The five largest US cloud and AI companies — Microsoft, Alphabet, Amazon, Meta, and Oracle — are on pace to spend roughly $775-800 billion on AI infrastructure in 2026, while US data center construction spending hit a record $75.2 billion annualized rate in July 2026, up 57% year-over-year. That combined spending puts the US on track toward a projected $7.6 trillion in cumulative AI infrastructure capital expenditure between 2026 and 2031, spanning compute, data centers, and power.
Data Center Investment in America 2026 – Introduction
Data center investment in the US reached a scale in 2026 that has no real precedent in modern corporate history. What began as a handful of hyperscalers racing to build enough computing capacity to meet surging AI demand has become the largest coordinated capital expenditure campaign ever undertaken by private companies, with the five biggest spenders alone committing more in a single year than most countries spend on their entire national infrastructure budgets. Q1 2026 earnings confirmed the scale of this commitment: Microsoft, Alphabet, Amazon, Meta, and Oracle collectively guided to roughly $775-800 billion in 2026 capital expenditure, a jump of about 64% from 2025 levels and nearly triple the roughly $238-410 billion the group spent as recently as 2024.
The construction side of this boom has become just as dramatic as the corporate spending commitments behind it. US Census Bureau data tracked private data center construction climbing from a $1.6 billion annualized rate in January 2014 to a record $75.2 billion in July 2026 — a figure that now exceeds general office construction, transportation infrastructure, and nearly every other category the Census Bureau tracks in its monthly construction survey. This report breaks down exactly where that money is going, who’s spending it, and what it means for the US economy, power grid, and semiconductor supply chain heading into 2027.
Interesting Facts about Data Center Investment in US 2026
| Category | Figure |
|---|---|
| Combined 2026 AI capex, Big 5 (MSFT, GOOGL, AMZN, META, ORCL) | ~$775-800 billion |
| Increase over 2025 hyperscaler capex | ~64% |
| US data center construction spending, July 2026 (annualized) | $75.2 billion |
| Year-over-year growth in construction spending | 57.2% |
| Full-year 2026 data center construction forecast | ~$700 billion (ConstructConnect) |
| Data centers’ share of all US construction spending | 2.3% |
| NVIDIA data center revenue, Q1 FY2027 (ended April 2026) | $75.2 billion, up 92% YoY |
| Projected cumulative AI infrastructure capex, 2026-2031 | $7.6 trillion |
| Additional US data center facilities in development | 1,500+ |
The headline number that best captures the scale of 2026’s data center boom is the speed of its growth rather than its absolute size. US data center construction spending grew nearly 79% in just two years — from $28.3 billion in April 2024 to $50.7 billion by April 2026, before accelerating further to $75.2 billion by July. That pace of growth has pushed data centers past general office buildings to become the largest single category within private office construction tracked by the Census Bureau, a milestone that would have seemed implausible just three years earlier, when data centers weren’t even broken out as their own category in federal construction statistics.
The other defining fact of 2026 is how concentrated the spending has become around a small number of companies and an even smaller number of hardware suppliers. NVIDIA’s $75.2 billion in single-quarter data center revenue — a 92% year-over-year increase — makes its quarterly data center business alone larger than the full-year revenue of most publicly traded technology companies, underscoring just how much of the “AI infrastructure spending” that hyperscalers report in their own earnings ultimately flows through a remarkably narrow hardware supply chain.
