JPMorgan Chase raises AI capital expenditure forecast to $5.5 trillion, expects Broadcom's AI revenue to exceed $150 billion by 2027.

According to Mars Finance, on June 21, JPMorgan Chase recently released a research report stating that it has raised its forecast for total AI capital expenditure by 2030 from $5.1 trillion to $5.5 trillion, with a corresponding increase in debt financing to $4.1 trillion. The report points out that the explosive growth in AI demand and the continued shortage of computing power are jointly driving this round of accelerated spending. Google's monthly AI token processing volume has reached 3.2 trillion times, while companies like Microsoft and Uber have exhausted their annual AI budgets within a few months this year. Regarding hyperscale cloud providers, the four major US supercomputing giants (Google, Amazon, Microsoft, and Meta) have a combined capital expenditure guidance of approximately $700 billion to $725 billion in 2026, representing a year-on-year increase of approximately 75%, and are expected to exceed $1.1 trillion in 2027. JPMorgan Chase predicts that their combined operating cash flow will exceed $900 billion in 2027, but this will still be insufficient to cover the massive spending plans, making debt and equity financing the norm. Regarding chip revenue, JPMorgan Chase projects Broadcom will achieve over $150 billion in AI-related revenue (including ASIC/XPU and AI networks) by 2027. Management has disclosed an order backlog exceeding $100 billion in 2027, far exceeding the bank's forecast of approximately $60 billion for 2026, considered a conservative estimate. The AI chip financing platform (AI XPV), jointly established by Broadcom, Apollo, and Blackstone, is also reportedly a landmark structural innovation in GPU financing. In terms of financing structure, the five major supercomputing companies have completed approximately $240 billion in external financing since the beginning of the year, with the high-rated bond market remaining the primary channel. It is projected that the issuance of AI-related high-rated bonds will reach $2.1 trillion over the next five years. Regarding power, JPMorgan Chase has raised its data center capacity growth forecast from 122GW to 138GW, but power remains the most critical bottleneck, and supercomputing manufacturers are actively exploring alternative solutions such as building their own power supplies.