- ↑
UTMCAP TMT - Presentation Final Push
Preliminary Search
- Datacentre infrastructure spend over time
- Then look at individual companies, who's spending the most on ai
- Look at customer's like samsung, intel.
- This would strengthen our case
- Trying to find ai demand over time
- Users for openai, gemini, etc.
- Revenue over time (less good proxy)
- Links
- https://www.statista.com/statistics/1559548/private-sector-data-center-construction-completed-us/
- https://programs.com/resources/data-center-statistics/
- https://www.barchart.com/story/news/744648/data-center-infrastructure-management-market-recent-trends-growth-outlook-size-share-future-scope-and-demand
- "The Data Center Infrastructure Management Market is anticipated to expand from USD 3.02 billion in 2024 to USD 5.01 billion in 2029 at a compound annual growth rate (CAGR) of 10.6% over the course of the forecast period."
- https://touchplan.io/wp-content/uploads/2025/11/MSIDataCenterReport.pdf
- This has a really good chart on page 5
- https://www.androidheadlines.com/2026/03/samsung-73-billion-ai-chip-investment-2026.html
billion spend on AI chips by Samsung in 2026.
- https://www.analyticsinsight.net/news/samsungs-310-billion-ai-investment-to-revolutionize-chip-and-data-centre-industry
- Samsung's
billion AI investment to revolutionize chip and data centre industry. - Doesn't really give the proper data compared to US total spend on datacentres infra per year.
- Samsung's
- https://arstechnica.com/gadgets/2025/12/after-nearly-30-years-crucial-will-stop-selling-ram-to-consumers/
- Crucial stops consumer RAM sales. Focusing on AI now.
- https://images.samsung.com/is/content/samsung/assets/global/ir/docs/2025_4Q_conference_eng.pdf
- You can see that memory sales have gotten higher and higher.
YoY.
- You can see that memory sales have gotten higher and higher.
- https://images.samsung.com/is/content/samsung/assets/global/ir/docs/2024_con_quarter04_all.pdf
- We can use this to show the CapEx over time.
- https://download.intel.com/newsroom/2025/corporate/29r29Xn/Intel-2Q2025-Earnings.pdf
- This shows intel's increase in DataCentre and AI revenue.
- Increase in AI demand over time
- Users
- https://openai.com/fr-FR/index/how-people-are-using-chatgpt/
- OpenAI said in September 2025 that ChatGPT had 700 million weekly active users. In December 2025, OpenAI also said it had more than 1 million business customers using its tools. In January 2026, OpenAI said revenue scaled with compute from $2B ARR in 2023 to $6B in 2024 and $20B+ in 2025. Those three figures together make OpenAI one of the strongest official demand narratives available.
- November 2022 - 100m
- March 2025 - 400m
- Sept 2025 - 700m
- October 2025 - 800m
- December 2025 - 800m
- February 2026 - 900m
- Gemini
- Gemini app over 400 million MAUs in May 2025
- more than 450 million MAUs in July 2025
- surpasses 650 million MAUs in November 2025
- https://openai.com/fr-FR/index/how-people-are-using-chatgpt/
- Proxied through Data centre and revenue
- https://investor.nvidia.com/news/press-release-details/2025/NVIDIA-Announces-Financial-Results-for-Fourth-Quarter-and-Fiscal-2025/
- Q4 FY2025 Data Center revenue: $35.6B, up 93% YoY
- Q4 FY2026 Data Center revenue: $62.3B, up 75% YoY.
- https://www.microsoft.com/investor/reports/ar25/index.html
- Azure surpassed $75B in annual revenue in FY2025, up 34%
- In FY2026 Q2, Azure and other cloud services revenue grew 39%
- https://investor.nvidia.com/news/press-release-details/2025/NVIDIA-Announces-Financial-Results-for-Fourth-Quarter-and-Fiscal-2025/
- Story:
- Higher DRAM / memory pricing would likely help LRCX indirectly by improving memory-maker margins and increasing incentives to spend on wafer-fab equipment for capacity and node transitions.
- Samsung's Memory business grew 62% YoY in 4Q25 and is explicitly ramping AI-linked products like HBM4, DDR5, and AI-related NAND, which supports stronger memory equipment demand.
- Micron's decision to exit the Crucial consumer business suggests memory supply is shifting away from low-value consumer channels toward higher-value AI and data-center demand.
- Continued downstream AI infrastructure growth, shown by NVIDIA's Data Center revenue up 93% YoY in Q4 FY2025, supports the view that memory demand and pricing should remain favorable for Lam's customers.
- Parameter growth:
- Users
AI Overview
Interpretation
You need presentation-usable U.S. data center infrastructure spend for the last ~4–5 years, preferably something you can cite cleanly and chart without relying on junk market blogs.
Assumptions
"Infrastructure spend" = primarily physical construction spend on data center buildings and built-in systems, not total AI capex including GPUs, servers, land, or utility upgrades.
False premises
None, but one correction: a lot of pages online mix up construction spend, construction starts, market size forecasts, and total AI infra capex. Those are not the same thing.