Hyperscaler Capital Expenditure in 2026
BIG-5 2026 AI CAPEX GUIDANCE (ESTIMATED RANGES, $ BILLIONS)
AMAZON ████████████████████████████████████████ $200-220B
ALPHABET ████████████████████████████████ $175-185B
META ████████████████████████ $115-135B
MICROSOFT ████████████████████ $110-175B (range across guidance updates)
ORACLE ████████ included in Big-5 total
| Company | 2026 Capex Guidance | 2025 Capex (for comparison) |
|---|---|---|
| Amazon | ~$200-220 billion | ~$125 billion |
| Alphabet (Google) | ~$175-185 billion (incl. $70-74B on data centers specifically) | ~$91 billion |
| Meta | ~$115-135 billion | ~$72 billion |
| Microsoft | ~$110-175 billion (raised multiple times through the year) | ~$90 billion |
| Combined Big-5 total (incl. Oracle) | ~$775-800 billion | ~$410-480 billion |
Data Source: Company earnings calls and SEC filings, Futurum Group, CFA Institute analysis
The scale of hyperscaler capital spending in 2026 is best understood through how often guidance has been revised upward throughout the year. Amazon, for instance, raised its full-year capex guidance to approximately $200-220 billion as the year progressed, driven primarily by continued buildout of AWS data centers and custom Trainium AI chips. Alphabet guided to $175-185 billion in total 2026 capital expenditure, with roughly $70-74 billion specifically earmarked for data center construction and networking equipment, while Meta’s range climbed to $115-135 billion, including a 1-gigawatt data center under construction in Ohio and a Louisiana facility that could eventually scale to 5 gigawatts.
Taken together, independent analysis from the CFA Institute and Futurum Group converged on a combined figure of roughly $775-800 billion for the five largest spenders in 2026, a number confirmed through companies’ own Q1 2026 earnings disclosures and representing an increase of approximately 64% over 2025’s already-record spending levels. CreditSights estimated that roughly 75% of this 2026 hyperscaler capex — approximately $545 billion — is specifically AI-related rather than general-purpose cloud infrastructure, reflecting just how thoroughly AI workloads have come to dominate these companies’ investment priorities. For more on the chip supply chain absorbing much of this spending, see our AI Chip Statistics report.
US Data Center Construction Spending in 2026
US DATA CENTER CONSTRUCTION SPENDING (SEASONALLY ADJUSTED ANNUAL RATE)
Jan 2014 █ $1.6B
Apr 2024 ████████ $28.3B
Apr 2026 ████████████████████ $50.7B
Jul 2026 █████████████████████████████ $75.2B
| Period | Annualized Construction Spending | YoY Change |
|---|---|---|
| January 2014 | $1.6 billion | — |
| April 2024 | $28.3 billion | — |
| April 2026 | $50.7 billion | +27.4% |
| July 2026 | $75.2 billion | +57.2% |
| Full-year 2026 forecast | ~$700 billion | +81% vs. 2025 |
Data Source: US Census Bureau Value of Construction Put in Place survey (C30), ConstructConnect, Axios
US Census Bureau data tells a clean, escalating story of just how fast physical data center construction has accelerated. The annualized spending rate crossed $50 billion for the first time in April 2026, up 27.4% from a year earlier, before surging further to a record $75.2 billion by July 2026 — a 57.2% jump over July 2025 levels. First-quarter 2026 construction spending alone hit $44.7 billion, a 28% year-over-year increase and the strongest single quarter on record at the time, with ConstructConnect’s full-year 2026 forecast projecting total annual spending approaching $700 billion, an 81% jump over 2025.
Data centers have now become genuinely dominant within the broader category of private office construction that the Census Bureau tracks, representing roughly 52% of all private office spending and surpassing not just general office buildings but also the entire transportation infrastructure category — airports, rail terminals, and marinas combined. Geographically, the South accounts for more than half of all planned data center spending currently in the pipeline, with more than 1,500 additional facilities in various stages of development nationwide, and the average individual data center project cost climbing to roughly $475 million, up sharply from $177.9 million just a few years earlier as facilities have grown larger and more power-dense. For more on how this construction boom is straining local power infrastructure and household electricity costs, see our Data Center Electricity Cost Statistics in US report.
