AI Infrastructure News 2026: US. Data Centers, Power, Chips & Investments to Watch

AI infrastructure news now shapes U.S. business headlines as much as chip launches or model releases. Since mid-August 2026, NVIDIA has backed an 8-gigawatt data center campus in Ohio. Google, NVIDIA and Emerald AI have launched an alliance for grid-friendly data centers. Meanwhile, the Senate has stalled a bill on who pays for data center power costs.

UsaBusinessDaily tracks AI infrastructure news for U.S. business and technology readers. Below, you will find what happened, why it matters and what to watch next. We cover the largest AI data centers in the USA, the companies behind them and the electricity numbers driving the boom. Every statistic names its source and year. Where forecasts disagree, we show both.

For wider coverage, see our AI news and trends and latest technology news sections.

Key facts from the latest AI infrastructure news

Metric	Figure and source
Global data center electricity, 2025	~485 TWh (IEA)
Global data center electricity, 2030	~950 TWh, just under 3% of world demand (IEA, April 2026 update)
Global data center electricity, 2026	565 TWh, up 26% from 447 TWh (Gartner, June 2026; different method from IEA)
U.S. data centers, 2030	649 TWh, or 11.8% of U.S. electricity; range 9.5%–15.3% (Berkeley Lab, June 2026)
OpenAI Stargate U.S. capacity	10 GW secured, announced April 29, 2026 (OpenAI); contracted, not all live
Ohio PORTS-Pike campus	First phase 4.25 IT-GW; up to 8 IT-GW (NVIDIA, August 17, 2026)

AI Infrastructure News: What Is Happening in 2026?

The short answer: the latest AI infrastructure news shows a buildout that has turned from a chip race into a power race. Four recent developments show how.

1. OpenAI says it has secured 10 GW of U.S. capacity (April 29, 2026)

OpenAI announced that its Stargate program has passed its goal of 10 gigawatts (GW) of U.S. AI infrastructure. The company had set that target for 2029 when Stargate launched in January 2025. In addition, OpenAI said it added more than 3 GW in the previous 90 days. Its first active site is in Abilene, Texas. Further sites are in development in Texas, New Mexico, Wisconsin and Michigan.

Why it matters: The 10 GW figure appears to be contracted capacity, not power that is running today. Data Center Dynamics noted that OpenAI’s CFO put its compute at 1.9 GW at the end of 2025. Therefore, read “secured” as “signed and planned,” because live capacity will arrive in phases over several years.

2. NVIDIA backs an 8 GW Ohio campus (August 17, 2026)

NVIDIA said it secured land, power and building capacity at the PORTS-Pike Technology Campus in Pike County, Ohio. SB Energy will build and operate the site, and OpenAI will lease it for 20 years. The first phase is designed for 4.25 IT-GW. NVIDIA also holds an option on another 3.75 IT-GW, which brings the total to 8 IT-GW. In addition, NVIDIA is investing $1.5 billion in SB Energy and providing credit support.

The partners also plan at least 10 GW of new power generation and $4.2 billion in regional grid investment with AEP Ohio, according to the announcement. Reuters separately reported a guarantee of up to $105 billion tied to lease and power payments.

Why it matters: In other words, NVIDIA is no longer only selling chips. It is helping finance the land, power and buildings those chips need. According to OpenAI’s project page, the first 800 MW is expected in 2028, the full build runs through 2032, and OpenAI pays only as completed capacity becomes available.

3. Google, NVIDIA and Emerald AI launch a grid-flexibility alliance (September 16, 2026)

The AI Energy Management Alliance (AEMA) promotes data centers that can cut or shift their electricity use when the grid is under strain. For example, a flexible facility can move workloads to other times, discharge on-site batteries or use paired power generation. The alliance launched with 18 partners, including Anthropic, Constellation, NRG, National Grid and AES.

Why it matters: Utilities often assume large facilities draw steady power. Therefore, flexible data centers could win faster grid connections if they prove they can reduce demand on request. That would ease one of the biggest bottlenecks in the sector.

4. The Senate stalls the Ratepayer Protection Act (late September 2026)

The House passed the Ratepayer Protection Act by 417 to 3. The bill aims to make data centers and tech companies pay for new electricity infrastructure instead of passing those costs to consumers. However, a Senate attempt to pass it by unanimous consent was blocked. Senator Martin Heinrich argued it does not go far enough. The bill also does not require states to adopt such standards.

Why it matters: Electricity bills are now a political issue for data center projects. As a result, expect local approvals, utility tariffs and state rules to shape where new projects get built.

What Is AI Infrastructure?

AI infrastructure is the set of computing hardware, networking, storage, software, power and cooling systems used to train and run AI models at scale. In practice, it means chips, servers, data centers and the electricity that feeds them.

