenditure forecast from $5.1 trillion in November to $5.5 trillion, primarily reflecting an increase in data center capacity growth expectations from 122GW to 138GW. Debt financing portion expected to be $4.1 trillion, project loan cost ratio average exceeds 85%, some projects exceed 90%. Hyperscaler cloud vendor capital expenditures continue to climb, expected to reach $650 billion in 2026, possibly exceeding $1.1 trillion in 2027, but operating cash flow expected to exceed $900 billion in 2027, early-stage investment returns positive. GPU and ASIC financing demand stands out. Future five-year AI accelerator financing expected to exceed $3 trillion, data center financing curve may stabilize around 2028, but chip financing will continue to grow until 2030. Customized ASIC adoption rate rising, expected ASIC to account for 42% of data center AI chip shipments in 2026, rising to 53% in 2027, ASIC shipment growth rate nearly triple that of commercial GPUs (+109% vs +39%). In power-constrained environments, per-watt performance advantage of ASIC becomes key competitiveness, customized ASIC data center costs approx $300-400 Billion/GW, while Nvidia GPU infrastructure exceeds $500 Billion/GW. Power is the key bottleneck. Data center expansion speed still constrained by power, base case scenario 2030 capacity growth 138GW, lower than unconstrained optimistic scenario (>150GW). Developers adopt behind-the-meter power, self-generated power and other innovative solutions to alleviate bottlenecks. Grid interconnection queues extend to 7 years in some regions, gas generation online in 2029 will relieve partial constraints. Investment-grade bond market is main force in AI financing. Expected future five-year investment-grade bond market to provide $2.1 trillion AI/Data Center financing, equivalent to $420 billion annually. Year-to-date AI-related capital expenditure financing in investment-grade bond market already reached $165 billion, surpassing all of 2025. Hyperscaler cloud vendors actively diversifying financing channels, non-USD issuance surges, year-to-date non-USD issuance accounts for 37% of global issuance (2025 was 14%). Leveraged financing and private credit have room for expansion. High-performance computing subsector weight in high-yield bond index expanded from 1.07% at end of 2025 to 2.75%. Expected 2026 HPC issuance $80 billion, 2026-2030 cumulative $350 billion. Private credit dry powder reaches $543 billion, capable of supporting AI financing. Securitization market expected 2026 US data center issuance $300-400 billion, currently 83% of SASB CMBS data center transactions located in markets with vacancy rate below 1%.

Analysis framework

Report adopts combined top-down and bottom-up analysis methodology. Top-down level, report starts with aggregate AI capital expenditure, decomposes into data center construction, GPU/ASIC procurement, power infrastructure etc., and forecasts financing needs for each part. Bottom-up level, report analyzes capital expenditure guidance, operating cash flow, financing gaps and financing strategies of five major hyperscaler cloud vendors (Google, Amazon, Meta, Microsoft, Oracle) one by one. Report also analyzes capacity, cost and adaptability to AI financing of each market type (investment-grade bonds, high-yield bonds, leveraged loans, securitization, private credit) according to capital market types. In power constraint analysis, report examines supply-side factors such as grid interconnection queues, generation capacity planning, behind-the-meter power supply solutions. In chip financing analysis, report evaluates asset characteristics such as silicon lifetime, residual value risk, maturity mismatch impact on financing structure.

Methodology notes

• Industry/Industrial Analysis FrameworkSupply and Demand Framework

  Power is supply constraint for AI data center expansion

  Report views power supply as key constraint condition for data center capacity growth, similar to traditional industry capacity bottleneck analysis. Power constraints determine upper limit of data center expansion, developers need to break bottlenecks through innovative solutions (such as self-generated power, behind-the-meter power supply).

• Company Fundamentals and Financial FrameworkFree cash flow analysis

  Capital Expenditure and Operating Cash Flow Financing Gap Analysis

  Report calculates financing gap by comparing capital expenditure guidance with operating cash flow forecasts of hyperscaler cloud vendors, thereby deriving external financing scale (debt, equity, project financing, etc.) required by each company.

• Valuation Method

  Loan-to-Cost (LTC) and Loan-to-Value (LTV) Analysis

  Report uses Loan-to-Cost (LTC) indicator to analyze project financing leverage levels, and points out that under market valuation rise background, 90% LTC actually equivalent to approx 60% LTV, reflecting financing space expansion brought by asset value rise.

• Industry/Industrial Analysis FrameworkUpstream-Middle-Downstream Chain Transmission

  Distribution and Financing Needs of AI Capital Expenditure in Each Industry Chain Link

  Report analyzes distribution of AI capital expenditure in industry chain links such as data center construction, chip procurement, power infrastructure, and evaluates financing characteristics and capital market adaptability of each link.