Where the Money Actually Goes: Chips, Power, and Shells
APPROXIMATE BREAKDOWN OF A TYPICAL AI DATA CENTER BUDGET
CHIPS / SILICON (GPUs, ACCELERATORS) ███████████████ ~50-60%
POWER/COOLING INFRASTRUCTURE ███████ ~20-25%
BUILDING SHELL/CONSTRUCTION █████ ~20%
| Spending Category | Approximate Share of Total AI Infrastructure Budget |
|---|---|
| Chips/silicon (GPUs, accelerators, memory) | ~50-60% |
| Power and cooling infrastructure | ~20-25% |
| Building shell/physical construction | ~20% |
| NVIDIA’s share of the AI accelerator market by revenue | ~80-90% |
| NVIDIA’s projected 2026 CoWoS wafer allocation (TSMC) | 595,000 wafers, ~60% of global CoWoS demand |
Data Source: NVIDIA SEC filings, Axios, Semiconductor Industry Association
Despite the eye-catching construction spending figures, the physical building itself represents a surprisingly modest share of what hyperscalers call “data center investment.” Industry analysis consistently puts construction — the concrete, steel, and physical shell — at roughly 20% of total project cost, with the remaining 80% split between power and cooling infrastructure and, by far the largest single category, the chips themselves. NVIDIA alone captures an estimated 80-90% of AI accelerator market revenue, meaning nearly every dollar hyperscalers report as “AI infrastructure spending” eventually flows through an extremely concentrated semiconductor supply chain anchored by a small number of chip designers and an even smaller number of advanced foundries.
That concentration shows up starkly in NVIDIA’s own financial results: the company’s data center segment generated $75.2 billion in revenue for the quarter ended April 2026 alone, a 92% year-over-year increase, making its single-quarter data center business larger than the full-year revenue of most publicly traded technology companies. NVIDIA’s dependence on Taiwan Semiconductor Manufacturing Company’s advanced packaging capacity is similarly concentrated — the company is projected to require 595,000 CoWoS wafers in 2026, representing roughly 60% of TSMC’s entire global CoWoS production capacity, a bottleneck that increasingly constrains how quickly new AI data center capacity can actually come online regardless of how much capital hyperscalers are willing to spend. For a deeper look at this chip supply chain and the broader US manufacturing push behind it, see our Semiconductor Manufacturing Statistics in US report.
The Long-Term AI Infrastructure Buildout Through 2031
PROJECTED ANNUAL AI INFRASTRUCTURE CAPEX
2026 ████████ $765-800 billion
2027 ████████████ $1.2 trillion
2028 ██████████████ $1.4 trillion
2031 █████████████████████ $1.6 trillion
| Year | Projected Annual AI Infrastructure Capex |
|---|---|
| 2026 | $765-800 billion |
| 2027 | ~$1.2 trillion (+50% vs. 2026) |
| 2028 | ~$1.4 trillion |
| 2031 | ~$1.6 trillion |
| Cumulative total, 2026-2031 | ~$7.6 trillion |
Data Source: Goldman Sachs, Bloomberg Intelligence, industry analyst projections
Analysts broadly expect 2026’s extraordinary spending levels to be merely the starting point of a much longer buildout. Baseline projections from major financial institutions put cumulative AI infrastructure capital spending across compute, data centers, and power at approximately $7.6 trillion between 2026 and 2031, with annual spending climbing from roughly $765-800 billion in 2026 to a projected $1.6 trillion by 2031 — meaning the current pace of investment, however extraordinary it already appears, is expected to roughly double again within five years. US hyperscaler capital spending specifically is projected to reach $1.2 trillion in 2027, a 50% increase over 2026, before climbing to $1.4 trillion in 2028.
That trajectory has prompted genuine debate among analysts about sustainability, since the aggregate spending now significantly outpaces the revenue growth AI products themselves have generated so far, raising questions about whether the current capex super-cycle reflects rational anticipation of future demand or a more speculative buildout that could eventually face a painful correction. What is not in dispute is the physical scale already locked in: Project Stargate, the joint venture spearheaded by OpenAI and Oracle, has committed nearly $400 billion of its planned $500 billion, four-year total across active sites, with its flagship Abilene, Texas campus already housing an estimated 450,000 NVIDIA GB200-equivalent GPUs.
