June 18, 2026—The Federal Reserve held the federal funds rate steady for the fourth consecutive time, and all three major U.S. stock indices closed lower. Yet, amid these macroeconomic headwinds, one underlying trend remains unshaken: capital expenditures (Capex) on AI infrastructure are expanding at a pace rarely seen in the history of technology. Nvidia (NVDA) closed the day at $204.65, demonstrating relative resilience despite broader market pressures.
This resilience is rooted in strong demand fundamentals. According to Goldman Sachs’ updated forecast released in June 2026, the four major hyperscale data center operators—Alphabet (Google), Amazon, Microsoft, and Meta—are expected to reach a combined Capex of $725 billion in 2026, up 77% from $410 billion in 2025. S&P Global’s estimates point to a similar figure, exceeding $700 billion. Including Oracle, the five largest cloud service providers (CSPs) are projected to spend a total of $760 billion on Capex, marking a year-over-year increase of 102.56%. The Magnificent Seven’s combined investment in AI and data center infrastructure is expected to hit $527 billion.
When capital of this magnitude converges in a single direction, understanding its flow becomes essential to grasping the industry’s overall landscape.
First Beneficiary Layer: Chips—The Primary Absorber of Capex
In this wave of capital, the chip sector stands as the most direct and concentrated beneficiary. A significant portion of the $725 billion Capex from the four major hyperscale cloud providers flows directly into GPUs, ASICs, and related semiconductor procurement. For every dollar spent on AI infrastructure, a substantial share returns to chip suppliers.
Spending guidance from these tech giants underscores this trend. Amazon (AWS) projects 2026 Capex at around $200 billion; Microsoft at about $190 billion; Meta’s guidance ranges from $115 billion to $145 billion; and Alphabet has raised its outlook to $180–190 billion. Combined, these four exceed $700 billion, with most of it directed toward AI data centers, proprietary chip development, and compute infrastructure.
Notably, this spending race is reshaping the financing structures of tech companies. In early June 2026, Alphabet completed an $84.75 billion equity offering—the largest single equity issuance in global history. Even with operating cash flow of roughly $174 billion over the past 12 months, Alphabet’s annual Capex of $180–190 billion is now outpacing its internal cash generation. Prior to this, Alphabet had already raised over $85 billion in the bond market, spanning six currencies and even issuing a rare 100-year sterling bond. Meta is also considering raising tens of billions through stock offerings.
This shift from "asset-light, high-margin" models to "asset-heavy, physical infrastructure" means chip suppliers now enjoy significantly extended demand visibility. Dell’Oro Group forecasts that Capex growth will accelerate further in the second half of 2026, driven by the rollout of NVIDIA Rubin systems and the hyperscalers’ custom accelerator platform upgrade cycles. Even during short-term sell-offs in chip stocks, the underlying demand remains robust—this is the core margin of safety that the Capex arms race provides for upstream suppliers.
Second Beneficiary Layer: Network Infrastructure—The "Vascular System" of Compute Power
Beyond chips, network infrastructure is the next major beneficiary. The rapid scaling of AI clusters is driving up bandwidth requirements both within and between data centers. Segments such as optical modules, switches, and high-speed interconnect chips are experiencing exponential growth in demand.
Research institutions highlight key investment opportunities in the global AI compute supply chain, particularly in CPUs, optical interconnects, and AI PCs—covering companies like TSMC, Nvidia, Intel, ASML, and Chinese players such as Zhongji Innolight. The rapid increase in token call volumes is causing persistent shortages of core compute resources in intelligent computing centers. The widespread adoption of AI inference workloads is pushing for lower network latency and higher bandwidth, prompting data center network architectures to evolve from 100G to 400G, 800G, and even higher speeds.
Morgan Stanley estimates that a 350% surge in token demand is one of the main factors driving hyperscale cloud Capex forecasts up from $450 billion to $800 billion. This data reveals how inference-side demand is pulling network infrastructure investment—every round of model inference requires data transmission across the network layer, so rising token volumes directly translate into increased demand for networking equipment.
Third Beneficiary Layer: Power and Cooling—The Invisible "Shovel" Business
Once compute capacity reaches a certain scale, power and cooling shift from "supporting costs" to "core bottlenecks." Gartner forecasts that global data center electricity consumption will reach 565 TWh in 2026, up 26% from 447 TWh in 2025. Power demand is set to rise from 104 GW to 132 GW, a 27% increase. By 2030, this figure is expected to hit 290 GW. AI-optimized servers will account for 31% of data center electricity consumption in 2026, and by 2027, their power use will surpass that of traditional servers.
