英伟达 (NVDA.US) 30 亿美元入股黑石旗下 Lancium：从芯片霸主到电力 “包工头”，AI 能源版图再下一城

According to Zhitong Finance APP, as the global AI competition shifts its focus from computing power to electricity, NVIDIA (NVDA.US) is redefining the boundaries of "AI infrastructure" with concrete actions. It has been reported that NVIDIA has agreed to invest up to $3 billion in Lancium, a power infrastructure developer supported by Blackstone Group, to acquire approximately 20% equity in the company behind the "Stargate" data center campus. This deal extends NVIDIA's AI landscape directly from the chip layer to the energy infrastructure layer—locking in electricity is becoming a more urgent strategic proposition than locking in customers in an era where the power consumption of AI chips is rising exponentially.

This transaction marks a historic extension for NVIDIA from its core chip business to the bottom layer of the AI value chain—energy infrastructure. The bottleneck of AI has shifted from computing power to the electricity required to operate that computing power.

Who is Lancium? The power operator behind "Stargate"

Founded in 2017 and headquartered in Houston, Texas, Lancium was co-founded by Michael McNamara, who serves as CEO. The company initially focused on energy transition technologies, with its core product, Smart Response software, enabling clean campuses to act as controllable load resources (CLR) for the grid. Its early business included energy-intensive applications like Bitcoin mining, gradually shifting focus to the power infrastructure of AI data centers.

Lancium's most critical asset is located in Abilene, Texas—a 1,000-acre "Lancium Clean Campus," which is the first operational base of the "Stargate" plan. The campus is designed to accommodate approximately 400,000 NVIDIA AI chips distributed across eight buildings. Lancium has currently locked in and developed 4 gigawatts of power on the Texas grid, with an additional 15 gigawatts of projects in progress.

Transaction Anatomy: Strategic Upgrade from "Selling Chips" to "Locking in Power"

The structure of this transaction clearly reflects NVIDIA's strategic intentions.

Phase One: NVIDIA invests $2 billion to acquire approximately 20% equity in Lancium. Based on this, Lancium's enterprise value is estimated at around $10 billion, including debt.

Phase Two: Once multiple campuses under Lancium meet established conditions for grid access, NVIDIA will invest an additional $1 billion, potentially increasing its ownership stake to about 30%.

This investment is purely equity-based and does not involve credit guarantees for data center construction and leasing. However, insiders reveal that NVIDIA is considering providing credit guarantees for other projects, including a large data center project in Ohio involving OpenAI and SoftBank, with a scale of up to 10 gigawatts.

Lancium's asset portfolio is already quite substantial. The company has secured 4 gigawatts of power contracts in Texas, including:

1.2 gigawatts supplied to OpenAI and Oracle at the Abilene "Stargate" campus; 900 megawatts for partner Crusoe's new data center for Microsoft in Abilene;

1 gigawatt contracted by QTS Data Centers in Turkey, Texas;

1 gigawatt under construction by Crusoe in Childress.

In addition to locked assets, Lancium has also laid out land in Texas with the potential to support 15 gigawatts of power, awaiting grid connection permits.

Lancium's Abilene "Clean Park" covers 1,000 acres and has 1.2 gigawatts of grid-connected facilities, which have received full approval from the Electric Reliability Council of Texas (ERCOT). This park is the first operational base of the "Stargate" program.

The "First Principles" of AI: From Computing Power Shortage to Power Shortage

NVIDIA CEO Jensen Huang's theory of the AI "five-layer cake" divides the AI industry from the bottom up into five levels: energy, chips, infrastructure and system software, AI models, and applications, with energy defined as the "first principle" of AI infrastructure.

This theory is becoming a reality. Huang recently estimated that the cost of building a 1-gigawatt data center could soon reach $80 billion to $100 billion. As AI clusters scale from tens of thousands to hundreds of thousands and millions, the bottleneck in power supply is more challenging than chip supply— the power demand of a super-large data center is equivalent to that of a medium-sized city, while grid upgrades and access often take years. Lancium is a key player in addressing this pain point.

From "Stargate" to Investment Landscape: NVIDIA is Transforming into an Infrastructure Company

The Lancium deal is the latest note in NVIDIA's accelerated transformation into a core investor in the AI infrastructure ecosystem. Reports indicate that in the quarter ending this April, NVIDIA invested a total of $18.6 billion in private companies and infrastructure funds, surpassing the total investment amount for the entire previous year.

So far this year, NVIDIA has issued $2 billion checks to cloud service providers CoreWeave and Nebius, and completed investments in network equipment companies Lumentum, Coherent, and semiconductor company Marvell. This series of actions indicates that NVIDIA is rapidly transforming from a chip seller to a core investor in the AI infrastructure ecosystem.

In the clean energy sector, NVIDIA has previously invested in the fusion company Commonwealth Fusion Systems and the nuclear energy company TerraPower founded by Bill Gates.

NVIDIA has a clear business logic for such investments: to secure scarce power resources for its large chip customer base in advance. When customers need to build AI data centers, NVIDIA can not only provide chips but also ensure power supply through its invested power infrastructure companies—this is a strategic upgrade from "selling shovels" to "controlling the water source."

NVIDIA's layout in the energy sector is far from an isolated case. By 2025, NVIDIA has made over 10 moves in clean energy development, grid management, and data center construction, including investments in the fusion company Commonwealth Fusion Systems and the nuclear energy company TerraPower founded by Bill Gates From nuclear fusion to grid access, NVIDIA is building a portfolio that covers the entire AI energy chain—Lancium is the latest move in this strategy, but far from the end.

Risks and Variables: Texas Governor's "Grid Pause Order"

Just before and after the trading news broke, a potential variable emerged. Texas Governor Greg Abbott ordered a pause on new data center grid access applications this week, requiring related projects to undergo further review.

This policy shift brings uncertainty to Lancium's expansion timeline. However, insiders point out that Lancium's 15-gigawatt project awaiting grid access is in a relatively favorable position—the review process with its partner utility company American Electric Power has entered the later stages.

The broader risk is that the power demands of AI infrastructure are facing social resistance and regulatory scrutiny. From Google's $15 billion data center in Andhra Pradesh, India, encountering environmental protests, to the grid access pause in Texas, AI's "power hunger" is evolving from a technical issue into a social governance issue