Hyperscalers Just Raised Their AI Capex Budget to $750 Billion. The Markets Financing It Are Under Pressure
The four major hyperscalers — Meta, Microsoft, Amazon, and Alphabet — have raised their combined AI capital expenditure budget to $750 billion for 2026, a figure that Globe and Mail market analysis identifies as "set to cross $1 trillion next year and likely to rise further beyond 2027." Goldman Sachs estimates the combined AI-related capital expenditure of these four companies will reach $5.3 trillion between fiscal 2025 and 2030. McKinsey projects AI capex will require $6.7 trillion worldwide by 2030. These are commitments from companies with combined balance sheets large enough to self-fund most of this investment, and procurement decisions already in execution that make these numbers visible in physical economic activity.
The financial market implications of $750 billion in annual AI infrastructure spending extend beyond technology equity valuations. The debt capital markets, project finance structures, and equity issuance programmes required to fund AI's physical infrastructure are creating capital market activity that fixed income, infrastructure, and equity professionals are actively navigating. The SK Hynix U.S. share sale — raising 43 trillion won on the assumption that U.S. institutional investors will continue to value AI memory exposure at premium multiples — is one visible piece of a much larger capital mobilisation that is reshaping how financial markets allocate capital to physical technology infrastructure.
The SK Hynix U.S. Share Sale as a Sentiment Test
SK Hynix's decision to raise capital through a U.S. equity offering rather than additional Korean market issuance reflects a specific arbitrage opportunity in how AI memory companies are valued across markets. U.S. institutional investors — whose portfolio mandates and sector allocation frameworks have been shaped by the Nvidia-led AI infrastructure narrative — have been willing to pay premium earnings multiples for AI memory exposure that Korean domestic investors, more familiar with the cyclicality of memory markets through multiple boom-bust experiences, have been more cautious about. The U.S. share sale is SK Hynix capitalising on that valuation premium to raise capital at a cost of equity lower than Korean market issuance would provide, while simultaneously giving U.S. institutional investors direct access to the AI memory supply chain through a listed equity instrument.
The timing of the SK Hynix share sale — beginning to trade on Friday, the same week that Samsung's record profit triggered a $100 billion market cap loss on cycle peak anxiety — creates a specific market sentiment test. If SK Hynix shares price well and trade positively on their Friday debut despite Samsung's Tuesday decline, it signals that U.S. institutional investors continue to differentiate between company-specific earnings delivery and sector-level cycle concern. If SK Hynix shares underperform at launch, it signals that Samsung's sell-off has contaminated the broader AI memory sector's near-term sentiment in U.S. markets in ways that will affect the cost of capital for subsequent Korean technology company U.S. capital market transactions. The SK Hynix debut is a market structure signal about how U.S. capital markets are pricing AI memory exposure after Samsung's record-profit-sell-off paradox.
Nuclear Power, Cooling Infrastructure, and the Capex Ecosystem
The $750 billion hyperscaler AI capex figure includes not just chips and servers but the full physical infrastructure stack that makes AI compute viable at data centre scale. Globe and Mail analysis specifically notes that "massive AI data centre growth is benefiting several nuclear power generator and reactor makers, construction giants, cooling and water purifying companies and industrial manufacturers." This description of AI capex's economic reach is commercially important for financial markets professionals modelling the investable opportunity associated with the AI infrastructure build-out: the direct AI chip and server market is already well-covered by equity research and institutional allocation, but the nuclear power, cooling infrastructure, and water purification markets that serve AI data centres represent less-covered but structurally compelling investment opportunities whose revenue is as tied to AI capex spending as Nvidia's revenue is.
NuScale Power, Oklo, and X-energy — small modular reactor developers seeking deployment at or near data centres for dedicated power supply — have seen significant valuation re-ratings as Microsoft, Amazon, and Google announced nuclear power purchase agreements to guarantee clean baseload power for AI data centres. The financial market structure of nuclear power for AI data centres is particularly complex: nuclear projects require 10 to 15 year development timelines, require project finance debt structures with utility-scale credit enhancements, and generate returns that are not linear with AI market growth but provide the long-term power price certainty that hyperscaler data centre operations require. The financial structuring of nuclear-to-data-centre power supply is one of the most sophisticated capital markets challenges currently being worked on in the infrastructure finance community.
The Morgan Stanley Warning and What Capex Discipline Would Mean
Morgan Stanley's report warning that semiconductor stock weakness "would likely continue as investors are bracing for more capex discipline in the near-term on the part of the hyperscalers" is the specific risk scenario that financial markets are pricing into technology sector valuations. "Capex discipline" in this context does not mean hyperscalers will stop spending on AI infrastructure — their committed programme backlogs, signed supply contracts, and competitive positioning requirements make a spending pause extraordinarily unlikely. It means that the rate of capex growth may slow from the 2025 to 2026 acceleration pace, creating a period where AI infrastructure supply catches up with demand rather than persistently lagging it. That supply-demand rebalancing in AI infrastructure would be deflationary for AI chip and memory pricing — compressing Samsung's and SK Hynix's extraordinary pricing power — while being fundamentally healthy for the long-term AI economy's ability to sustain the productivity gains that justify the cumulative investment.
For financial services professionals modelling AI infrastructure investment, the distinction between "capex discipline" as spending reduction and "capex discipline" as growth rate normalisation is crucial. The hyperscalers' $750 billion 2026 commitment is not at risk of reversal — executed contracts represent irreversible near-term spending. The question is whether the 2027 trajectory toward $1 trillion maintains its current momentum or moderates to a more sustainable growth rate that allows AI revenue — from enterprise software licensing, cloud AI service subscriptions, and consumer AI product pricing — to grow into the capital investment already deployed. The timing of that revenue catch-up is the most commercially important technology finance question for H2 2026.
What This Means for Market Participants
Financial services professionals advising on AI infrastructure investment should separate the demand reality from the valuation question. The $750 billion hyperscaler AI capex commitment for 2026 is documented, contracted, and in execution — it is not a forecast but a spending programme. The valuation question is whether the market has priced the AI infrastructure investment cycle's full expected earnings impact into semiconductor, data centre, and AI infrastructure company valuations at a pace that leaves insufficient earnings upside to sustain momentum when growth rates normalise from current extraordinary levels. The SK Hynix U.S. share sale debut on Friday will provide the most current data point on how U.S. institutional capital markets are answering that valuation question in real time.