AI Data Center Capital Expenditure Surpasses $1 Trillion: An Investment Map of the Entire Industry Chain from GPUs to REITs

In the first half of 2026, the five major hyperscale cloud service providers—Microsoft, Amazon, Google, Meta, and Oracle—collectively raised their capital expenditure guidance. Morgan Stanley's internet stock team, after analyzing Q1 earnings reports, forecast that by 2026, the combined capital expenditure of these five hyperscalers will reach approximately $800 billion, further rising to $1.2 trillion in 2027, significantly upwardly revised from the previous estimate of $450 billion. Additionally, according to another Morgan Stanley report, hyperscalers will drive about 40% of the cash capital expenditure of Russell 1000 companies between 2026 and 2028, totaling over $2 trillion.

Bank of America Securities also substantially raised its expectations. Its analyst Vivek Arya’s team predicts that in 2026, global hyperscale cloud providers’ AI capital expenditure will exceed $800 billion, a 67% year-over-year increase, and will surpass $1 trillion in 2027. This judgment is based on the fact that Alphabet, Microsoft, Meta, and Amazon’s quarterly revenues all exceeded Wall Street expectations, with AI and cloud service demand becoming the main growth drivers—Meta raised its 2026 capital expenditure guidance to $125–$145 billion, and Amazon’s AWS segment grew 28%, the fastest pace since 2022.

A longer-term perspective is also worth noting. Data from Marvell cited at the 2025 AI Investor Day indicates that global data center capital expenditure was $435 billion in 2024, expected to reach $593 billion in 2025, and could break through $1 trillion by 2028, with a compound annual growth rate (CAGR) of 20% from 2025 to 2028. If the five largest US tech companies’ combined capital expenditure reaches about $650 billion in 2026, these figures are largely consistent.

NVIDIA CEO Jensen Huang stated at the GTC conference in March 2026 that the Blackwell and Vera Rubin AI chip series will achieve at least $1 trillion in cumulative revenue by the end of 2027, doubling the previous forecast of $500 billion. NVIDIA CFO Colette Kress, during the earnings call, pointed out that as proxy AI begins to penetrate various industries, AI infrastructure spending could reach $30–$40 trillion annually by the end of the decade.

Below, we systematically analyze the core beneficiary segments of this round of capital expenditure, from upstream to downstream.

GPU Chips and ASICs: NVIDIA Leads, Broadcom and Marvell Benefit Simultaneously

NVIDIA (NVDA) is the most direct beneficiary of this cycle. In fiscal year 2026, NVIDIA achieved revenue of $215.9 billion, up 65% year-over-year, with a GAAP gross margin of 71.1%. Data center is the absolute core, with full-year revenue reaching $193.7 billion, up 68%, accounting for about 90% of total revenue. In Q4, data center revenue was $62.3 billion, up 75% and hitting a quarterly high. From the revenue structure, hyperscalers are NVIDIA’s largest customer group for data center business.

With the continued volume of the Blackwell platform and upcoming Vera Rubin shipments, inference-side growth is accelerating. Huang defined the current stage as an “inference inflection point,” emphasizing that real-time computing demands for AI systems are becoming a new growth engine. However, not all of the $1 trillion infrastructure expenditure flows into GPUs. The ASIC custom chip market is expanding at an even faster pace.

Broadcom (AVGO) dominates this market. According to Yahoo Finance, Broadcom is expected to capture about 60% of the AI server compute ASIC market by 2027, with ASIC shipments projected to double. CEO Hock Tan stated during the earnings call that the AI chip market opportunity for the three major customers in fiscal 2027 will be between $60–$90 billion, and Broadcom is expected to secure a “reasonable share.” According to Dongwu Securities research, Marvell forecasts global data center capital expenditure will exceed $1 trillion by 2028, with AI acceleration compute related to $349 billion, and the ASIC market size revised upward to $55.4 billion, with a CAGR of 53% from 2023 to 2028.

The demand surge for AI servers also propagates to storage. HBM and high-performance DRAM have become another rigid demand outside of AI chips, with SK Hynix, Micron Technology (MU), and Samsung Electronics as key beneficiaries.