Answer
The best core source is the U.S. Census construction spending series, because it is actual U.S. spend data, not a consultant forecast. Census' historical data page is the underlying source, and Our World in Data has already turned it into a clean reusable chart. Census describes its "value of construction put in place" series as the dollar value of construction work done in the U.S., and its historical files include private construction broken down by type. Our World in Data states explicitly that its chart is based on the Census private construction series, with data center construction as a category, and adjusts it for inflation using BLS producer price data. (Census.gov)
The cleanest chart to use is this one: Our World in Data — "Monthly spending on data center construction in the United States." It is open-access, presentation-friendly, and based on Census + BLS. It also clearly explains scope: it includes on-site building construction work, materials, labor, and contractor costs, but excludes land, maintenance, and IT hardware such as servers and storage. That makes it great for a "U.S. data center construction spend" slide, but not for a "total AI infrastructure spend" slide. (Our World in Data)
For the direction of travel, several reputable sources are consistent: by mid-2024, NAIOP reported data center spending had risen for 12 straight months and reached $27 billion at a seasonally adjusted annual rate in May 2024, up 69% year over year, and equal to 33% of private office construction. By late 2025, Reuters reported U.S. data center construction spending hit about $40 billion at a seasonally adjusted annual rate in June 2025, following a 50% surge in 2024. (naiop.org)
For a simple annual headline series, one useful secondary source based on Census is Wolf Street: it says full-year 2025 U.S. data center construction spending reached $41 billion, up 32% from 2024, more than 100% above 2023, and 344% above 2020. I would treat this as a convenient summary, not the primary source, and still cite Census/OWID underneath your chart where possible. From those percentages, the implied rough historical path is about $9.2B in 2020, $31.1B in 2024, and $41.0B in 2025; the exact 2021–2023 annual totals should be pulled from the raw Census monthly file before you put numbers on a slide. (wolfstreet.com)
For supporting geography slides, Reuters has a strong map-based angle: the U.S. is poised for a major capacity expansion, with filings indicating potential growth from about 15 GW to over 150 GW, led by Virginia, Texas, and Pennsylvania. That is not spend data, but it pairs well with a spend chart to show where the money is likely concentrating. (Reuters)
For a more polished industry visual, the MSI / MOCA Systems reports are usable as secondary chart sources. The Q1 2026 construction economy report includes a strong "current vs planned data center capacity across key U.S. states" graphic, and the special report says U.S. data center spending was about $11 billion in 2022 and projects a much larger 2026 market. The 2022 point is useful, but anything forward-looking from MSI belongs in a forecast slide, not your historical spend slide. (MOCA Systems Inc.)
What I'd use in your deck
Slide 1: Main historical spend chart
Use the OWID chart based on Census/BLS. Title it something like:
"U.S. data center construction spending has surged since 2022". (Our World in Data)
Slide 2: Annual headline callouts
Use these cautiously:
-
2020: about $9.2B implied from the 2025 "+344% vs 2020" figure
-
2024: about $31.1B implied from 2025 being 32% higher
-
2025: $41.0B full year
Mark 2020 and 2024 as derived from reported growth rates unless you verify them directly from the raw Census file. (wolfstreet.com)
Slide 3: Geographic support
Use Reuters/MSI state-capacity map language: Virginia and Texas remain the dominant build-out hubs, with Pennsylvania and several Sun Belt/Midwest markets gaining share. (Reuters)
Best sources to pull from
-
U.S. Census historical construction data — raw underlying source. (Census.gov)
-
Our World in Data chart — easiest reusable public chart built from Census + BLS. (Our World in Data)
-
Reuters / Bank of America Institute coverage — current milestone framing for 2024–2025 growth. (Reuters)
-
NAIOP — useful 2024 benchmark and share-of-office comparison. (naiop.org)
-
MSI / MOCA Systems — supporting visuals and forecast framing, not primary history. (MOCA Systems Inc.)
Tradeoffs
Cheapest viable option: use the OWID chart only. Fast, free, credible, but monthly rather than neatly annual.
Middle ground: OWID chart + Reuters/NAIOP callouts. Better story, still fast.
Best overall: pull the raw Census monthly series and aggregate it into annual U.S. spend for 2021–2025, then pair it with a state-capacity visual. That gives you the cleanest investor-style slide, but it takes a bit more work.
Risk: MSI, ConstructConnect, and other industry reports are useful, but they often mix starts, pipeline, and spend. Don't let those slip into the same chart.
What I'm least certain about
The exact annual 2021–2023 dollar totals without directly aggregating the raw Census monthly file. The trend is solid; the precise yearly values should be verified from the Census download before you put hard numbers on a slide.
I can turn this into a slide-ready list of 3–5 exact charts with captions and source lines next.
Summary
- https://ourworldindata.org/grapher/monthly-spending-data-center-us
- Very good chart. Exactly what we need.
- https://www.naiop.org/research-and-publications/magazine/2024/fall-2024/business-trends/diving-deeper-into-construction-spending/
- https://wolfstreet.com/2026/02/28/construction-spending-on-data-centers-factories-powerplants-and-office-buildings-boom-bust-and-in-between/