IBM and NVIDIA describe it in similar terms: compute, storage, networking and the software that manages them. In 2026, however, power and cooling sit at the center of the definition, because they limit how fast everything else can grow.

Layer	What it does	2026 pressure point
AI models	Generate text, images, code, and predictions	Training and inference demand grow together
GPUs and accelerators	Process AI workloads in parallel	AI-optimized servers use 175 TWh in 2026 (Gartner)
Servers and racks	House the chips	Denser racks produce more heat
Networking	Links thousands of chips into one cluster	Delays leave costly chips idle
Storage	Holds training data and model files	Must feed chips without gaps
Cooling	Removes heat from equipment	Cooling and other infrastructure use 195 TWh in 2026 (Gartner)
Power	Supplies electricity	The main bottleneck, per Gartner
Data center	Houses the full stack	New campuses now plan 1 GW or more

Table 1: The AI infrastructure stack. Layer descriptions are USA Business Daily’s summary; figures are from Gartner (June 2026).

Why Is AI Infrastructure Growing So Fast?

Two kinds of work drive the growth: training and inference. Training builds a model, while inference runs it every time someone sends a prompt. As more companies build AI into products, inference workloads multiply.

The hardware numbers show the shift. Gartner estimates that electricity use by AI-optimized servers grew about 84% in 2025 and will grow another 84% in 2026. In contrast, conventional servers grew about 1% in both years. By 2027, Gartner expects AI-optimized servers to use more power (258 TWh) than conventional ones (200 TWh).

Money also follows demand. The IEA reports that capital spending by five tech companies (Alphabet, Meta, Microsoft, Amazon Web Services and Oracle) topped $400 billion in 2025. It expects that figure to rise about 75%, to $715 billion, in 2026. Most of it goes to data centers.

In short, AI demand becomes chip orders, chip orders become data center campuses, and campuses become electricity demand.

Biggest U.S. AI Infrastructure Projects

Much AI infrastructure news centers on a handful of giant campuses. Epoch AI, a research group that tracks large AI data centers, updated its rankings in September 2026. Because its figures are estimates built from public information, the table below combines Epoch’s numbers with company announcements.

Project	Operator	Location	Capacity (IT power)	Status
Colossus 2	xAI	Memphis, Tennessee	~946 MW now; ~1,531 MW planned	Operating; full build 2027
New Carlisle	Anthropic and Amazon	Indiana	~910 MW now; ~1,925 MW planned	Operating; full build 2028
Fairwater Atlanta	Microsoft	Georgia	~636 MW	Operating
Stargate Abilene	OpenAI and Oracle	Abilene, Texas	~421 MW	Operating (first Stargate site)
Fairwater Wisconsin	Microsoft	Mount Pleasant, Wisconsin	~2,263 MW planned	Projected 2028
Stargate New Mexico	OpenAI	New Mexico	~1,750 MW planned	Projected 2028
Hyperion	Meta	Louisiana	~1,676 MW planned	Projected 2028
PORTS-Pike	OpenAI, SB Energy, NVIDIA	Pike County, Ohio	4.25 IT-GW first phase; up to 8 IT-GW	First 800 MW expected 2028; build through 2032

Table 2: Largest U.S. AI data centers. Sources: Epoch AI (September 2026) for operating and planned MW estimates; NVIDIA and OpenAI (August 2026) for the Ohio campus. All figures are estimates or announced targets.

By site count, Epoch’s tracker shows Texas (13 sites), Virginia (7), Ohio (5), Nebraska (4) and Iowa (4) hosting the most large AI data centers. Note that megawatt figures here describe IT power, which is the electricity used by computing equipment alone.

How Much Electricity Does AI Infrastructure Use?

Power is the most cited theme in AI infrastructure news this year. The numbers below show why.

Data centers used about 485 TWh of electricity worldwide in 2025, according to the IEA. The agency projects roughly 950 TWh by 2030, which is just under 3% of global electricity use.

AI also drives much of that growth. The IEA reported that electricity use by AI-focused data centers rose about 50% in 2025, compared with 17% for data centers overall.

In the United States, the IEA expects electricity demand to grow about 2% a year from 2026 to 2030, with data centers behind roughly half of the increase. Berkeley Lab’s June 2026 update goes further. It estimates that U.S. data centers could use 649 TWh in 2030, or 11.8% of U.S. electricity, within a range of 9.5% to 15.3%. By comparison, its 2024 report had projected 6.7% to 12% by 2028.

