What Happened
The debate over whether the artificial intelligence build-out is a bubble got a pointed answer this week from one of Wall Street's largest banks. In its midyear 2026 outlook, published June 24, JPMorgan Global Research raised its estimate for total global AI-related capital expenditures through 2030 to $5.5 trillion, up from the $5.1 trillion it had projected earlier in the year. Rather than flag the spending as a warning sign, the bank framed it as a durable and increasingly profitable cycle — a thesis it summed up in the blunt rhetorical question that headlines much of the coverage: what bubble?
The upgrade is driven by two forces, according to the bank: a larger wave of physical capacity expansion than previously modeled, and a shift in how that expansion is being financed. JPMorgan raised its estimate for debt tied to the AI build-out to $4.1 trillion, citing loan-to-cost ratios that now average above 85% and in some deals exceed 90%. In other words, the industry is not merely spending more; it is borrowing more aggressively to do so, taking advantage of favorable credit conditions while they last.
Why It Matters
At the center of the forecast is what JPMorgan calls the "AI upstream" — the data centers, chips, and supporting infrastructure that everything else in the industry depends on. That layer remains heavily concentrated in the United States, which the bank estimates commands roughly 85% of all AI and machine-learning venture capital. The build-out is the unglamorous foundation beneath every chatbot and coding assistant, and its sheer scale is what turns a software story into a heavy-industry one.
For readers in Seoul, there is a direct line to the local economy. JPMorgan expects meaningful spillover benefits to flow to China, South Korea, and Taiwan precisely because of their central roles in the semiconductor supply chain. The memory chips, advanced packaging, and fabrication capacity that AI accelerators require are disproportionately produced in this region, which means a spending wave decided in U.S. boardrooms lands as order books in East Asian fabs.
The spending itself is staggering in absolute terms. JPMorgan expects hyperscaler capital expenditures — the combined outlays of the cloud giants — to reach $650 billion in 2026 and surpass $1.1 trillion in 2027. The five biggest spenders, Meta, Alphabet, Microsoft, Amazon, and Oracle, already poured an estimated $342 billion into capex in 2025 alone, and Microsoft on its own is guiding toward roughly $190 billion this year.
How Markets Are Reacting
The bank's central claim is that this spending is being done from a position of strength rather than desperation. JPMorgan projects that hyperscaler operating cash flow will exceed $900 billion by 2027, comfortably above the capital they are committing, and notes that early returns on AI investment are already positive. That profitability argument is what separates the current cycle, in the bank's telling, from the speculative manias investors instinctively compare it to.
Markets have been receptive but not uncritical. Sentiment turned visibly more confident in late June, with a rally in chipmakers such as Micron as some traders set aside bubble fears, and Qualcomm using its investor day to lay out a data-center strategy it expects to generate more than $15 billion in annual revenue by 2029. Yet even JPMorgan concedes the open question: the scale of capacity being built assumes AI demand will keep growing fast enough to fill it, and investors remain genuinely divided on how long it will take for returns to match the money going in.
What Comes Next
The next constraint is physical, not financial. JPMorgan estimates that 122 gigawatts of new data center capacity is slated for deployment between 2026 and 2030, a figure that turns electricity supply, grid interconnection, and power generation into the gating factors for AI growth. The story of the next few years may be less about model releases and more about transmission lines, substations, and where the power to run all this silicon will actually come from.
Financing will be the other thing to watch. JPMorgan forecasts that high-grade corporate debt will account for more than $2.1 trillion of data-center financing over five years, with about $150 billion in U.S. hyperscaler bond issuance in 2026 and another $100 billion equivalent raised abroad. The bank also flags a structural risk hiding in plain sight: the entire cycle is concentrated among a small cluster of technology companies, so any single one pulling back on spending could ripple through the whole industry's growth outlook.
Closing Thoughts
What makes the report notable is not the headline number but the confidence behind it. For two years the reflexive framing around AI infrastructure has been the question of when, not if, the spending would outrun the revenue. JPMorgan's answer is that, for now, the math still works — cash flows exceed outlays, returns are positive, and the borrowing is a deliberate strategy rather than a sign of strain.
The caveat is in the phrase "for now." A forecast that leans on favorable credit conditions, continued demand growth, and the discipline of a handful of giant spenders is only as sound as those assumptions. If they hold, the $5.5 trillion estimate may yet prove conservative. If any of them slips, the same concentration that makes the cycle so powerful today is what would make a downturn move quickly. Either way, the center of gravity in AI has unmistakably shifted from clever software to the trillions of dollars of physical infrastructure being poured beneath it.
한글 요약
JPMorgan글로벌리서치가 6월 24일 발표한 2026년 중간 전망에서 2030년까지 전 세계 AI 관련 설비투자(capex) 추정치를 기존 5조1000억 달러에서 5조5000억 달러로 상향했다. 은행은 이 막대한 지출을 거품 신호가 아니라 지속 가능하고 점점 더 수익성 있는 사이클로 규정하며, AI 투자가 이미 플러스 수익을 내고 있다고 강조했다. 상향의 배경은 예상보다 큰 물리적 용량 확장과 자금 조달 방식의 변화로, AI 구축에 묶인 부채 추정치도 4조1000억 달러로 높아졌다. 대출 대비 원가 비율(LTC)은 평균 85%를 넘고 일부는 90%를 웃돈다.
전망의 핵심은 데이터센터·반도체·전력 등 'AI 업스트림' 인프라다. 이 영역은 미국에 집중돼 있지만(미국이 AI·머신러닝 벤처투자의 약 85% 차지), JPMorgan은 반도체 공급망의 중심인 한국·대만·중국으로 낙수 효과가 흐를 것으로 봤다. 하이퍼스케일러(메타·알파벳·마이크로소프트·아마존·오라클)의 설비투자는 2026년 6500억 달러, 2027년 1조1000억 달러를 넘어설 전망이며, 영업현금흐름은 2027년 9000억 달러를 상회해 투자 규모를 웃돌 것으로 추정됐다. 마이크로소프트 한 곳만 올해 약 1900억 달러를 집행한다.
다음 관문은 재무가 아니라 물리적 제약이다. 2026~2030년 새로 배치될 데이터센터 용량은 122기가와트(GW)로 추정돼, 전력 공급과 송배전망이 AI 성장의 병목으로 떠올랐다. 자금 측면에서는 5년간 2조1000억 달러 이상이 우량 회사채로 조달되고, 2026년 미국 하이퍼스케일러 채권 발행만 1500억 달러에 이를 전망이다. 다만 이 사이클이 소수 빅테크에 집중돼 있어, 한 곳이라도 투자를 줄이면 산업 전체 성장에 큰 충격이 될 수 있다는 구조적 위험도 함께 지적됐다.
참고: Fortune, Data Center Dynamics, Fortune.