Nvidia and Amazon invest billions in their own power infrastructure for AI

Nvidia is investing up to $3 billion in Texas-based infrastructure company Lancium, which develops power capacity and land for AI data centres. Meanwhile, Amazon is building its own gas-fired power plant in Pecos County, Texas, with a maximum capacity of 7.65 gigawatts. Both announcements came to light in early August 2026 and illustrate how significant the energy demands of the AI sector have become.

The investments are not only large in scale, they also show that tech giants are increasingly seeking to manage their own power supply for AI infrastructure, rather than leaving that entirely to grid operators or energy companies.

Global data centre electricity consumption stood at approximately 415 terawatt-hours in 2024, roughly 1.5 percent of total worldwide consumption. Analysts attribute much of the rise in recent years to the growth of AI workloads.

Nvidia takes a 20 percent stake in Lancium

Nvidia announced on 9 August 2026 that it would invest an initial $2 billion in Lancium, equivalent to approximately a 20 percent stake. An additional $1 billion is tied to specific milestones, including securing grid connections. Based on the deal, Lancium is valued at roughly $10 billion, including its land and power assets.

Lancium, founded in Houston in 2017 and now headquartered in The Woodlands, Texas, already has four gigawatts of power under contract in Texas. In addition to that existing contract, the company is developing sites for a further 15 gigawatts. CEO and co-founder Michael McNamara leads the company, which is also partially owned by Blackstone Group.

Lancium's most prominent site is a data campus near Abilene, Texas. That campus has a 1.2-gigawatt grid connection approved by grid operator ERCOT and is being expanded to eight buildings totalling approximately four million square feet. Each building is designed for up to 50,000 Nvidia GB200 NVL72 systems. The Abilene site is also the first operational Stargate location, the data centre project being developed jointly by SoftBank, OpenAI and Oracle with an investment budget of up to $500 billion for US AI infrastructure.

A second campus, near Childress in Texas, also holds an ERCOT-approved grid connection of one gigawatt. Construction is scheduled for the third quarter of 2026. Lancium is considering a public listing in 2027.

Amazon builds gas plant that could become America's most polluting

Amazon confirmed on 7 August 2026 that it is building a gas-fired power plant at the GW Ranch site in Pecos County, Texas. The project had already received a permit from the state of Texas in January 2026. The New York Times reported on the scale of the plant on 8 August 2026.

The plant, featuring 35 natural gas turbines and a maximum output of 7.65 gigawatts, is permitted to emit 33 million tonnes of CO₂ per year. By way of comparison, this would make the plant the single largest individual source of carbon dioxide in the United States. The facility is initially intended to operate independently of the Texas electricity grid and to supply Amazon's data centres in the region directly with power. For cooling water, the plant uses non-potable, brackish groundwater.

Amazon spokeswoman Margaret Callahan confirmed the project to the press. The company co-founded The Climate Pledge in 2019 and is targeting net-zero CO₂ emissions by 2040. However, Amazon's total emissions have risen every year in recent years, a trend analysts link to the substantial growth of data centres for AI. Through 40 projects in Texas, Amazon has enabled 10 gigawatts of carbon-free generation capacity for its existing data centre operations.

Tension between climate targets and energy demand

Together, the two projects expose a fundamental tension within the AI sector. Companies such as Amazon have publicly committed to ambitious climate targets, while concrete investments in fossil-fuel power generation are growing in scale. The 33 million tonnes of CO₂ the Texas plant is permitted to emit annually sit in direct contradiction to Amazon's own net-zero target for 2040.

With its investment in Lancium, Nvidia is pursuing a different strategy: the chip company is focusing on power infrastructure itself, without directly building fossil generation capacity. Lancium works with data centre partner Crusoe, with which it had already announced a deal in July 2024 for a 200-megawatt AI-focused data centre outside Abilene as the first phase of an expansion to 1.2 gigawatts.

For the broader Stargate consortium, which aims to invest up to $500 billion in US AI infrastructure, Lancium and its Abilene campus represent the first concrete operational link in that chain. The pace at which that capacity is being built, partly driven by the Nvidia investment, reflects how urgent the race for computing power and the associated energy is at present.

For the broader European and Dutch AI sector, developments in the United States serve as a benchmark. The energy demands of AI data centres are also a factor here, but European regulations on CO₂ emissions, permitting and grid congestion make it virtually impossible to build fossil backup capacity at this scale in Europe. This creates both a competitive disadvantage in terms of speed and an incentive to invest earlier in alternative, low-carbon power solutions. Investors and policymakers seeking to facilitate AI infrastructure in the Netherlands or elsewhere in Europe will need to treat the power question as a structural bottleneck, not a technical footnote.