Meanwhile, Gartner’s June 2026 forecast is higher still for the world. It expects data centers to use 565 TWh in 2026 and more than 1,200 TWh by 2030. AI-optimized servers will account for 31% of data center power use in 2026, up from about 20% a year earlier. Gartner also expects the U.S. to use about 204 TWh in 2026, roughly 36% of the global total. Analyst Linglan Wang said AI capacity is now “constrained by power availability.”

Source	Scope	Baseline	Forecast
IEA (April 2026)	Global	485 TWh in 2025	~950 TWh in 2030
Gartner (June 2026)	Global	447 TWh in 2025	565 TWh in 2026; 1,200+ TWh in 2030
Berkeley Lab (June 2026)	U.S. only	~4.5% of U.S. electricity in 2024	649 TWh (11.8%) in 2030; range 521–843 TWh

Table 3: Three data center electricity forecasts. They use different methods and boundaries, so do not add or compare them directly.

Why do the numbers differ? Because each organization counts different equipment and uses different assumptions, the results vary. For example, Gartner puts 2025 at 447 TWh, while the IEA says 485 TWh. Berkeley Lab covers only the U.S. and builds its estimate from expected equipment shipments. Treat each forecast as a separate view, and always check which one an article is quoting.

Companies Building America’s AI Infrastructure

Dozens of companies appear in AI infrastructure news. The table groups the most visible ones by role. This list is informational and is not investment advice.

Role	Companies	What they do
Chips and systems	NVIDIA, AMD, Broadcom, Dell, Supermicro, HPE	Supply GPUs, accelerators, networking and AI servers
Cloud and hyperscalers	Microsoft, Amazon, Google, Oracle, Meta	Build and run large campuses; rent capacity to AI labs
AI developers	OpenAI, Anthropic, xAI	Consume compute for training and inference; co-develop sites
Specialized AI clouds	CoreWeave	Rent GPU capacity; OpenAI is a known customer
Data center operators	Equinix, Digital Realty, QTS	Provide colocation space, power and connectivity
Power and cooling	Vertiv, SB Energy, AEP Ohio	Supply cooling, electrical equipment and grid or generation capacity

Table 4: Who builds what. Project examples: Microsoft (Fairwater), Meta (Hyperion), xAI (Colossus 2), Anthropic and Amazon (New Carlisle).

Partnerships now cross these categories. For example, OpenAI signed a deal to use 2 GW of Amazon Web Services’ Trainium chips in early 2026, according to Data Center Dynamics. Meanwhile, NVIDIA is financing a campus that OpenAI will lease. As a result, the line between chipmaker, landlord and customer keeps blurring.

GPUs Are Only One Part of AI Infrastructure

AI infrastructure news often focuses on GPUs, but buying them is the easy part of the story. Keeping them powered and cool is harder. Gartner estimates that cooling and other supporting equipment will use 195 TWh in 2026, up 22.6%. That is more than the 175 TWh used by AI-optimized servers themselves.

Cooling design is therefore a competitive issue. Microsoft’s Fairwater campus in Mount Pleasant, Wisconsin is reported to use a closed-loop liquid cooling design that avoids water use for cooling. Networking matters too, because AI clusters link thousands of chips and delays between them leave expensive hardware idle.

Where Is the U.S. AI Infrastructure Boom Happening?

Recent AI infrastructure news points to these states.

• Texas: Home to the Stargate Abilene site and the most large AI data centers Epoch tracks (13).
• Ohio: Site of the PORTS-Pike campus in Pike County, built on the former Portsmouth Gaseous Diffusion Plant area. Epoch tracks five large sites in the state.
• Virginia: Seven tracked sites. Northern Virginia remains a long-established data center hub.
• Indiana: New Carlisle hosts the Anthropic and Amazon campus, about 910 MW today by Epoch’s estimate.
• Tennessee: Memphis hosts xAI’s Colossus 2, the largest operating AI data center Epoch tracks.
• Georgia: Microsoft’s Fairwater Atlanta is estimated at 636 MW.
• Wisconsin: Microsoft’s Fairwater campus in Mount Pleasant is projected to reach about 2,263 MW by 2028. Stargate also lists a Wisconsin site in development.
• New Mexico and Michigan: Both host Stargate sites that OpenAI says are in development.
• Louisiana: Meta’s Hyperion campus is projected at about 1,676 MW by 2028.

The Biggest AI Infrastructure Challenges

The main AI infrastructure challenges in 2026, in rough order of urgency:

1. Power availability. Gartner says grid supply will be insufficient for future data center construction. Power access now decides which projects move forward.
2. Grid connections. Large campuses need new generation and transmission. The Ohio project alone includes $4.2 billion in grid investment.
3. Cooling and water. Denser chips create more heat, so operators are turning to closed-loop and liquid cooling to limit water use.
4. Who pays for the grid. The stalled Ratepayer Protection Act shows the fight over passing costs to households and small businesses.
5. Financing. NVIDIA’s credit support and OpenAI’s long leases show how much capital these projects need before any chip runs.
6. Forecast uncertainty. Berkeley Lab’s 2030 range runs from 521 to 843 TWh, so planners face a wide gap between low and high cases.
7. Project reversals. The Decoder reported that OpenAI saw a Texas expansion rejected, paused a UK project, and dropped a Norway site. Announced capacity does not always get built.