Server Assembly and System Integration

After chips are manufactured, the assembly of servers and system integration directly meet the demand for AI infrastructure spending. Major players include HPE, Dell Technologies, and Supermicro (SMCI). IDC data shows the global data center market was about $347 billion in 2024, expected to grow to $627–$650 billion by 2030, with server infrastructure investment being the main incremental driver.

Network Equipment: Upgrading AI Cluster Interconnects

AI cluster scale is expanding from thousands of cards to tens of thousands or even hundreds of thousands, driving demand for high-speed Ethernet and InfiniBand devices in backend network connectivity through scale-up and scale-out architectures.

Arista Networks (ANET), with its data center switches, leads among hyperscalers. Cisco (CSCO), as a traditional enterprise network leader, is accelerating its transformation into the AI data center network market. Their switches, routers, and optical modules will directly benefit from the continuous expansion of network bandwidth in 2026–2027.

Data Center Cooling: Liquid Cooling Market Accelerates

The power density of AI servers continues to rise—GPU rack power consumption has jumped from a few kilowatts in traditional servers to tens of kilowatts or higher—prompting a shift from air cooling to liquid cooling in data centers.

Market research firm Grand View Research’s GMI report indicates the global data center liquid cooling market was about $3.3 billion in 2025, expected to reach $10.55 billion by 2030, with a CAGR of 26.1%. Another research firm projects the 2024 market size at $870 million, growing to $10.7 billion by 2030, with a CAGR of 51.93%. Guotou Securities forecasts that the global new data center liquid cooling system market could exceed $50 billion by 2030, with a CAGR of 22% from 2026 to 2030.

In terms of competitive landscape, Vertiv (VRT) led the liquid cooling market in 2025 with over 11.3% share. The top five vendors (Vertiv, Schneider Electric, Rittal, Stulz, Boyd) collectively hold about 35%. Schneider Electric (SU), listed in Europe, also offers a full suite of liquid cooling solutions. Previous industry data estimates Vertiv’s market share in liquid cooling technology exceeded 60%, though figures vary across sources; GMI’s official market share report is the most authoritative.

Power Supply: Gigawatt-Level Rigid Bottleneck

Power demand for AI data centers is becoming the most severe bottleneck in infrastructure development. Evercore ISI reports that announced incremental power demand exceeds 125 GW, designating 2026 as a “critical year” for the power industry. SemiAnalysis forecasts global data center IT power demand will surge from 49 GW in 2023 to 96 GW in 2026, with AI consuming about 40 GW. Vertiv predicts global data center power demand will reach 140 GW by 2029, adding 100 GW in five years.

On the supply side, two types of companies benefit most: independent power producers selling electricity via spot or long-term contracts, and utilities holding large-scale generation assets.

Vistra Corp (VST) exemplifies this sector. In January 2026, Vistra signed a 20-year PPA with Meta, providing over 2,600 MW of zero-carbon power starting from late 2026 and reaching full capacity by 2034. According to Investing.com, Vistra’s diversified generation portfolio and retail business give it significant resilience and flexibility amid rising power demand.

NextEra Energy (NEE) is a leading US renewable energy provider. In March 2026, NVIDIA announced partnerships with six US energy giants including AES, Constellation Energy, NextEra, and Vistra, aiming to unlock up to 100 GW of idle US grid capacity for AI data centers. Google also signed a 25-year PPA with NextEra to restart the Illinois nuclear plant.

Energy Storage: From Backup Power to Grid Interaction

The power load of AI data centers is not constant but exhibits significant peaks and valleys driven by batch scheduling of training tasks and inference requests. This makes energy storage systems crucial not only as backup power but also for grid interaction and cost management.

The value of energy storage mainly lies in three aspects: smoothing load fluctuations to avoid high peak electricity prices; earning revenue as grid frequency regulation resources; and bridging power when grid capacity is insufficient. This trend benefits grid-scale storage integrators like Fluence Energy (FLNC), which see ongoing order growth.