• Fixed Income and Credit AnalysisSpread analysis

  Relative Value Analysis of Tech Sector Credit Spreads and Market Index

  Report analyzes changes in tech sector credit spreads relative to market indices, pointing out tech spreads tightened from 34bp widening at end of 2023 to current 10bp widening, and evaluates spreads of data center bonds relative to hyperscaler cloud vendor bonds.

• Industry/Industrial Analysis Framework

  Total Cost of Ownership (TCO) Comparison between Customized ASIC and Commercial GPU

  Report compares Total Cost of Ownership between customized ASIC and commercial GPU, pointing out ASIC can achieve 30-40% TCO reduction, mainly due to elimination of supplier margins, cheaper ethernet networks and power savings.

• Cycle and Prosperity Framework

  Dynamic Balance Analysis of Token Consumption Volume and Per-Token Cost

  Report analyzes dynamic relationship between token consumption volume growth (logarithmic) and per-token cost decline (linear), pointing out need 1-2 years for cost decline to catch up with usage growth, this is typical manifestation of Jevons Paradox in AI field.

Asset mapping & comparison

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

• Alphabet (GOOGL)

  Beneficiary: Cloud revenue accelerating 63%, cloud backlog orders double to $462 Billion, actively pre-financing FY27-28 capex through debt and equity

  Strengths

  Strong operational capability, cloud profit margin expanding to 33%, liquidity $127 Billion

  Weaknesses

  Capex will consume almost all internal cash flow, off-balance sheet commitments growth needs attention

  Comparison

  Most diversified in non-USD issuance (61% non-USD)

  Risks

  Compute constraints limit revenue growth, off-balance sheet leases and guarantee commitments increase

• Amazon (AMZN)

  Beneficiary: AWS re-accelerating, cloud revenue growth 28% (15 quarters fastest), custom silicon becoming core story

  Strengths

  Retail business providing support, operating margin record 13.1%, financial flexibility high ($143 Billion cash)

  Weaknesses

  Free cash flow dropped from $26 Billion to $1 Billion, cumulative financing gap nearly $97 Billion

  Comparison

  Similar to Alphabet, pre-financing FY27 capex

  Risks

  Capital expenditure exceeding internal cash flow generation, reliance on external financing

• Meta Platforms (META)

  Beneficiary: Revenue growth 33%, ad impressions growth 19%, AI engagement improving

  Strengths

  Annual operating cash flow over $130 Billion, operating margin 41%

  Weaknesses

  Free cash flow turning negative this year, FY27 deteriorating, off-balance sheet obligations growing significantly

  Comparison

  Most active among hyperscalers to shift expenditure to off-balance sheet

  Risks

  Three-year cumulative financing gap approx $97 Billion, contract commitments increased to $238 Billion

• Microsoft (MSFT)

  Beneficiary: Cloud revenue $55 Billion (+29%), AI business over $37 Billion run rate

  Strengths

  AAA rating, low debt burden ($46 Billion), currently fully financing with internal cash flow

  Weaknesses

  Financing curve lags Alphabet and Amazon approx 1 year, FY27-28 financing gap $900-130 Billion

  Comparison

  Only hyperscaler not utilizing investment-grade bond market

  Risks

  Expected large-scale issuance starting second half 2026, market has not tested such scale permanent debt financing

• Oracle (ORCL)

  Beneficiary: OCI revenue $6 Billion (+93%), RPO increased to $638 Billion

  Strengths

  Contract revenue visibility $638 Billion, prepaid structure improving capital efficiency

  Weaknesses

  Most reliant on external financing among hyperscalers, gross debt $130 Billion

  Comparison

  First hyperscaler to include equity financing in capital plan

  Risks

  Trajectory depends on revenue growth achieving plan, lease commitments $261 Billion not reflected on balance sheet

• NVIDIA (NVDA)

  Beneficiary: GPU and custom ASIC financing demand growing, future five-year AI accelerator financing over $3 Trillion

  Strengths

  Commercial GPU market leader, gross margin over 70%

  Weaknesses

  Custom ASIC adoption rate rising may erode market share

  Comparison

  ASIC shipment growth far exceeding GPU (+109% vs +39%)

  Risks

  Custom ASIC TCO advantage (30-40% reduction) may drive customers to self-develop chips

• Broadcom (AVGO)

  Beneficiary: Custom ASIC adoption rate rising, 2027 AI revenue expected above $150 Billion

  Strengths

  Cooperating with Google, Meta etc. on custom ASIC, 2027 backlog orders over $100 Billion

