What bubble? JPMorgan says the $5.5 trillion AI capex explosion is profitable–for now
JPMorgan Global Research is doubling down on a central thesis for 2026: the surge in AI investment is not only durable, but increasingly profitable. In its midyear outlook, the firm points to a broadening capital expenditure cycle anchoring growth expectations. At the center is the “AI upstream” buildout—data centers, chips, and supporting infrastructure—still heavily concentrated in the U.S., which commands about 85% of AI and machine learning venture capital. Spillover benefits are expected in China, South Korea, and Taiwan, given their roles in semiconductor supply chains. JPMorgan now estimates total global AI-related capital expenditures will reach $5.5 trillion through 2030, up from $5.1 trillion. The increase reflects higher capacity expansion and a shift in financing. The bank raised its estimate for debt financing tied to the AI buildout to $4.1 trillion, citing higher loan-to-cost ratios. But the scale of spending can raise questions about whether AI demand will grow quickly enough to justify the capacity being built. While cloud providers such as Amazon Google, and Microsoft report rising AI revenue, investors still remain divided on how long it will take for returns to match investment levels. Regarding financing conditions loan-to-cost ratios average above 85%, with some exceeding 90%, reflecting both favorable credit conditions and perceived value creation. In some cases, investors are assigning outsized equity value to expansion—for example, a $15 million-per-megawatt investment translating into a $25 million increase in market capitalization. Hyperscalers—the primary drivers of AI infrastructure spending—appear to be entering this phase from a position of strength. JPMorgan expects their capital expenditures to reach $650 billion in 2026 and exceed $1.1 trillion in 2027. Profitability remains intact, with operating cash flow projected to surpass $900 billion by 2027. Still, the investment cycle remains concentrated among a small group of technology c