Data Center REITs: Structural Opportunities in Land and Construction

The physical construction of AI data centers also creates structural investment opportunities in land holding and data center operations. Data center REITs are the most direct beneficiaries. WisdomTree notes that hyperscalers are partnering with companies like Digital Realty and Equinix—focusing on large-scale construction with robust power and cooling, and interconnection hubs for AI workloads. Data center REITs offer predictable cash flows from long-term leases, with lease terms ranging from 10 to 20 years, providing strong pricing power.

Major data center REITs include: Equinix (EQIX), the world’s largest data center REIT, valued at about $108 billion, with a presence in key global markets; Digital Realty (DLR), the largest wholesale data center REIT, with a strong position in core markets; and Iron Mountain (IRM), transitioning from traditional document management to a data center operator, serving approximately 240k clients across 61 countries.

Industry Chain Beneficiaries Overview

Upstream core: GPU chips led by NVIDIA (full-year data center revenue $193.7 billion); ASIC/custom chips led by Broadcom, with three major clients expected to reach a total XPU and network market size of $60–$90 billion by 2027; storage mainly by Micron (MU), with HBM demand providing revenue elasticity.

Midstream assembly and interconnection: server assembly mainly by HPE, DELL, Supermicro; network equipment mainly by Arista (ANET) and Cisco (CSCO).

Downstream supporting: liquid cooling led by Vertiv (VRT), with 11.3% market share; power supply mainly by Vistra (VST) and NextEra Energy (NEE); energy storage represented by Fluence (FLNC); data center REITs by Equinix (EQIX), Digital Realty (DLR).

Gate US Stock Trading: Connecting Crypto Assets with US Stock AI Infrastructure Investment

While analyzing beneficiaries along the AI industry chain, investors face a practical and urgent question: how to efficiently allocate between crypto assets and US stocks? Gate launched official US stock trading services in June 2026, allowing users to trade over 10k US stocks and ETFs directly within the Gate platform using USDT, covering NYSE, NASDAQ, NYSE Arca, NYSE American, BATS, and other major US markets and liquidity networks, with fractional trading starting from 0.01 shares, lowering entry barriers for US stock investment.

The core advantages of Gate’s US stock trading include three aspects. First, compliance structure—Gate has a strategic partnership with SEC-registered broker-dealer Alpaca, employing a comprehensive clearing agreement covering execution, clearing, settlement, custody, dividends, and corporate actions. Second, capital efficiency—users can switch positions between crypto assets and US stocks without account conversions; when crypto markets fluctuate, USDT in the same account can be immediately used to invest in AI infrastructure stocks. Third, no overnight holding fees—unlike perpetual contracts with funding rates or CFD products, Gate’s stock spot trading involves no funding or overnight costs, making it more suitable for medium- to long-term allocation. Additionally, Gate’s partner broker is a SIPC member, providing certain protections for securities assets under specific conditions.

To trade US stocks on Gate, users should: update the Gate app to the latest version (Android support available, iOS requires 8.21.5+); complete KYC verification and ensure regional eligibility; navigate to the “TradFi” section at the bottom of the app, then enter the US stocks zone; transfer USDT into the US stocks account via trading or asset pages, then participate in real-time stock and ETF trading. Currently, intraday trading is supported, with plans to expand to 24/7 trading gradually.

Conclusion

Morgan Stanley’s report indicates that in 2024, total capital expenditure of hyperscalers was only $260 billion, but expectations for 2026 are close to $800 billion, reaching $1.2 trillion in 2027—more than four times the growth in just three years. From chip manufacturing to power supply, liquid cooling systems, and data center REITs, the entire AI infrastructure industry chain is undergoing systemic re-pricing.

At the same time, this round of capital expenditure race also faces significant risks. First, whether profit growth can keep pace with capital expenditure intensity is the biggest uncertainty—if monetization of AI services underperforms, the credit quality and cash flow sustainability of hyperscalers will need reassessment. Second, whether US grid infrastructure expansion can match data center growth, and whether chip shortages will squeeze other sectors like consumer electronics, pose broader economic transmission risks. Investors participating in AI infrastructure themes should carefully evaluate the risk-return characteristics of each segment and adopt prudent allocation strategies.