JPMorgan Chase Insights: AI CapEx Nearing $870 Billion, Power and ROI Emerging as Next Challenges

TL;DR
· TrendForce has raised the global top nine CSPs' 2026 CapEx to around $830 billion, with a year-on-year growth rate increasing to 79%.
· Microsoft, Google, Meta, and Amazon continue to increase investment, driven by chip price hikes, data center, and power facilities upgrades.
· The semiconductor, data center, and power supply chain directly benefit, but the pace of AI revenue realization will still impact market patience.

The AI infrastructure budget of global cloud giants is still being revised upwards.

A TrendForce press release in May revealed that the total 2026 CapEx of the top nine CSPs globally has been raised to around $830 billion, with the year-on-year growth rate increasing from 61% to 79%. This figure covers Google, AWS, Meta, Microsoft, Oracle, as well as ByteDance, Tencent, Alibaba, and Baidu.

JPM-related strategic materials also indicate that the 2026 CapEx of large hyperscalers has exceeded $600 billion, but different statistical approaches to leasing, power, land, Chinese cloud companies, and emerging AI cloud platforms do not align. A more conservative interpretation is: regardless of a narrow or broad approach, the AI infrastructure budget for 2026 has not cooled down.

This round of spending is not just about buying more servers. GPUs, custom ASICs, network equipment, data center land, power access, and cooling systems have all contributed to the budget increase. For the market, AI infrastructure investment has shifted from tech companies' internal investment to semiconductor orders, data center leasing, utility investments, and the free cash flow of large tech companies.

A Broader $830 Billion Perspective, Cloud Providers' Budgets Continue to Rise

The $830 billion provided by TrendForce is the total sum of the top nine CSPs globally. This is broader than statistics focusing solely on large U.S. cloud providers and includes Chinese cloud companies.

This is also the reason why current AI CapEx figures can vary. A narrow approach is closer to the self-owned CapEx of a few American hyperscalers, while a broad approach includes financial leasing, data center construction, power-related investments, and additional cloud platforms. The numbers cannot be directly added together, but the trend is clear: investment in AI infrastructure continues to increase.

The disclosed figures of several top companies are already indicative of the magnitude of this trend.

Microsoft's Investor Relations materials show that the 2026 calendar year capital expenditure is approximately $190 billion. Alphabet's first-quarter earnings call and SEC filing indicate a 2026 CapEx expectation of $180 billion to $190 billion. Meta's first-quarter announcement and SEC filings provide a 2026 capital expenditure range of $125 billion to $145 billion, including financing lease principal.

Amazon's official statement earlier stated that the company's total capital expenditure in 2026 is around $200 billion. Third-party estimates in the market also suggest higher figures for AWS-related expenses, but this does not mean that Amazon has officially issued guidance of over $230 billion specifically for AWS.

In other words, the cloud giants have not slowed down due to the AI payback period controversy. They are still pre-building data centers, purchasing chips, and securing power for training, inference, and enterprise AI needs in advance.

Out of Microsoft's $190 Billion, $25 Billion Comes from Component Price Increases

Microsoft's numbers best reflect the complexity of this round of upward revisions.

In its FY2026 third-quarter earnings call, the company mentioned that the 2026 calendar year CapEx is around $190 billion, with approximately $25 billion coming from the impact of higher component prices. In other words, the increase in capital expenditure does not entirely correspond to a synchronous increase in the number of servers, GPUs, or AI computing power.

This is crucial for investors. The budget expansion is partly driven by demand and partly by cost.

On the demand side, training and inference continue to expand. Cloud providers need to purchase more GPUs, build larger clusters, and are also advancing in-house ASICs to reduce the unit cost of computing power. On the cost side, the increase comes from price hikes in high-end chips, storage, network equipment, power equipment, construction, and key components.

Therefore, what the semiconductor, network equipment, data center, and power sectors are seeing is order and construction demand, while large tech company shareholders are also seeing cash flow pressure, depreciation pressure, and margin pressure. If AI revenue growth can keep up, the market will continue to give these expenditures time. If revenue realization lags behind investment, the controversy will return to a more direct question: Can the money spent be turned into revenue fast enough?

Data Center Vacancy Rate Drops to 1.6%, Power Becomes a Harder Constraint

What supports the continued escalation of the cloud giants is not just management's stance but also supply-side tightness.

CBRE IM data shows that the vacancy rate in major North American data center markets is around 1.6%, with a pre-leased percentage of 74.3% for capacity under construction. This indicates that many new data centers have been locked in by cloud providers and AI customers before being delivered.

For CSPs, delaying by one year may mean missing out on customer demand. For data center operators, power equipment, and utility companies, order visibility is extended.

High-end chips are also in short supply. Next-generation AI chips like Nvidia's Blackwell are in high demand, and supply schedules will continue to affect server delivery. Even as cloud providers push for in-house ASICs, GPUs remain a key resource for training and high-end inference clusters.

An even harder-to-bypass constraint is power. AI data centers cannot just start working once servers are purchased; they also require substations, power access, backup power, cooling systems, and long-term power contracts. With a high proportion of U.S. power generated from natural gas, the additional load from data centers will also impact gas, electric grid infrastructure, and utility investments.

This is also why the AI capital expenditure benefit chain continues to expand outward. GPUs and ASICs receive orders first, followed by data center developers and lessors benefiting, driving power equipment, natural gas, electric grid expansions, and cooling system installations. However, the further down the chain, the more the projects rely on local approvals, grid connection progress, labor, and supply chain deliveries.

Money Is on the Way, but Revenue Realization Speed Needs to Catch Up

The most overhyped aspect of this AI capital expenditure cycle is equating a budget increase directly with AI commercialization being already achieved.

Reality is more nuanced. While the cloud giants are indeed ramping up, data centers are indeed under pressure, and high-end chip orders are still strong, part of the capital expenditure growth comes from component price increases, and power and delivery issues may slow down actual computational power deployment. Just because the 2026 capital expenditure is larger on the balance sheet does not mean available computing power will increase at the same rate.

What investors are more concerned about is not whether cloud providers will continue to spend money but how quickly this money will turn into revenue. The AI training demand supported the first round of investment; inference, enterprise applications, and cloud service pricing power will determine subsequent returns.

If the cloud revenue, software revenue, and productivity gains brought by AI cannot cover depreciation, energy, and financing costs, the market's patience for massive CapEx will decrease. The money is already on the way, chips and data centers are in line, and what might truly hinder this expansion cycle is not the budget but power access, project delivery, and revenue ramp-up speed.