Power Demand and Grid Constraints in 2026
US DATA CENTER ELECTRICITY DEMAND
2023 BASELINE (LBNL) ████████████████ 176 TWh (4.4% of all US electricity)
2026 (IT + COOLING, TOTAL) ████████████████████████████ 75.8 GW
| Metric | Figure |
|---|---|
| US data center electricity consumption, 2023 baseline | 176 TWh (4.4% of total US electricity) |
| US data center power demand, 2026 | 75.8 GW |
| Share of all new US electricity demand growth attributable to data centers | ~50% (IEA) |
| EIA forecast, 2026 electricity load increase from data centers | +1.9% nationally |
| Microsoft’s unfulfilled Azure order backlog (power-constrained) | $80 billion |
Data Source: Lawrence Berkeley National Laboratory, International Energy Agency, EIA, Goldman Sachs
The physical constraint increasingly limiting how fast new data center capacity can come online isn’t capital or even chip supply — it’s electricity. Lawrence Berkeley National Laboratory’s baseline estimate of 176 terawatt-hours of annual data center electricity consumption as of 2023, already 4.4% of total US electricity use, has grown into a 2026 total power demand figure of roughly 75.8 gigawatts, with the International Energy Agency attributing approximately half of all new US electricity demand growth directly to data centers. That dynamic means essentially every new power plant being built in the US today is being built, directly or indirectly, to serve AI infrastructure load.
The clearest illustration of this grid bottleneck is Microsoft’s own disclosed $80 billion backlog of unfulfilled Azure cloud orders — not demand that hasn’t materialized, but paying customers actively waiting for power-constrained data center capacity that simply doesn’t exist yet. Utilities have responded by planning historic capital investment of their own: broader US utility spending tied specifically to AI data center demand is projected to reach approximately $1.4 trillion between 2026 and 2030, a buildout analysts warn could pass nearly $700 billion of its cost through to ordinary residential ratepayers via higher electricity bills over the same period.
Frequently Asked Questions
How much are US tech companies spending on data centers in 2026?
The five largest spenders — Microsoft, Alphabet, Amazon, Meta, and Oracle — collectively guided to approximately $775-800 billion in 2026 capital expenditure, confirmed through Q1 2026 earnings disclosures, an increase of roughly 64% over 2025 spending levels.
What is the current US data center construction spending rate?
US Census Bureau data showed data center construction spending reaching a record $75.2 billion seasonally adjusted annual rate in July 2026, up 57.2% from July 2025. Full-year 2026 spending is forecast to approach $700 billion.
Which company is spending the most on data centers in 2026?
Amazon is generally the single largest spender, with 2026 capital expenditure guidance around $200-220 billion, much of it directed toward AWS data centers and custom Trainium AI chips.
How much of data center spending actually goes to chips versus construction?
Industry estimates put the physical building shell at roughly 20% of total project cost, with power and cooling infrastructure accounting for another 20-25%, and chips and silicon — dominated by NVIDIA, which holds an estimated 80-90% of the AI accelerator market — making up the remaining 50-60%.
How much total AI infrastructure spending is projected through 2031?
Analyst projections put cumulative US AI infrastructure capital expenditure, spanning compute, data centers, and power, at approximately $7.6 trillion between 2026 and 2031, with annual spending projected to climb from roughly $765-800 billion in 2026 to $1.6 trillion by 2031.
Is the US power grid able to keep up with data center demand?
Not fully. Data centers now account for an estimated 50% of all new US electricity demand growth, and companies like Microsoft have disclosed tens of billions of dollars in unfulfilled cloud orders specifically because power-constrained data center capacity isn’t yet available, making electricity supply the primary near-term bottleneck on the sector’s growth.
What is Project Stargate?
Project Stargate is a joint venture led by OpenAI and Oracle aimed at building out massive AI computing infrastructure in the US, with a planned total investment of $500 billion over four years. As of 2026, nearly $400 billion has already been committed across active construction sites, including its flagship Abilene, Texas campus.
Disclaimer: The data research report we present here is based on information found from various sources. We are not liable for any financial loss, errors, or damages of any kind that may result from the use of the information herein. We acknowledge that though we try to report accurately, we cannot verify the absolute facts of everything that has been represented.
