Nvidia is reportedly putting up to $3 billion into Lancium, a Texas power-infrastructure developer tied to the Stargate data-center project. The reported structure is simple enough. An initial $2 billion buys roughly 20% of Lancium. Another $1 billion may follow if grid-connection milestones are hit.
That last clause is the useful part.
The AI industry has spent two years talking as if the bottleneck is model quality, GPU supply, or maybe memory bandwidth if the speaker has recently read a chip slide deck. Those all matter. But the Lancium deal is a reminder that the constraint is moving down the stack, into land, substations, interconnect queues, and power contracts.
You can buy GPUs and still not have compute.
Lancium is not just another data-center landlord. The company has been building power-heavy sites in Texas, including the Abilene campus associated with Stargate, the large AI infrastructure project backed by SoftBank, OpenAI, and Oracle. Reports put Lancium's land and power-connection portfolio at roughly a $10 billion enterprise value. The Decoder also noted that Lancium has about 4 gigawatts of power under contract in Texas and sites under development for up to 15 more gigawatts.