  Comparison

  Key player in ASIC market, collaborating with Apollo/Blackstone on AI XPV platform

• CoreWeave (CRWV)

  Beneficiary: New cloud vendor, direct beneficiary of GPU financing demand

  Strengths

  First vendor to issue leveraged loans in High Performance Compute subsector

  Weaknesses

  Scale relatively small, market capacity limited

  Comparison

  Similar to Iren etc., suitable for relatively simple financing structures

Key data

• 2030 AI Capital Expenditure Forecast$5.5 TrillionHigher than November forecast of $5.1 Trillion
• Debt Financing Portion$4.1 TrillionAccounts for major portion of AI capital expenditure
• 2030 Data Center Capacity Growth138 GWHigher than November forecast of 122 GW
• 2026 Hyperscaler Cloud Vendor Capital Expenditure$650 BillionPossibly exceeding $1.1 Trillion in 2027
• 2027 Hyperscaler Cloud Vendor Operating Cash FlowAbove $900 BillionEarly investment returns positive
• Next Five-Year GPU/ASIC Financing DemandAbove $3 TrillionChip financing will continue to grow until 2030
• Investment-Grade Bond Market AI Financing (5 Years)$2.1 TrillionEquivalent to $420 Billion annually
• 2026-2030 HPC Issuance Cumulative$350 Billion2026 expected $80 Billion
• 2026 ASIC Share of AI Chip Shipments42%Expected to rise to 53% in 2027
• ASIC Shipment Growth Rate (2026)+109%Commercial GPUs at +39%
• Customized ASIC Data Center Cost$300-400 Billion/GWNvidia GPU infrastructure exceeds $500 Billion/GW
• Agentic Workload Token Consumption Multiple23.4xRelative to Conversational Workloads
• Report AI Cost Overrun Enterprise Proportion71%2026 Survey Data
• Revenue Required to Achieve 10% ROIC$700 BillionSteady-state Revenue Requirement
• Private Credit Dry Powder$543 BillionCapable of supporting AI financing
• 2026 Data Center Securitization Issuance Target$300-400 Billion$11.2 Billion issued year-to-date

Impact & implications

Report believes AI capital expenditure wave will have profound impact on multiple capital markets. In investment-grade bond market, AI-related issuers already account for 15.5% ($1.4 trillion), effectively becoming largest sector. Hyperscaler cloud vendors shifting from occasional issuers to structural presence, impacting index weighting, spread behavior and sector relative value. Tech sector credit spreads widened to cheapest level since 2004, report upgraded sector allocation from Underweight to Neutral. In leveraged financing market, HPC subsector has become important component of high-yield bond index, 2026 YTD return 10.64%, far exceeding high-yield index 1.63%. In securitization market, data center assets凭借 ultra-low vacancy rates (1.37%) and quality tenant credit, become highlight in CMBS/ABS market. Report points out if hyperscaler cloud vendors cannot achieve expected capital returns, situation similar to 5G investment may occur — investment is competitive necessity, but financial returns may fall below expectations. However, AI search applications already show revenue growth signs, Google adopting AI mode and Gemini has increased platform overall search volume.

Risks

• Power supply constraints may limit data center capacity expansion speed
• Grid interconnection queue lengthening (extending to 7 years in some regions)
• Regulatory review and cost sharing resistance (New Jersey, Pennsylvania, Virginia, Maryland etc. states)
• Rising costs and reduced supply of technical workers (power line electricians)
• Uncertainty in GPU/ASIC lifespan (3 years vs 7 years major impact on debt capacity)
• Maturity mismatch in chip financing (market prefers long-term, but silicon lifetime shorter)
• Adoption stagnation risk: costs too high may prevent certain use cases until additional compute further reduces token costs
• Hyperscaler cloud vendors may fail to achieve expected AI infrastructure investment returns
• Securitization market investors skeptical about residual value risks
• Large campuses >1GW difficult to adapt to current securitization market capacity

What to watch

• Pace of gas generation online in 2029 (alleviating power constraints)
• Sustainability and market depth of hyperscaler cloud vendor non-USD issuance
• Changes in custom ASIC adoption rate (expected 53% in 2027)
• GPU leasing price trends (indicator of residual value risks)
• Dynamic balance between agentic workload token consumption and per-token costs
• Deployment pace of private credit in AI financing
• Absorption capacity of data center securitization market for large campus financing
• Execution status of hyperscaler cloud vendor equity financing plans
• Regulatory treatment of behind-the-meter power supply solutions
• Progress in AI revenue monetization and achievement of capital expenditure returns