Nvidia on the cap table: $100B pledged to OpenAI, roughly $10B a gigawatt

The conditionality is the part worth reading. OpenAI's money is released progressively as each gigawatt goes in ^[1], and Lancium's final tranche depends on thresholds that include grid hookups ^[4]. Capital priced against energized capacity rather than signed intent puts an interconnection schedule in front of a chip vendor's own cheque book.

Divide the pledge by the footprint and the rate is up to $10 billion per gigawatt of deployed systems ^[1]. At the other end of the stack, $2 billion for roughly a fifth of Lancium implies about $10 billion of equity value ^[2], close to the enterprise value The Information attached to that portfolio of land and power connections ^[5]. The asset underneath is the 1,000-acre Abilene campus, the first operational Stargate site ^[12].

Nvidia's reason for underwriting substations sits in its own product marketing: GPU utilization is the data center's most valuable resource, and time spent waiting constrains an AI factory's revenue ^[9]. It sells the racks. Whoever controls the power connection decides when those racks start billing, and Cloverleaf is the party that works with utilities, energy providers and investors to secure it ^[11].

Which leaves the $30 billion framing unsupported here. Across these four sources, no Nvidia position sits at the application layer; the disclosed stakes are one model lab and a set of power and site developers ^[4]. What the material does ground is $4.5 billion of named commitments outside OpenAI ^[3], an undisclosed Cloverleaf minority that the Wall Street Journal sized in the hundreds of millions ^[6], and an OpenAI intention roughly 22 times the rest of the named total combined ^[5].