JPMorgan: AI capital expenditure raised to $5.5 trillion, with Broadcom’s AI revenue expected to exceed $150 billion by 2027

BlockBeats message, June 21 — JPMorgan recently released a research report stating that the forecast for total AI capital expenditure before 2030 has been raised from $5.1 trillion to $5.5 trillion, and the related debt financing scale has been increased in parallel to $4.1 trillion. The report notes that the breakout of AI demand and the continued shortage of computing power are jointly driving the acceleration of this round of spending, with Google’s monthly AI token processing volume already reaching 3.2 quadrillion times, and companies such as Microsoft and Uber exhausting their full-year AI budgets within just a few months this year.

In terms of hyperscale cloud providers, the four major U.S. supercomputing giants (Google, Amazon, Microsoft, Meta) have issued combined capital expenditure guidance of approximately $700 billion to $725 billion for 2026, representing an increase of about 75% year over year. It is expected that in 2027, the figure will exceed $1.1 trillion. JPMorgan expects their combined operating cash flow to exceed $900 billion in 2027, but this will still be insufficient to cover the massive spending plans, making debt and equity financing a common approach.

On the chip beneficiaries side, JPMorgan expects Broadcom to achieve more than $150 billion in AI-related revenue in 2027 (including ASIC/XPU and AI networking). Management has disclosed that order backlog in 2027 will exceed $100 billion, far above the bank’s estimate of about $60 billion for 2026, which is viewed as a conservative characterization. Broadcom’s AI chip financing platform (AI XPV), jointly established with Apollo and Blackstone, is also reported to be a landmark structural innovation in the GPU financing space.

Regarding financing structure, the five major supercomputing companies have completed approximately $240 billion in external financing from the beginning of the year to date. The higher-rated bond market remains the main channel, and it is expected that over the next five years, the issuance scale of AI-related higher-rated bonds will reach $2.1 trillion.

On power, JPMorgan has raised its data center capacity growth forecast from 122GW to 138GW, but power is still the most critical bottleneck. Supercomputing vendors are actively exploring alternative solutions such as building their own power supplies.