Gartner Research Director Linglan Wang notes, "The surge in compute-intensive AI workloads is driving unprecedented growth in data center power consumption. Today, AI compute capacity is constrained by power supply, making reliable power delivery a new battleground for scaling and profitability in the global AI race."
On the cooling front, the high power density of AI servers is accelerating the shift from air cooling to liquid cooling. Once a niche approach, liquid cooling is now becoming mainstream infrastructure for high-power intelligent computing centers. JPMorgan predicts that the global AI server liquid cooling market will soar from about $8.9 billion in 2025 to over $17 billion in 2026. The chilled water unit market is expected to grow from $1.6 billion in 2026 to $12.7 billion by 2030. Companies like Delta Electronics, Auras, Chaun-Choung, and Wiwynn are positioning themselves to capture this system upgrade opportunity.
Morgan Stanley projects that data centers will face a 55 GW power shortfall. This supply-demand imbalance means that companies in power equipment, nuclear energy, and renewable energy projects are becoming key beneficiaries of Capex spillover. Unlike chips, power and cooling are not direct absorbers of Capex—they represent "rigid derivative demand" that emerges as Capex expands, with growth rates closely tied to the pace of data center construction.
Global Data Center Capex: A Trillion-Dollar Macro Narrative
Bringing these layers together, the global picture of data center capital expenditure becomes clearer. Dell’Oro Group has raised its 2026 global data center Capex outlook to over $1 trillion. The top four U.S. cloud providers have increased their data center Capex by 78%.
Looking further ahead, Goldman Sachs predicts that the four major hyperscalers alone could reach $5.3 trillion in Capex by the end of 2030. Nvidia CEO Jensen Huang projected during the May earnings call that global AI data center investment will soar to $3–4 trillion by 2030. While the precision of these forecasts remains to be seen, the direction is unmistakable: AI infrastructure investment shows no signs of slowing down.
The ROI Challenge: When Will Profits Materialize?
However, the flip side of massive Capex is uncertainty around ROI (return on investment)—the central point of contention in today’s market.
The Capex-to-revenue ratio has climbed to historic highs. Meta’s Capex-to-sales ratio for 2026 is projected at about 54%, Microsoft’s at 47%, and Alphabet’s at 46%. Goldman Sachs analysts note that consensus estimates put hyperscale cloud Capex at $770 billion in 2026—equivalent to 100% of operating cash flow.
This capital intensity means that linear growth in cloud revenue alone is no longer sufficient to quickly absorb the depreciation and financial pressure brought by Capex. The four major hyperscalers are shifting from the traditional model of "self-sustained expansion through operating cash flow" to a leveraged model that "relies heavily on capital markets—debt and equity—to fund asset-heavy physical infrastructure."
CITIC Securities still identifies three key mechanisms that could improve cloud ROI: accelerated year-over-year cloud revenue growth, increased AI adoption rates, and continued expansion in cloud service demand. However, the timeline for ROI improvement remains an open question. Gartner data shows that global AI spending will reach $1.76 trillion in 2025 and is expected to rise to $2.60 trillion in 2026. Within this massive spending pool, infrastructure investment accounts for the largest share (55%), but the speed at which returns are realized still depends on how quickly inference-side commercialization takes hold.
Conclusion
The 2026 AI Capex arms race is no longer just a "cash-burning contest"—it marks a full-scale restructuring of the industry value chain, from chips to power, from compute to energy. Annual spending of $725–750 billion—by any measure—constitutes an economic reality that cannot be ignored.
The logic of this beneficiary chain is clear: chipmakers (especially NVIDIA) are the first and largest direct absorbers of Capex; network infrastructure suppliers gain incremental orders as cluster sizes grow; and power and cooling providers benefit as "rigid derivative demand," with gains materializing later in the build-out cycle. The timing and elasticity of these three layers differ, but together they form a comprehensive industry narrative.
For the market, the real test may not be whether Capex continues to rise—current guidance from the major players makes that answer quite clear—but rather when, and how efficiently, these investments will translate into sustainable profits. The validation of ROI will determine whether this arms race ushers in a new technological era or simply repeats chapters from the dot-com bubble. Until that answer emerges, every link in the beneficiary chain will continue to absorb the largest capital flood in tech history.
