JPMorgan raises its AI capital expenditure forecast to $5.5 trillion and predicts Broadcom's 2027 AI revenue will surpass $150 billion.

Huoxing Finance reports that on June 21, J.P. Morgan released a research report upgrading its forecast for total AI capital expenditures before 2030 from $5.1 trillion to $5.5 trillion, with corresponding debt financing scaled up to $4.1 trillion. The report notes that surging demand for AI and persistent compute shortages are jointly driving this acceleration in spending; Google alone now processes 32 quintillion AI tokens per month, while companies like Microsoft and Uber have already exhausted their full-year AI budgets within just a few months this year. Among hyperscale cloud providers, the four major U.S. supercomputing giants—Google, Amazon, Microsoft, and Meta—are collectively guiding 2026 capital expenditures of approximately $700 billion to $725 billion, representing a year-over-year increase of about 75%. This is expected to surpass $1.1 trillion by 2027. J.P. Morgan forecasts their combined operating cash flow will exceed $900 billion by 2027, yet still fall short of covering their massive spending plans, making debt and equity financing the norm. In terms of chip beneficiaries, J.P. Morgan expects Broadcom to generate over $150 billion in AI-related revenue by 2027 (including ASIC/XPU and AI networking). Management has disclosed backlog orders exceeding $1 trillion for 2027, far surpassing the bank’s prior estimate of approximately $60 billion for 2026—a figure already considered conservative. Broadcom’s joint AI chip financing platform, AI XPV, formed with Apollo and Blackstone, is also highlighted in the report as a landmark structural innovation in GPU financing. Regarding financing structures, the five major supercomputing companies have completed approximately $240 billion in external financing year-to-date, with the high-grade bond market remaining the primary channel; AI-related high-grade bond issuance is projected to reach $2.1 trillion over the next five years. On power, J.P. Morgan has raised its forecast for data center capacity growth from 122 GW to 138 GW, but electricity remains the most critical bottleneck, prompting supercomputing firms to actively explore alternative solutions such as self-built power infrastructure.