What AI Infrastructure Means for U.S. Businesses

Here is what AI infrastructure news means for everyday U.S. companies.

• Access to AI services. OpenAI argues that more compute lets it serve users more reliably and lower costs over time. However, no source guarantees lower prices, so budget for changes either way.
• Electricity costs. Businesses in regions with heavy data center growth should watch utility rate filings and state rules. The ratepayer debate may affect commercial bills.
• Supplier opportunities. Project announcements for the Ohio campus cite about 35,000 construction jobs and 2,500 operating jobs. Electrical, construction and cooling suppliers sit near these projects.
• Adoption planning. Small firms do not need their own data centers. Instead, they need reliable AI tools. Start with our guides to AI tools for small businesses and agentic AI explained.
• Search and marketing. Cheaper, faster AI changes how customers find answers. See AI and digital marketing and what is GEO for practical steps.

AI Infrastructure Trends to Watch Through 2030

These trends appear most often in AI infrastructure news and analyst forecasts.

• Flexible data centers. The AEMA alliance pushes facilities that can cut demand during grid stress in exchange for faster connections.
• Inference-heavy workloads. Gartner expects AI-optimized servers to outuse conventional servers by 2027.
• Liquid cooling and denser racks. Cooling demand is rising 22.6% in 2026, which pushes operators toward liquid designs.
• On-site and dedicated power. The Ohio plan includes at least 10 GW of new generation. Some sites add on-site turbines or batteries.
• Larger power scale. Gartner projects data center power demand of 290 GW by 2030, up from 104 GW in 2025.
• New financing structures. Expect more credit guarantees, long leases, and vendor-backed deals.
• Policy action. Federal and state rules on cost allocation and interconnection will shape timelines.

For a small-business angle on these shifts, read our AI trends small businesses should watch in 2026.

Frequently Asked Questions

What is AI infrastructure?

AI infrastructure is the hardware, software, networking, storage, power and cooling used to train and run AI models at scale. It includes GPUs, servers, data centers and the electricity supply behind them.

Why is AI infrastructure growing so quickly in 2026?

Demand for training and inference keeps rising. Gartner says AI-optimized server power use grew about 84% in 2025 and will grow about 84% again in 2026. Large companies are also raising capital spending on data centers.

What companies are building AI infrastructure?

Key players include NVIDIA, AMD, Microsoft, Amazon, Google, Meta, Oracle, OpenAI, Anthropic and xAI. Specialized providers such as CoreWeave, Equinix, Digital Realty and Vertiv supply capacity, space, power and cooling.

Where are AI data centers being built in the U.S.?

Epoch AI tracks the most large sites in Texas, Virginia, Ohio, Nebraska and Iowa. Major individual projects are in Tennessee, Indiana, Georgia, Wisconsin, New Mexico, Louisiana and Michigan.

How much electricity do AI data centers use?

The IEA estimates global data centers used about 485 TWh in 2025 and could reach about 950 TWh by 2030. Berkeley Lab estimates U.S. data centers could use 649 TWh in 2030, or 11.8% of U.S. electricity.

Why do AI data centers need GPUs?

GPUs run many calculations at the same time, which suits AI training and inference. Standard processors handle such parallel work far less efficiently. That is why AI-optimized servers now draw a fast-growing share of data center power.

What are the biggest challenges facing AI infrastructure?

Power availability, grid connections, cooling, financing and cost allocation lead the list. Gartner says AI capacity is constrained by power, and Congress is debating who should pay for new grid costs.

What will AI infrastructure look like by 2030?

Expect larger campuses, more on-site power, liquid cooling and flexible demand. Gartner projects data center power demand of 290 GW by 2030. Forecasts vary, so watch for updates from the IEA and Berkeley Lab.

Conclusion:

The core story of AI infrastructure news in 2026 is simple. AI demand is growing faster than power, cooling and financing can easily follow. Projects in Ohio, Texas, Indiana and Wisconsin show the scale. Meanwhile, the Senate debate and the AEMA alliance show that electricity is now as important as chips.

Want more U.S. technology and business updates? Explore our U.S. business trends coverage and our AI news and trends hub. To stay ahead, subscribe to the USA Business Daily newsletter for more AI infrastructure news and U.S. business reporting.